VideoLAN Developers Reject Spy Agency Backdoors to Keep VLC Secure

VideoLAN开发者拒绝情报机构后门，坚守VLC安全 在全球视频播放领域，手写汇编代码的性能始终优于现代编译器；与此同时，由于企业依赖少数无偿志愿者，互联网基础设施在结构层面上依然脆弱。 本次讨论深入探讨了支撑全球互联网视频运转的底层开源基础设施，重点聚焦于VLC和FFmpeg。开发者详细阐述了多媒体编程严苛的性能要求、手写汇编语言始终优于现代编译器的原因，以及对专有编解码器进行逆向工程的复杂流程。他们还谈及了志愿者维护人员所面临的巨大压力，包括开源安全激励机制的不匹配、企业对公共代码库的依赖，以及在面对国家级情报机构的施压时，坚决不妥协用户隐私的坚定立场。 FFmpeg 与 VLC 作为 YouTube、Netflix、Chrome 和 OBS 等全球主流平台的隐形底层基石，在坚持零广告、无追踪的前提下，VLC 全球下载量已突破 65 亿次。 (30:07) 开源多媒体生态遵循严苛的“唯代码质量论”，完全不看贡献者的企业履历或背景，而支撑这套庞大生态的核心维护者极少，VLC 仅有 5 人，FFmpeg 仅约 10 至 15 人。 (89:18) 为在数十亿台设备上榨干硬件极限性能，关键编解码器严重依赖手工优化的汇编语言（核心代码占比近 80%），其运行速度较常规 C 代码及编译器自动向量化提升达 10 到 62 倍。 (121:08) 针对两家不同情报机构要求在 VLC 媒体播放器中植入后门的诉求，开发团队坚决予以拒绝，并明确表态若必须妥协安全性，宁可直接关闭该项目。 (228:59)

VLC and FFmpeg form the foundational bedrock of global digital video, operating as an interdependent open-source ecosystem driven by uncompromising code quality, hardware-level optimization, and resistance to corporate and state control. Project maintainers Jean-Baptiste Kempf and Kieran demonstrate that sustaining this critical infrastructure requires strict architectural principles and personal independence despite intense patent minefields and systemic maintainer burnout. Architectural Symbiosis and Resilient Media Decoding VLC and FFmpeg operate in a mutual symbiosis described as a "binary star system"—analogous to Android and Linux—sharing developers, cross-funding, and low-level code without a centralized command authority. When compiled for Windows, VLC incorporates roughly 16 million lines of code, of which only 1 million reside in VLC's native repository and 2 million originate from FFmpeg, with the remainder coming from external dependencies. The projects democratize media handling by allowing individual users to run the exact same software stack utilized by trillion-dollar tech conglomerates. Rooted in VLC's late-1990s origin as a campus streaming solution over packet-dropping UDP networks, the software operates on a strict zero-trust principle regarding input data. Both VLC and FFmpeg ignore three-letter file extensions, treating them merely as demuxer priority hints while directly analyzing internal file payloads. This payload-inspection architecture allows the software to seamlessly decode damaged streams, partial downloads with missing end-of-file index metadata, and engineered edge cases such as polyglot files that function simultaneously as valid ZIP and MP3 formats. The Assembly Mandate and Hardware-Level Optimization Multimedia playback imposes an unyielding real-time execution constraint: decoders have exactly 16 milliseconds to render a frame before visual output collapses. With FFmpeg estimated to run across 100 million to 1 billion CPUs concurrently worldwide, raw execution efficiency directly dictates hardware and energy costs. Consequently, core performance depends on mastering computer architecture—including CPU pipelining, Arithmetic Logic Units, and Single Instruction Multiple Data (SIMD) vector processing—rather than high-level programming abstractions. The maintainers reject conventional software engineering orthodoxies that rely on modern compiler auto-vectorization, asserting that human-optimized assembly consistently outperforms compiler-generated code by wide margins. For example, the AV1 decoder dav1d is composed of 79.9% assembly and 19.6% C, yielding performance up to 62 times faster than standard C implementations. While modern memory-safe languages like Rust provide advantages for greenfield parsing and networking layers, maintainers note that rewriting legacy engines in Rust provides an illusion of safety, as performance-critical assembly hot paths bypass Rust's memory boundary checks regardless, requiring verification tools like checkasm instead. Corporate Extraction, Maintainer Burnout, and State Resistance (77:33) The open-source multimedia infrastructure operates on a precarious human foundation, with core maintenance falling on a tiny volunteer pool—roughly five maintainers for VLC and ten to fifteen for FFmpeg. Despite this limited capacity, trillion-dollar corporations regularly treat community bug trackers as zero-cost enterprise vendor pipelines bound by SLAs. In one cited incident, Microsoft Teams staff demanded high-priority fixes for an issue affecting Microsoft products, yet offered only a small, one-time payment of several thousand dollars when maintainers requested a long-term maintenance contract. This labor strain is intensified by security research practices, including automated "AI slop" bug submissions and the systemic inflation of trivial glitches into high-priority CVE vulnerabilities on decades-old, obsolete codecs. Beyond corporate friction, the project has actively resisted state-level interference; VideoLAN leaders previously rejected requests from two separate intelligence agencies seeking backdoors in VLC media player, establishing that the team would sooner shut down the project entirely than compromise its software integrity. Navigating Patent Minefields Toward Ultra-Low Latency (224:00) Video compression relies on exploiting human biology by discarding invisible color data in the YUV color space, with each successive generation (from MPEG-2 to H.264, HEVC, and AV1) targeting a 25% to 50% bandwidth reduction for equivalent visual fidelity. However, deployment remains heavily restricted by patent minefields. The fractured, uncapped licensing pools surrounding HEVC forced device manufacturers to strip support and prompted streaming platforms to back royalty-free alternatives like VP9 and AV1, requiring open-source developers to exercise meticulous legal and technical creativity to avoid active patents. Beyond standard distribution and offline playback, open-source development is expanding toward teleoperation via Project Kyber. Built to support real-time human feedback loops for remote surgical equipment, drones, and autonomous vehicles navigating edge cases, Kyber bypasses TCP head-of-line blocking by multiplexing synchronized video, audio, and gamepad inputs over QUIC and UDP. The platform aims to compress end-to-end network latency down to an industry target of four milliseconds on low-cost computing hardware. The sustained independence of FFmpeg and VLC highlights an unresolved paradox in modern technology: global digital media and cultural preservation rely on low-level open-source tools maintained by a handful of unpaid or underfunded engineers. While these developers continue to advance compression standards and low-latency robotics protocols through strict meritocratic code review, their long-term resilience remains constantly threatened by corporate free-riding, legal intellectual property barriers, and maintainer exhaustion.

Jean-Baptiste Kempf 与开发者团队阐述了开源多媒体项目 FFmpeg 与 VLC 的技术基石、工程哲学及现实困境：这一生态系统凭借对代码质量和手工汇编性能的极致追求支撑着全球视频基建，却在万亿美元级企业的索取、安全警报滥用和极高维护负担下艰难维系脆弱的人力平衡。 视频压缩的生物工程与技术演进 视频数据压缩的本质是针对人类视觉感知特性的主动信息降解。与追求完整还原的无损压缩不同，现代视频编码依赖将原始数据缩减 100 到 200 倍。工程实现上采用 YUV 色彩空间分离明度与色度，利用人眼对色彩细节敏感度较低的生物学弱点对色度进行下采样，在人眼难以察觉的前提下直接削减近半原始文件大小，随后借助离散余弦变换等频域处理进一步提升效率。从 H.264 到 AV1 的各代标准均沿袭反变换、帧内预测、运动补偿及熵编码的核心框架，但每代技术均实现约 25% 至 50% 的压缩率提升。在分发体系中，I 帧作为完整静态图像，P 帧记录单向预测差异，而计算最复杂的 B 帧则依赖前后双向预测数据，满足从流媒体自适应分发到专业级无时间压缩 ProRes 剪辑的广泛需求。 极简硬件优化击败编译器主流神话 在视频播放必须在严格的 16 毫秒窗口内交付下一帧的严苛环境下，低级系统优化关乎软件生死。行业内普遍认为现代高级编译器配合自动向量化足以取代手写汇编，但现实数据打破了这一技术假设：在性能关键的代码热路径中，手写 SIMD 汇编的执行效率可达到标准 C 代码的数倍，特定场景下差距甚至高达 62 倍。以 AV1 软件解码器 dav1d 为例，其代码库包含 79.9% 的手写汇编语言、19.6% 的 C 语言以及 0.5% 的其他语言。面对当前转向 Python、JavaScript 乃至依托大语言模型的“氛围式编程（vibe coding）”潮流，维护者指出硬件级别的极致优化无法通过高层抽象替代，理解 CPU 流水线、SIMD 矢量处理及算术逻辑单元等计算机体系架构，依然是确保全球每日数十亿小时播放能耗经济性的基石。 双星共生架构与极端容错设计 VLC 与 FFmpeg 之间并非竞争对手，而是一种类似“Android 与 Linux”的互利双星系统。VLC 凭借跨终端的广泛普及为底层的 FFmpeg 提供海量真实世界边缘案例与极端文件输入，同时提供资金与代码反哺；而 FFmpeg 则作为基础音视频库驱动着包括 VLC、Google Chrome 及 YouTube 在内的全球数字媒体基础设施。为了在不可靠的网络和早期数字盗版时代生存，VLC 从诞生起便确立了“不信任输入”的防御原则，主动忽略三字母文件后缀名，直接通过探针探测文件有效载荷。即便遇到帧分辨率与旋转角逐帧动态跳变的极端 MKV、完全由 ASS 字幕渲染构成的视频，或是缺失结尾索引元数据的残缺文件，播放器也能绕过标准错误并持续解析渲染。 开源治理、专利雷区与商业寄生困局 (77:33) 媒体工程因密集数学算法成为与蜂窝通信并列的“专利雷区”。MPEG 系列标准后期因 HEVC 许可池碎片化和无上限专利费引发行业反弹，促使科技阵营转向由开放媒体联盟打造的免版税替代方案 AV1，开源开发者甚至需要通过“双重创造力”微调算法来规避专利侵权。在许可机制层面，libVLC 核心引擎从 GPL 转向 LGPL，消除了外部应用和商业游戏引擎集成时被强制开源整体项目的法律障碍。然而，开源社会契约正遭受商业机构的单向索取：万亿美元级科技巨头在云端封装开源工具对外收费，或如微软团队在公共漏洞跟踪器上要求志愿开发者遵守其企业级服务等级协议（SLA），却仅提供象征性的几千美元单次补偿。此外，成千上万的历史贡献者使得 FFmpeg 在法律上无法效仿数据库厂商通过修改开源许可证来自保。 维护者生态耗竭与安全警报的异化 (73:15) 尽管全球媒体网络依赖 FFmpeg 和 VLC 运行在数以亿计的 CPU 之上，其实际维护工作却仅由十多位无偿核心志愿者维系。这一极其脆弱的人力基础正在承受自动化安全工具与虚假威胁的双重冲击。近年来，安全行业大规模利用人工智能脚本和模糊测试工具挖掘边缘漏洞，将极其罕见的冷门编解码器轻微故障或单个像素显示溢出冠以高严重级别漏洞评分（CVE），导致维护者面临如同遭受拒绝服务攻击般的“狼来了”警报泛滥。这一系统性枯竭曾具体演变为 XZ Utils 供应链投毒事件——攻击者通过高压消耗独力维护者以获取提交权限。维护团队同时面临着因终止旧平台支持而遭遇包含白色粉末的实体死亡威胁等极端网络暴力，保持项目生机的关键往往落在了不计名利的年轻汇编开发者与纯粹编码热情的极客精神之上。 该节目系统性展示了现代互联网视频流媒体的技术底层——它并非建立在无瑕的商业契约之上，而是依靠极少数开源开发者以手工汇编、逆向工程和道德自律构建的脆弱工程奇迹；在面临专利围剿、万亿级企业白嫖以及 AI 自动化滥报的持续挤压下，这一关键数字基础设施的可持续性与安全性依旧缺乏制度性保障。

Jean-Baptiste Kempf: We care about excellent code. We don't care who you are. Like maybe you're a dog. I don't care. Jean-Baptiste Kempf: If we had to compromise our software, we would shut it down. This is clear. Kieran Kunhya: regrets are attacks on your mind. [70:05] Kieran Kunhya: building architects that never have to go to site [77:33] We are not your supplier. If you want me to be your supplier... I will send you a contract and SLAs. [83:30] Teenagers have written more assembly in FFmpeg than Google engineers. [90:00] John Collison: museum of passion projects [90:00] Tim Sweeney: Many things in the world only happen because an awesome person decides to do it. This is the case with VLC. [229:00] If we had to compromise our software, we would shut it down. [253:40] Talk is cheap, send patches [253:40] FFmpeg, nothing is beyond our reach [256:06] Je ne regrette rien [256:06] Linus Torvalds: Most good programmers do programming not because they expect to get paid or get adulation by the public, but because it is fun to program.

[0:00] - The important is, is your code good? We care about [0:03] excellent code. We don't care who you are. Like maybe you're a dog. I don't [0:07] care, right? I don't care where you come from. I need to look at your code. [0:11] Oh, yeah, but I'm an engineer at this very large company in [0:15] Italy, in Germany, in the US. We don't care. We care about the [0:19] quality of your code because this is what defines our community [0:23] and which means that we have a lot of people who contribute who are some very different [0:27] backgrounds and very introverted. Sure. But that's okay, right? [0:31] - FFmpeg is probably one of the biggest CPU users in the world. Everything we've just [0:35] said in the past couple of minutes, every sentence is someone's [0:39] lifetime's work. There are books about every sentence. So the level of complexity [0:44] in many cases is inordinate. [0:45] - FFmpeg has one hundred thousand lines of assembly for all the codecs. [0:50] - For all codecs. Mm-hmm. [0:50] - And just this one has two hundred and forty thousand. Every cycle [0:55] matters. We are talking about probably three billion [0:58] devices which are going to decode video nonstop because, for [1:02] example, thirty percent of the video from Netflix is now in AV1, [1:06] fifty percent of YouTube. [1:07] - This is what peak video codecs should look like. [1:11] Seventy-nine point nine percent assembly, nineteen point six percent C, [1:15] and zero point five percent other. [1:18] - And what's incredible is with those tweets, which is factual, people get crazy. [1:25] - For the last two years, they go crazy. No, intrinsics is fine. The compiler- [1:28] - You can optimize your compiler. Auto-vectorization, it's your fault. You don't [1:32] understand. And we've tried that forever, right? [1:35] - For two years, and two years later, showing hundreds of examples [1:40] of handwritten assembly. No, no, no, you're doing it wrong. The compiler can do this. [1:43] The intelligence agencies tried to, like, say, "Can you put a backdoor in VLC?" [1:48] - Yes. Two of them. [1:50] - Well, what did you say? [1:51] - No. Well, I was a lot less polite. [1:54] - Basically saying, "Hell no." [1:56] - Like, if we had to compromise our software, we would shut it down. This is clear. [2:00] - Any tweets Kieran, you regret? [2:04] - Tweets I regret? [2:05] - Or is it like that, how does the French song go? Regret nothing. [2:08] - Don't regret anything. No, it's because regrets are attacks on your mind. [2:17] - The following is a conversation all about FFmpeg and VLC [2:21] with Jean-Baptiste Kempf and Kieran Kunhya. [2:25] FFmpeg is an open source software system that is [2:30] the invisible backbone behind YouTube, Netflix, Chrome, [2:33] VLC, Discord, and basically every platform that [2:37] touches video or audio on the internet. [2:41] It can decode, encode, transcode, stream, and [2:45] play almost any video or audio format ever [2:49] created. To me, it is one of the most incredible software [2:53] systems ever developed, and it's all done by volunteers. [2:59] VLC is also a legendary piece of software. It is an [3:03] open source media player that plays basically anything you throw at [3:07] it, any format, any platform, no ads, no tracking. [3:11] It has been downloaded over six billion times, and [3:15] again, for me, it has been one of my favorite pieces of [3:18] software ever, with the most legendary logo, which I, [3:22] of course, had to honor in this conversation by wearing the [3:27] VLC traffic cone hat the whole time. [3:31] So again, above all else, thank you to the incredible volunteer [3:35] engineers who put their heart and soul into this code that has been [3:39] used and loved by billions of people. Thank you. [3:43] And about the two great engineers and human beings I'm talking to in this episode, [3:50] Jean-Baptiste is the president of VideoLAN and [3:54] is a key figure behind VLC and FFmpeg. Kieran is a longtime codec engineer, FFmpeg [4:01] contributor, and the man behind the now infamous [4:05] FFmpeg account on Twitter/X that I recommend [4:09] everybody follow for the memes and for the [4:13] unapologetic celebration of open source and great low-level software engineering. [4:20] Let me also say that it's inspiring and humbling that [4:24] so much of modern civilization rests on software built by [4:28] people who are not chasing fame or money, but are [4:31] obsessed with the craft of engineering. [4:34] We live in a world where billions of people consume video every day without [4:38] ever thinking about the invisible machinery underneath it. But that [4:42] machinery matters. Open source infrastructure matters. [4:46] It is one of the great examples of human beings quietly collaborating [4:50] across borders to build something useful, durable, and elegant for the rest of us. [4:57] And so this conversation is not just about codecs and [5:01] media pipelines. It is also about the deeper spirit of [5:05] engineering and generosity that makes projects like FFmpeg possible. [5:11] Again, I can never say it enough. Thank you. [5:16] This is the Lex Fridman Podcast. To support it, please check out our [5:20] sponsors in the description, where you can also find links [5:23] to contact me, ask questions, give feedback, and so on. [5:28] And now, dear friends, here's Jean-Baptiste Kempf and Kieran Kunhya. [5:35] So the legend goes VLC can open everything. What's the [5:39] weirdest thing that you know that it can open? [5:43] - You know, there is a ton of people who are using VLC to record VHS [5:46] videos, right? Like, it's just like you plug it with a capture card and you [5:50] can basically record VHS video. [5:52] - Well, how does that work? [5:53] - Basically, it's, you know, those type of capture card where you can put a Peritel [5:57] in or- ... or RCA, and you put that, and actually VLC can play those type of cards, [6:02] and there is a module which allows to control directly some of those [6:06] VCR camcorders. We support DVD audios lately, right? We spent [6:10] the summer working on DVD-Audio support, and [6:14] like there is no, no one's making any DVD audio support. There is a custom [6:17] encryption schemes. [6:19] - What about Lucasfilm? [6:20] - Oh, yeah, and there is of course all the weird codecs support, game codecs supported [6:23] by FFmpeg. [6:25] - The one Star Wars video game, the first ten- second opening sequence, someone has gone and [6:29] implemented that and made sure that's bit exact on one disc that existed at [6:33] one time of one little sequence in the game. [6:36] - And then funnily, there was a... At one VideoLAN conference, we made a [6:40] competition to make the weirdest and most horrible file ever [6:44] ... and see if VLC could play it. [6:46] - What did it end up being? What's the file? [6:48] - It was an MKV file made by Derek- [6:52] ... which each of the frame was changing resolution, aspect [6:56] ratio- ... rotation and it was like- [6:59] - Did it work? [7:00] - Yes. And there was another one where [7:03] the whole video was actually animated subtitles, right? SSA, right? So- [7:08] - Yeah. I remember that, yeah [7:08] - ... each, this one was- [7:10] And so each frame was a black frame, but on top of that there was a, [7:14] a subtitle that was animated for each frame. [7:16] - There was a file that's a valid ZIP and a valid MP3 at the same time or something like that, [7:20] so. [7:20] - So yeah, we'd made a competition of stupid files. [7:23] - And it worked. It opened all of the stupid files. [7:27] - Yes. [7:27] - By the way, [7:28] For people who are not familiar, I am wearing a hat. Would it be fair to say this [7:32] is the best worst logo of all time, the cone? [7:36] - Yeah, by far, right? The logo of VLC is so iconic, right? [7:40] Like we are a team with a small number of people and the icon [7:44] is known everywhere. I go to middle of nowhere in India or in [7:47] China, people know the cone, right? And 25% of [7:51] the website traffic that comes to our main website is [7:55] cone player, right? So, so many people don't know VLC, right? They know the cone player. [7:59] - That's the thing they Google for is cone player. [8:01] - Yeah. They go on Google and they put cone player and they download VLC, right? [8:05] So that's iconic. And once we tried to change it as a joke, right? We said [8:09] it was going to be a type of uh, [8:12] caterpillar construction and we said that during April 1st- [8:18] ... and we had around 10,000 emails saying, "No, don't change the logo," and so on, right? [8:21] So it's so iconic, right? It's so distinctive, right? If you want to do a [8:25] video player, you're going to put a play button on a TV, right? And that's a YouTube, [8:29] YouTube logo, right? It's unoriginal. This one is orange, right? [8:33] - Yeah. [8:33] - It's very bright and it's weird. [8:37] - And it's ridiculous and it's absurd and it's hilarious. It becomes meme and meme becomes [8:41] culture. Yeah. [8:41] - And you keep it and you know about it and you know that in 20 years, like you [8:45] still have, going to have the cones and remember, oh yeah, that was a video player. [8:49] - Yeah. And we'll talk about, you know, the mission of FFmpeg being [8:53] a kinda the archival aspect of it. So you can think about 1,000 years from [8:57] now we'll have all these videos that only VLC can open. Humans, [9:01] human civilization has already destroyed itself multiple times and the [9:05] only thing that will remain is this like, you know, the cockroaches will be [9:09] crawling around and it'll be the VLC logo- ... with some of the archival [9:13] footage that VLC can open. And the aliens will show up and they'll press [9:17] play and they'll get to see it all 'cause- [9:18] - Well, really, really hope so, right? But there is also so many memes where people say, "Well, I'm sure I can put [9:22] a pancake inside my DVD drive and VLC will play it." Like- [9:25] - Can they? [9:26] - No, we tried. It doesn't. Um- [9:27] - Doesn't. [9:28] - ... but we actually have a video of us trying that. Didn't work. [9:31] - A codec for physical reality, I don't know what that would even look like. [9:34] - There was a guy who did that, right? He printed a small cone, right? [9:38] Like the ones we distribute as goodies and inside he put an [9:41] RFID chip which was his way of playing a movie, right? And so he- [9:46] ... put this on a RFID player and when he put that it was playing like The [9:50] Last Star Wars and so on. So instead of having like DVD boxes, he had like [9:54] VLC cones all around and he plugged that and that was like physical objects. [9:59] - So the thing that we're talking about is everything around video [10:03] codecs, video encoding, video decoding, video streaming, video [10:07] player client that I'm wearing on my head, the entire [10:10] ecosystem enabling free media. We'll talk about FFmpeg, we'll talk [10:14] about VideoLAN, VLC and all the other incredible video [10:18] technology that is used probably by billions of people. So JB, [10:25] you're the lead developer behind the legendary VLC player. [10:29] Kieran, amongst many other things, you're lead developer behind the legendary [10:33] FFmpeg handle on Twitter. And both of you have spicy opinions I would say. [10:40] So today we wanna talk about FFmpeg and VLC. [10:45] For context for people who are not aware and I'm sure [10:49] basically everybody listening to this have used these two [10:53] technologies probably regularly without knowing it. [10:57] So FFmpeg underlies basically most video on the internet [11:01] including YouTube, Netflix, Chrome, Firefox, of course [11:05] VLC and countless other video platforms. It [11:08] is estimated that over 90% of video processing workflows online and [11:12] offline involve FFmpeg. VLC has been downloaded at least 6.5 billion times. But [11:22] likely that number, 'cause it's impossible to really count the number [11:26] is much higher than that. Virtually any operating system [11:30] supports virtually any media format. [11:35] The limitation being it can't open pancakes. So, [11:39] Can we just lay out some of the basics to help people [11:43] understand what's involved in all of this? So when we press [11:47] play on a video player like VLC, what happens? What-- [11:52] How does it go from the file or the stream to the [11:56] pixels on the screen and the sound on the speaker? What are the big [11:59] stages to be aware of? [12:01] - So there are several stages, right? The first stage is to get from [12:05] an address, right, which is the type of URL, to [12:09] give you a byte of streams, right? So this would be, for example, HTTP, [12:14] file, DVD, right? You give the path to the media, and it gives [12:17] you a stream of data. [12:19] - The stream needs to be cut up by what's known as the container, the demultiplexer or [12:23] demux. We'll try and keep the jargon light throughout this, but it [12:27] needs to go and start demarcating video and audio frames. So it just gets data from the [12:31] operating system blocks at a time and needs to start cutting these frames up [12:35] into compressed data. It then needs to start doing simple [12:39] parsing of the video frames- ... mainly to figure out whether that codec is GPU [12:44] decodable or needs to fall back to software. We're very sort [12:48] of used to assuming the GPU will play all of these things. There'll be hardware [12:52] acceleration. I think it's up to forty- five percent of files are not GPU [12:55] decodable. So these need to be probed. They need to be detected. There can [12:59] be variants of a given codec, some of which are decodable on the [13:03] GPU. Different vendors of GPU might have different capabilities, [13:07] so those need to be detected. So if it's GPU capable, you pass it through [13:11] to the GPU black box. So now if there's a software fallback, [13:15] that means in the beginning is to first do deentropy [13:19] coding, so removing the mathematical coding of the bit stream. So this [13:22] uses capabilities such as Huffman coding or arithmetic [13:25] coding to actually decompress the mathematical layer of the bit stream. [13:30] We then need to start reading the syntax elements for intra prediction. So intra [13:34] prediction are like still images of the video, so your I-frames. [13:40] So this works and operates in the spatial domain. So you do your intra prediction in spatial [13:44] domain. You have a residual because your prediction isn't quite [13:47] matching that of reality. So you've made a prediction, but then there's a little bit [13:51] left, and that's what's known as the residual. This is stored in the frequency [13:55] domain, and these are quantized to decompact their space. [13:59] We then need to do the inverse transform to bring them back to the [14:03] spatial domain and apply these residuals. [14:05] - So a lot of the process of the decoding is this thing is compressed. [14:10] And you have to predict the highest quality thing that's [14:14] supposed to go there. I-frame- ... is the best representation you have spatially. [14:19] And then there's a lot of temporal compression that can happen [14:23] depending on the codec, and then you're predicting. You're predicting what [14:26] the reality that was captured in this rawest form. [14:29] - Yeah, because what people don't realize is that the compression on [14:33] video and audio is [14:36] one hundred times, right? Like, people don't realize how compressed we, we [14:40] do, right? For audio, you move, you compress by, when you go from [14:44] normal audio to MP3, you compress by ten times, right? When, when you [14:48] move to video, you need one hundred times, two hundred times, right? So you need [14:51] to remove all the details, but that you don't care about [14:56] because all the compressions that we do, and that's very important, people forget about [14:59] that, is to be viewed by humans, right? [15:02] So all the codecs, either for audio, mimic basically how your [15:06] ear works, right? And a lot of things about, like, the [15:09] response on the ear and same for your eyes, right? And so, for [15:14] example, on video, we don't work on RGB, right? Everyone expects [15:18] to work in RGB. We don't, right? We move to YUV, which is [15:22] basically one is luminance, brightness, and the other are colors. [15:26] And this matches your eyes, where inside your eyes you have the cones and the buttons, right? [15:30] With some of them look on brightness and more on, on the other on colors, right? So [15:34] we need to compress a lot, and so we need to degrade. But in order to [15:38] degrade, we need to match the human perception, and this is why it's so [15:42] difficult. And then we need to use the maximum power, [15:46] mathematical power, very complex technologies. We move to the [15:49] frequency domain, as Kieran said. We do a ton of [15:53] dequantizing, in order to get the best compression, but it still looks good. [15:58] - You're trying to compress in order to maximize the highest [16:02] quality thing for human perception. [16:04] - That is correct. And this is very important, right? Compression is not like [16:08] a ZIP, right? A ZIP, you have data in, you get data out, [16:12] right? And you try with all the ZIP compression to arrive with [16:17] the limit. Here we are degrading the signal, right? And so we [16:20] need to degrade both the audio and the video signal in the best way [16:24] possible. And we can do that, but it involves, first, [16:28] a lot of theoretical knowledge about how the eye works, but [16:32] it, a lot of mathematical change, a lot of mathematical tricks, [16:36] right? For example, when you move to RGB and you do go to [16:40] YUV, for example, what we do very often is that [16:44] we scale down the resolution of the color compared to the brightness. [16:48] And most of the time, just this without compression, it [16:51] divides the size by two, but most people don't see it, [16:55] right? And so on and so on, right? And then you go [16:59] to very complex mathematical change. So of course [17:04] Fourier transforms, which de facto are not Fourier transforms, they are like [17:07] discrete cosine transform, but that's the same idea. So frequency domain [17:11] we split the video by blocks, right? So that's why when it's [17:15] wrongly decoded, you see those blocks and badly encoded, you see those blocks, [17:19] and so on, to arrive to compression states that are [17:23] insanely high, right? And each generation of the codec is like thirty percent less- [17:29] ... for the same quality, right? And this requires amount of [17:32] power of computational power that is huge. [17:35] - No, no, but you should elaborate. It's thirty percent better, but an order of [17:39] magnitude, perhaps, perhaps even two orders of magnitude more [17:42] compression power. That's the big difference. [17:45] - What do you mean by compression power? [17:47] - Sorry, CPU power to achieve that level of compression. [17:49] - Oh, yeah. So you have to be able to leverage the CPU and sometimes GPU, like [17:53] you mentioned. And then we should mention that a lot of this [17:56] programming is done at the lowest possible- [18:01] ... stack, whether it's C and of course, as the legendary— [18:05] ... Twitter handle re-emphasizes over and over, a lot of assembly. [18:10] - So what happens globally is that you have an address, right? Which gives you [18:14] with the operating system, a stream of bytes, a stream of data, right? And this is the first step. [18:18] And the second step arises with demuxing, where you're going to separate audio, [18:22] video, subtitle in type of different tracks. And then on each of those [18:26] tracks, you're going to decompress them, decode them, either audio with an [18:30] audio codec, video to video codec, and subtitle to subtitle codec. [18:34] And once you've decompressed those type of things, you have raw images, [18:38] raw, and then you're going to talk with your graphics card and your screen and [18:42] display that. And same for the audio, you're going to talk to your audio card, which [18:46] then is going to go in analog to your audio speakers. [18:50] - And everything we've just said in the past couple of minutes, [18:53] every sentence is someone's lifetime's work. There are books about- [18:56] ... every sentence. So the level of complexity [18:59] in many cases is inordinate. You know, it's, it's... Every sentence [19:03] has thousands of people working on this- [19:06] ... in industry as a whole, books written about [19:10] it. So there's a lot of detail, there's a lot of subtleties, there's a [19:14] lot of both academic and practical realities, both of which matter. [19:20] - Uh, we mentioned codecs, but I don't think you mentioned containers. [19:24] So what, what are the actual containers [19:28] for some of the stuff we're talking about? So people are familiar with MP4, [19:33] uh, MOV, MKV. So anyway, what are containers versus the thing that goes inside? [19:41] - So the container is what we call also the muxer, right? When I say demuxing, it [19:45] means decontainerizing, right? So actually, if you look, [19:49] mux means multiplexer and demultiplexer, right? [19:53] Mux and demux are those. And same, a codec is actually coder, decoder, right? [19:58] Um, and so containers are this collection [20:02] of multiple tracks, right? So it's a, what normal people call the file [20:06] format, but it's a bit more, um, subtle than that. [20:10] But the most known one, of course, is MP4, but when I started, [20:14] it was AVI, right? AVI was the, the video format from- [20:17] ... from Microsoft, and MOV, M-O-V, which became [20:21] MP4, was a format from Apple. In the open source community [20:25] one of the person that is still active on VideoLAN is called Steve Lhomme and started, [20:29] This Matroska format, which is, like, a bit more complex and [20:33] more future-proof. And there are so many others. [20:38] - So, I mean, there's a, it's a pretty common thing, and maybe it'll even happen in [20:42] this conversation, that people confuse container [20:45] and the codec, right? So confuse MP4 and H.264, for [20:49] example. Is that a horrible violation? [20:51] - No, it's not, because technically the name of H.264 is MPEG-4 Part 10. Because [20:59] MPEG-4 is actually a meta specification which has [21:03] several things in it, right? There is the Part 2. [21:07] so there is, like, audio codecs, right? AAC de facto is MP4 audio- [21:11] ... something. There are actually several video codecs, right, inside the [21:15] MPEG-4 specification. One of them is MPEG-4 Part 10, [21:19] called also AVC, called also H.264. Right? So [21:24] it's completely the fault of the industry to make things difficult [21:28] to understand. So that's very difficult so that people [21:31] then don't understand why sometimes you talk about MPEG-4 Part 10, [21:35] where you mean H.264, and why it's not MP4. [21:38] - So you can technically shove in all kinds of different codecs inside [21:42] containers and horribly so. [21:44] - But broadly speaking, though, MP4 is understood to generally [21:48] be H.264 plus AAC audio. 99% of the time that's that, and that, [21:55] the rest are de minimis, the small effects, you know, edge effects really compared to that. [21:58] So it's not the end of the world. There, there are people [22:02] who do get annoyed by that. But also in reality, something like VLC, just to point out, [22:06] the file may say .MP4, but it may be something completely different, and that's [22:10] one of the challenges both FFmpeg and VLC have is the real world is a [22:14] completely different place to a three- letter file format. [22:17] - And this is very important to say, right? Like, for example, in VLC and in FFmpeg, [22:21] we discard the file format, right? We look [22:25] into the file to understand what's in it because so many [22:28] people, like, they say, "Oh, it's a video, it should must be MP4," but technically it's [22:32] an MOV or maybe it's a MKV, right? So we [22:36] analyze in real time everything that we have, and we don't trust- ... the format. [22:42] - So what information does the fact that it's .MP4 give you? [22:45] - It helps, right? It gives you a hint, right? Just like, oh, it's [22:49] finished by .MP4. I will start first by opening, [22:54] probing it with the MP4 container demuxer [22:58] to see, well, it should be that. But I don't trust it, and if I'm lost, I say, "Okay, [23:02] maybe I'm going to try it." So it bumps the priority of the module. [23:06] - So how do you get to... just to take a bit of a tangent there. [23:11] You know, the dumb thing is if you try the MP4, [23:15] but it turns out it's a different codec than you would have expected, [23:19] Most players just break there. [23:21] - Yes. [23:22] - Yes. [23:22] - And so how do you not break? There's just philosophically, I'm sure [23:26] there's a bunch of stumbling blocks along the way where it's [23:30] easy to just break and stop, freak out. That's it. How does VLC not? [23:34] - This is why VLC is popular. But the reason is because [23:38] actually VLC was, is just a client of a streaming [23:42] solution called VideoLAN from a very long time [23:46] ago, from the late '90s. And when you're playing video [23:50] which are on UDP, right, in network, they might be damaged, right? [23:54] So you don't trust your inputs, and this is very important into the security is that you [23:58] don't trust your inputs. So everything in VLC is prepared to work with broken files. [24:06] And it's a philosophical idea from the beginning, and everything [24:10] is engineered into that. And it's a culture, right? And so, [24:14] for example... And VLC became very popular on that because a long time ago when people [24:18] were pirating content which they do a lot less today- [24:23] - And none of us ever have- [24:24] - No, of course not. Um— ... the metadata to play some files like AVI is at, [24:30] at the end of the file, right? And when you're downloading, you don't have that, right? [24:34] So VLC was just like, "Hey, this file is broken, but I'm still going to try to [24:38] interpret it," and this was very useful. [24:41] - We hinted at the awesomeness of the various different stages. [24:45] We hinted at the awesomeness of codecs, the depth and the richness and the complexity [24:49] of everything involved there. What— Let's try to define [24:53] what is a video codec? What's involved there? What does it mean to [24:57] compress something? You already started to hint at it— [24:59] ... but can we elaborate a little bit more? [25:01] - So there's a huge amount of redundancy in any video both [25:05] spatial and temporal, and the point of any video codec is to remove [25:09] this redundant data, use mathematical properties as part of this reduction [25:13] process. So more often than not, using several orders of magnitude more [25:17] compute to compress because that's more costly versus both costly [25:21] both financially and in CPU resources— [25:23] ... versus the decompression. So it's asymmetric in that respect. Uh, [25:27] often the case because compression is done once, but there could be lots of viewers of [25:31] another file. So to take that information and compress it by [25:36] 100x, 200x, removing redundant information and [25:40] using mathematical properties to make that small, but also have properties [25:44] such as error resilience. So as, as JB suggested, [25:48] VLC in the beginning was, was used to play UDP network feeds, and UDP [25:51] network feeds lose packets. And so some of the design goals of a [25:55] codec is also to be recoverable. [25:58] You need to actually be able to join a stream. It's not necessarily a file. You need to join, [26:03] get on the decoding process, and start decoding. [26:06] - And, and to give a more image to people [26:10] who are not familiar, right? Like, when you're going to see any type of movie, right? [26:14] You're going to see the camera is going to pan, right, and travel. And [26:18] you realize that, for example, all the background is the same from, for, like, [26:22] a minute, right? Or— [26:23] ... thirty seconds, right? So you can reuse the cloud that you see uh, on [26:27] the background, you can reuse that from a frame to another, right? [26:31] And so it's, gets the more, the more [26:35] memory you have, the more power, the more comparisons you can make, right? And so [26:39] the more compressed you can be. And most of the modern codecs are [26:43] basically doing that. [26:44] - So just to make it even more explicit. So what is video? Video is a [26:50] bunch of pixels often RGB. [26:54] You have three values, and you have a grid of pixels, and you have, [26:58] let's say, twenty-four or thirty or sixty, [27:02] frames a second, and you just have all these pixels [27:05] repeating and showing different stuff- [27:08] ... thirty times a second. And so the question, the philosophical, [27:12] the technical question is, how can I compress all of that, [27:17] store all of that at 100x? [27:20] - Yep. Or 1,000x, right? [27:22] - 1,000x. [27:22] - The target is 1,000x, right? [27:24] - And the goal is when you say redundancy, what is redundant? Meaning stuff at best [27:32] that humans wouldn't notice if it was missing. [27:35] - So for example, you have a picture of a cloud, right? And from the [27:39] next frame, they're still going to be the same cloud, so it's redundant. You could just put it once and [27:43] not do it, right? Or you have a black background behind me, for [27:47] example. The black is the same on the whole picture, right? So you can say, "Well, you know, in [27:51] this picture, take the pixels that you have on the top left and the one [27:55] on the top right. I'm not going to give the value. I'm just going to tell you it's the same at the top [27:58] left." And then you can say for frame one reuse [28:02] something from the previous frame or the previous, previous frame, and so on and [28:06] so on, right? So you could... Basically, it's [28:10] unlimited, but then it's limited in terms of memory or in terms [28:14] of compute power. Because, for example, if you need to compare pixels [28:17] on two hundred frames in the past on 4K resolutions, it's a huge amount of compute. [28:26] - And then when you're showing it, you have to do the decompress of all of that. [28:30] So is it the codec, the, has the [28:33] encoding and the decoding is a coupled process that you're developing? [28:37] - Yes, exactly, right. And those are two different [28:41] trade-offs, right? Are you going to compress more? But then it might be [28:45] more difficult to decode. Are you going to [28:49] comp- to make it a codec that is more complex to encode and easier to [28:53] decode? Are you going to make a codec that is easier to encode because you need to be [28:56] fast, but then the, the client side, the, the player is going to spend more [29:00] time? That's why you have so many different type of codecs, is that it's not [29:05] always easy. And to make it even more complex, modern codecs [29:09] like AV1, AV2, or VVC are actually not codecs. They are [29:12] a collection of tools, right? There are multiple tools, multiple [29:16] codecs in the same codec to, depending on the image, get the more compression. [29:21] - So just to elaborate, codecs like AV1, VVC [29:26] have a much wider, have a wide audience. It could be a screen share [29:30] content, it could be video, it could be animation. All of these [29:34] require different coding tools. So [29:38] what happens these days is a collection of tools are put in and called [29:42] AV1 and called AV2, called VVC to allow for different use [29:45] cases. So you may be on Zoom and sharing your PowerPoint, [29:49] and then you need to show the audience a video. That codec needs to [29:53] start changing its tool set depending on the content to compress in a different way. [29:59] - And like you said, there's a bunch of incredible engineers behind each part of [30:03] that, each part of the tools that make up AV1, for example. [30:06] Uh, so we've kind of danced around it. We talked about VLC, [30:10] the logo, the hat. Let's talk about FFmpeg. What, what is FFmpeg exactly? [30:17] - FFmpeg is basically the low-level libraries for codecs, so compressions [30:26] and decompression, muxers and demuxers, and filters. It's— [30:30] The core is this, and then you have several tools which allow you to create a [30:34] type of pipeline to process any type of video files. [30:38] And it's used as a library absolutely inside [30:41] everything from VLC to Chrome to your smart TVs, to [30:46] basically any video that you see online you usually use [30:49] FFmpeg. And FFmpeg in it has all those [30:53] type of tools, and sometimes depend on other libraries like [30:57] x264, libvpx, and others, right? So it's really now the de facto tool to process [31:05] images. [31:06] - From a philosophical level, I think it's incredible that your home [31:10] videos, your grandmother's home videos and trillion-dollar corporations [31:15] effectively are on a level playing field using the same technology stack. [31:21] It's— it wouldn't be a surprise, you know, these big companies just have three [31:24] thousand-line FFmpeg commands. [31:27] There are some that use the API, but there are some that just have long command lines. [31:31] - So yeah, there's a bunch of tools, like literally command line tool, [31:35] FFmpeg, of course, FFprobe. There's libraries, libavcodec, libavformat, libavfilter. [31:43] But the FFmpeg on the command line— [31:47] is, like, legendary because you can cut— Like, there's so many [31:51] parameters. You can customize everything to hell. [31:53] - It's a language. It's an actual language. [31:55] - It's an actual— yeah, you could think of it as a programming language. [31:57] - Yeah, of course, I'm sure. Because— so most of the people, they're going to take FFmpeg, [32:02] file in, file out, and specify the format, right? But you can-- We've seen [32:07] thousands of characters, and we've seen also, like, people, like, doing [32:11] programming generation of command lines [32:14] to make FFmpeg. There is a ton of people who are using AI to generate [32:18] command lines for FFmpeg because you have no idea what it is. But you can do- specify so [32:22] many filters right on the command line, right? So [32:26] FFmpeg is this collection of toolbox for multimedia processing that everyone uses. [32:34] And everyone that is watching your videos is also using it, right? You're on [32:38] YouTube. Well, it's FFmpeg on the client side. Well, the your [32:42] server side, on the server side. The client side is probably Chrome. Well, you're using [32:45] FFmpeg also. [32:46] And you're using OBS to record. Well, it's FFmpeg, right? You're using a ton [32:50] of important, like, big box, professional boxes. Well, it's very [32:54] possible that inside some part of FFmpeg is running. [32:57] - I mean, there's like so many, just to give people an idea, like I use [33:01] FFmpeg a lot on, on everything. Just trivial stuff like, [33:06] Take a video, add an intro video and an outro [33:09] video, and fade one into the other like what is it [33:13] called? Dip to black, like where it dips and then shows [33:17] the next video and does the same thing with audio. There's [33:21] like a cross dissolve of the audio. It's quiet, it quiets the audio and [33:25] makes it loud again. And then there's a bunch of stuff like [33:29] showing the captions on screen card, like baking the [33:33] captions in. You can customize the font. [33:36] You can do all kinds of layering of audio and video. There's a [33:40] million things and of course, all of that works [33:45] like magically with basically any codec. [33:49] Like anything you can shove in on the audio and the video side, it works. [33:53] - But it's like if you look at, for example, [33:56] you can do things that you would do with Adobe After Effects- [33:59] ... in command line on FFmpeg, right? It's, and it's very interesting because, for [34:03] example, for imaging, there is not such tool. [34:07] There is a few tools, but not with the breadth of FFmpeg. [34:10] - So ImageMagick has a similar kind of- [34:13] - Yes, but you will not- [34:14] - ... spirit, but it- [34:14] - ... do some filters, complex filters. You don't have the equivalent of Photoshop- [34:18] in command line, right? But for video, you have FFmpeg in command line. [34:22] - Yeah. It's incredible. I mean, it's like an example of a thing when a [34:26] bunch of great people get together and they get a vision, and they [34:29] stick by that vision for many years, which is incredible. [34:33] - And the vision behind, and the same for VLC and FFmpeg, is that [34:38] we make everything that is very complex easy to use for the normal [34:44] people, for everyone. [34:45] Right? Our goal is to make something that is insanely complex technically [34:49] and make it easy to use, right? And people, they use VLC, they [34:53] drop a file. They don't realize how complex the file is, but they [34:57] play it. Or, or people put any type of thing inside FFmpeg with [35:01] complex filters, and it just works like magically, right? And people- [35:05] And this is our mission, right? Make very complex things. [35:08] - We wouldn't be here and you wouldn't be here if this [35:11] required, you know, a traditional television studio setup. [35:16] It's tools like FFmpeg that democratize this. The podcast and [35:19] streaming revolution, the YouTube revolution- [35:22] was caused. You know, FFmpeg was a big player in that because it [35:26] democratized this technology that was once in the nineties, for example, you needed [35:32] equipment that cost hundreds of thousands of dollars to do compression. It was the size of a car, [35:37] and now everybody has that at almost an exact level playing field, and [35:41] that's something that's so remarkable. [35:42] - It gave voice to a lot of people. And just to clarify, we say you, [35:46] you wouldn't be here, not the human, but the podcast. [35:49] - The podcast. Oh, sorry. You as a... Sorry. [35:50] - I would still... VLC did not have anything to do on a biological [35:55] level- ... at creating me as a human. [35:56] - But, but it's like you realize also everything moved from text to images [36:00] and images to video, right? [36:02] Look at social networks. Video is everywhere. It's the most [36:05] powerful medium there is, right? And when you see [36:09] shorts and, and in Reels and in TikTok, right? It's [36:13] amazingly powerful to give... Video is amazing for [36:17] that, right? But the complexity is important. [36:20] - This is what people don't realize. I mean, this is really it [36:24] gave power to the individual all across the world. That's real [36:28] freedom. And I think, I can't believe it, but we still haven't mentioned the [36:32] actual obvious thing for people who are not familiar, which [36:36] it's open source, and there's a open source [36:39] community of users and developers behind it. So [36:43] it's really, it's a movement. So, like, we'll talk a bunch in a [36:47] bunch of different ways about the community behind it. But can you speak to the open [36:51] source element? So when we say what is FFmpeg, it's an open source project. [36:57] - Yeah. So FFmpeg, VLC, x264, VideoLAN, everything [37:01] we do is fully open source. And for the people who don't understand how open [37:05] source is, my usual analogy is about a chocolate cheesecake, right? Um, [37:09] usually for you, when you want to buy your cheesecake, you go to a bakery, they give you the [37:12] cheesecake. The other one way of having a cheesecake is have your [37:16] grandma give you a recipe of how to make that. When we do open source, [37:20] we give you the chocolate cake, and we give you the recipe to [37:24] actually remake the same cake, but at the same time tell you how [37:28] to build the oven and also how you're allowed to modify [37:32] the recipe and resell it to someone else. [37:35] And this is because software is just a very long recipe of small [37:39] instruction. Computers are not very clever. They go very, very [37:43] fast. So a normal program has tens of billions of instructions [37:46] instead of the tens when you have your chocolate recipe. So [37:50] a lot of the software industry was about selling software, like where you [37:54] just have like the final cheesecake. In open source, we [37:58] give you everything, and that managed to get a [38:02] lot of people work together, right? Because then you decide that you're going to [38:06] make the best program, the best recipe for video, and you create [38:10] communities. In FFmpeg, since the beginning of FFmpeg, probably [38:15] two thousand to three thousand- [38:17] - In the thousands, yeah [38:17] - ... people have contributed from the beginning, right? And then it's exactly like the [38:21] Linux kernel, right? The Linux kernel has probably ten thousand people [38:24] contributing everywhere, and they get together, well, [38:28] mostly online, right? So they virtually get together to create the [38:32] best tool for something. And on FFmpeg and VLC, it's just like, [38:38] well, this codec doesn't work, so I'm going to work on the codec, and I'm going [38:42] to add the support for this file inside FFmpeg, so it will be [38:46] beneficial to everyone. Because again, we work for the greater good. We work for [38:51] everyone, and that is what open source is. [38:54] - And we should mention, depending on the licensing, [38:57] You could probably build a billion-dollar, maybe even a trillion-dollar company [39:04] around basic... as a wrapper to... [39:07] - Well, yes- ... people do. [39:09] People do, right? There was a lot of problems with mostly, [39:13] Cloud providers who are basically running some open source tools, [39:17] In the cloud and just give you the API to [39:21] access to that. And there was a lot of um, [39:25] databases like Mongo or Elastic who changed their license in order [39:29] to avoid those type of scenarios. [39:31] - This is a question we get a lot in FFmpeg is, "Why don't you do that?" [39:35] And you can't. We have, we have thousands of contributors, some of whom aren't even alive anymore. [39:40] It would need all of their agreement to do that, and JB will go maybe a bit later [39:44] and talk about how challenging that process was in VLC to do the re-licensing. [39:49] - The license is a social contract in terms of Rousseau de [39:53] facto of the community. The community does [39:57] not agree on much beside the license. People go [40:01] around, discuss around because of the license, and that [40:05] also allow those license fork, right? Sometimes the community splits, but [40:09] it's possible because of the license and to merge back. [40:12] And we've seen that so many times, right? GCC and GC, [40:15] And EGCS in the past. We have seen, for example, all the web [40:19] browsers, right? They started as web, like KHTML, which becomes WebKit and [40:23] then which becomes Blink, right? So open source [40:27] license is like the core of the community [40:30] and people are coming from all around the world, very different type of religion, [40:37] Political borders. They work [40:41] in the same way on a project to solve a specific problem, [40:45] and the specific problem we're working on is to make multimedia easy for everyone. [40:50] - Uh, looking it up on Perplexity here, looking at the different open source licenses. [40:54] Most major open source licenses fall into two buckets: [40:58] permissive, very few conditions, and copyleft, share-alike [41:02] requirements for derivatives. Below is a brief practical summary of the [41:06] main ones you'll see in the wild. MIT license, BSD, ISC, Apache, GNU GPL, GNU AGPL. [41:20] Where's LGPL? Yeah, LGPL. Let's see. [41:23] There's the Mozilla Public License. There's Eclipse Public License. It [41:27] goes on. There's a lot of variety. I mean, I think the really popular [41:31] ones is MIT, GPL, LGPL- [41:34] - Yeah. And BSD. BSD [41:36] - ... and BSD, Apache. Sometimes you'll see- [41:38] - Apache as well [41:38] - ... Apache. Unlicense, that's an option. Attempts to dedicate code to the public domain with a fallback [41:42] permissive license. [41:43] - There are many licenses for many different things. What people don't understand [41:47] that public domain is something that doesn't exist worldwide, right? [41:51] So it's all the open source licensing [41:55] use the copyright law, right, the international copyright law, in order [41:59] to give rights on how you use the software or how you [42:03] modify. It's de facto a copyright license [42:06] contract that you give to the end user or to the developer. And [42:10] so you have like the first one, which are basically very permissive, MIT, BSD. [42:14] You give the code and basically you do whatever you want, right? [42:19] You take it, you want, you modify, you do what you want. And this [42:23] is popular for JavaScript and the type of BSD operating system. [42:29] - So some of them, one of the parameters is whether they require attribution, [42:33] meaning if you use the code, you have to say- [42:35] - Yes. So in those types of permissive licenses, some you need to say if you [42:39] use it, which is called attribution, and some you don't. And then there is [42:43] the other part of license which are copyleft, [42:46] where you need to give back to the community your [42:50] modifications and with different strings attached. [42:53] some weak copyleft license, like the Mozilla Public [42:57] License, to some which are a bit stronger like a [43:01] GPL, or even very strong like AGPL. So all of [43:05] those are different types of licensing that depends on what [43:09] your goals are and how you want to structure your community, which is why I spoke about [43:12] social contract, because this is very important to understand. [43:16] FFmpeg and VLC are mostly GPL or LGPL. The Linux kernel is GPL [43:24] but Android is Apache. A ton of JavaScript frameworks that are using are mostly MIT. [43:32] All the BSD kernels, OpenBSD, NetBSD are of course BSD. And so it's [43:39] philosophical change on how you want people to contribute back- ... basically. [43:44] - So I think you talked about that, you've moved at one point from GPL to [43:52] LGPL on certain parts of the project. What... Can you describe the difference [43:55] between the two, and what does it take to move to, I guess, a more permissive... [44:00] So that direction is more permissive. LGPL is more permissive than GPL. [44:04] - Yeah. So you have to realize that you can always go from [44:07] more permissive to less permissive, right? Because of course, [44:12] those licenses are basically statements, and so if you restrict, [44:16] you can always restrict more, right? So in a GPL project, you can take [44:20] MIT code, but you cannot do the opposite, right? Because they are more [44:24] constrained to match. Indeed, in fact, I changed the core [44:29] of libVLC, which is the engine of VLC- ... from GPL to LGPL. [44:37] And there were two reasons to do that. The first one is that so people can [44:41] use the VLC engine, libVLC, into [44:45] third-party applications. So a lot of applications which are playing video [44:49] on your phone or on your tablet are actually VLC engine in it- [44:52] ... which is calling FFmpeg in it. Um, so that was [44:57] one of the ways to create one of the companies I created, which is doing consulting [45:01] and integration of those types of applications where you integrate VLC [45:05] into third-party solutions like inside game engines or [45:09] stuff like that. With GPL, you couldn't do that because that means you needed to open source [45:13] everything, and those are for a lot of, like, commercial companies who don't want [45:17] that. [45:18] - So you can create a company with LGPL, you can create a company around it. [45:22] - Yes. [45:22] - You can do a commercial thing. You don't have to open source it. [45:24] So that's a big, big leap. [45:25] - So you could play video in your game. [45:27] - Yes. [45:27] - The problem is I'm a game developer, and I want to play some videos- [45:32] ... and I don't want to be forced to open source the entire game just to play those videos. [45:36] So that's where the consulting business, the libVLC LGPL- [45:40] ... allows you to do that. The LGPL, the library GPL as it used to be known, [45:43] allows you to do that. [45:44] - And FFmpeg is exactly the same. It force... LGPL [45:48] forces you to give back what you change on this component, this- [45:52] ... library, which is why it's library GPL. [45:55] And so you can use FFmpeg as LGPL into, like, [45:59] any type of application, even non-open source, but you need to give back the [46:03] modification you did on FFmpeg. Same on libVLC. [46:06] - Is it limiting from an open source perspective to go GPL? [46:10] Because if you-- if your library, if your code is GPL, it means you're not... [46:17] You're basically discouraging companies from building a business- [46:22] - Yes [46:22] - ... around it, right? Is that, is that fair to say? [46:23] - It depends on the company, but the company whose business model requires [46:27] the source, the application to be closed source, yes, it's limited. [46:32] So that's why, for example, I moved to LGPL. The second reason is a, [46:35] a bit more obscure, is that the terms and conditions of the, [46:39] App Store, the Apple App Store for iOS makes it very [46:43] complex to have GPL application on it, while it's easier to have [46:47] LGPL applications on it. So VLC on Windows and on Mac, [46:52] And on Linux is GPL. The core is LGPL. [46:55] But on iOS the iPhone version and the Apple TV [46:59] version is a type of different license called the MPL. [47:03] And yes, I went and changed the license and it was a long story. [47:08] - Yeah. So I think basically to change the license you have to contact all the [47:12] contributors. [47:12] - Yes. It's very important to understand that open source [47:16] projects are what we call in the US copyright law [47:20] joint work, or in civil law collective [47:23] works or collaborative works, is that you work all [47:27] together in terms of the same goal, and then you create one software, which is one release. [47:32] But the copyright is kept by all the individuals. [47:36] Some open source projects don't do that. They force copyright assignment, but this is not what we do. [47:40] We're communities. So everyone has basically copyright on what [47:44] they changed. And this copyright stays even [47:48] if at the end your contribution was deleted because the new [47:51] contribution was based on your previous one, right? So if you want to properly [47:55] re-license, you need to find all the contributors. And at that time, I [47:59] had to contact more than three hundred and fifty people. And [48:03] sometimes, well, they're just an email, right? So it's... you need to actually track down. [48:07] I actually, like, travel to some place to go someone that I [48:10] was like, sorry, that I'd found online to see a-- to [48:14] go to their job and say, "Well, you licensed that. Can you-- do you [48:18] want to change from GPL to LGPL?" Most of the times they don't even care. They wanted to [48:22] help VLC. But also it brought me to very complex situation. I [48:26] arrived to the work of a person who was a factory worker. [48:30] Um, and I said, "Well, I need to you to sign that," [48:35] because it was his son who died who actually wrote the code, right? [48:40] So I had to explain all those type of open source meaning, [48:43] and no, I was not a company trying to rip out the two line or five [48:47] line that that guy did- [48:49] ... but was useful, and the whole community agreed on that, and he had [48:53] no idea I was a factory worker. This com-- And I was a lot younger, right? Like [48:57] it was fourteen years ago, and like, like I was [49:01] almost in tears, right? It's very difficult, right? We are talking about lives of people and [49:05] he explaining, and we went talk about the photo of this guy, right? [49:09] So it's important to do it right and to do it correctly. [49:13] But yes, that means tracking down everything because every contribution works. [49:18] There are some project who don't respect that, and we do re-licensing a bit, like, [49:22] aggressively. [49:23] But as I said, it destroyed the whole heart of the community because it's-- we [49:27] only agree on the, on the license, so that's important. [49:31] - I would emphasize the community is such a wide-ranging group of people. [49:35] There's people in the Syrian war zone with electricity part-time. There's, [49:41] there's all people from all walks of life- [49:44] ... rich, poor, young, old. So it's quite remarkable to get, [49:50] you know, a group of people aligned on something. And that's an [49:54] achievement in itself. [49:56] - Yeah. It's incredible. And a lot of them are introverts, so you [50:00] coming to find them and getting them and getting them to answer an email might [50:03] be quite difficult. [50:05] - Most of us are introverts, right? You need to be more precise. You have [50:09] extremely introverts, extremely, extremely introverts and introverts, right? [50:12] It's just like a whole spectrum of different people. It doesn't matter. The [50:16] important is, is your code good? [50:19] Is your code great? Is your technology great? We care about excellent code. [50:24] We don't care who you are. Sorry, it's just like we have no idea to check. [50:28] We cannot check, right? Like, maybe you're a dog. I don't care, right? [50:32] I don't care where you come from. I need to look at your code. And this is [50:35] important because people don't understand that, and they come to the community and send them some [50:39] patches, and they get rejected, and they don't like that because, [50:44] I mean, you're just like, "Sorry, it's not up to our standards." "Oh, yeah, but I'm [50:48] engineer at this very large company in Italy, in Germany, [50:52] in the US." We don't care. We care about the quality of your code [50:56] because this is what defines our community, and which means that we have a [51:00] lot of people who contribute who are some very different backgrounds and, [51:04] and very introverts, sure. But that's okay, right? [51:07] - So one of the legends of the community is of course, [51:11] Linus Torvalds, who created Linux and is [51:15] a longtime maintainer of the Linux kernel. As the [51:19] legend goes, he can be pretty harsh on this meritocratic process of [51:23] reviewing the code and saying it's not good enough. Can you just speak [51:27] to the legend of Linus Torvalds? [51:29] - Linus is one of a kind, right? And [51:33] I would even go and say that what he did on Git is more interesting than what he [51:37] did on the Linux kernel. [51:39] He's very harsh, but what people don't see is usually when he's [51:43] harsh to, it's people who are maintainer of part of the [51:47] kernel, right? So they know him, right? So he's not very harsh like [51:51] that to everyone. The thing is, what he created in his room [51:55] is basically powering every server online, right? Even [51:59] at Microsoft cloud called Azure, I'm quite sure seventy, eighty [52:02] percent of the servers are running Linux. All your Android phones are running [52:06] Linux. What he did with the power of open source, sure, [52:12] is amazing. And yes, the quality of the Linux [52:15] kernel is very high, and yes, it's difficult, but [52:20] we cannot compromise on that. We cannot compromise on quality [52:24] because in the end, and you have to understand that, is the core [52:28] community of VLC is five people. The core community of FFmpeg is ten to fifteen, [52:34] and we are the ones who are going to maintain your code, right? [52:38] Because one thousand contributors in the timeline and just ten staying, it's [52:42] one percent chance that someone comes and stays. One percent. [52:46] So you will have change of job, change of wives, you have children, you [52:50] have accident in life. You're going to change jobs, whatever. You're not going to [52:54] come back. It's most likely. So we are the one going to maintain your code. [53:00] It needs to be maintainable. It needs to be excellent. [53:05] And yes, sometimes that means that you need to rework your work because it was good, but [53:09] it's not excellent, and we need excellence because we are very [53:12] few to maintain something that is critical for the whole. [53:15] - But we should also mention that there is some spiciness, some harshness to the [53:19] language that's sometimes used when you're keeping this high bar of excellence. [53:25] Is there something to say to that? [53:27] - It's, it's true, right? It's also the fact that, for example, what we're doing [53:31] is low level. It's extremely technical. You get into this [53:35] community. The tone gets very like [53:39] a type of-- It's a subculture, right? So people who arrive from the external are [53:43] basically not known to the subculture. Most of those people [53:47] around FFmpeg and VLC, we do VideoLAN DevDays, VDD every, [53:51] every year. They are so fun in real life, and they love it. [53:54] But it's true that you're online and sometimes, like, the tone, you [53:58] don't realize how it is. But that's okay. [54:01] - It's a culture. I mean, you get this in the gaming culture. There's pretty harsh, [54:04] intense, the way people communicate, and it's-- everyone [54:08] understands that the way you show love and respect just looks different in [54:11] different communities. Sometimes people... It depends. If it's a [54:15] book club, usually people are going to be much sweeter. If it's an open source [54:21] project that's very high stakes and used by millions of people- [54:24] - But it's very not often insults that you see, for example, [54:28] in the gaming, right? And so Linus' tone is a bit unusual [54:32] even for the open source community. It's more like it's more harsh on the results, [54:36] saying, "No, this is not good. This is crap." Those type of things that you will see. [54:40] - Try not to make it about the person, make it about the code. [54:42] - Yes. [54:43] - It's very, very matter of fact, and I think you've got to look at it in terms of, you know, [54:47] the famous FFmpeg is developed almost entirely by volunteers, and that's true, and you've got [54:51] to imagine someone's done a hard day's work at their day job. They come home. [54:56] You know, terseness might be a thing, you know, it... And that's not something to take [55:00] personally. [55:01] - You're tired, you're busy, but you still care about this open source stuff. [55:05] But you may not be able to explain and handhold someone on every subtle detail. [55:10] - And also you have to [55:12] realize that most people don't speak English as native language. [55:16] And this is especially for open source projects like [55:20] FFmpeg and VLC, which are mostly centered out of Europe. Sometimes like [55:26] people who are from the US or, or just like are very [55:30] not happy about the tone, but most of the time it's also like they don't know better, right? [55:34] It's difficult. The language is-- English is a difficult language. There is so many [55:37] subtleties and tone and so on that you don't have, right? [55:40] So often it's also difficult in those type of community about like [55:44] different cultures and languages. [55:46] - So as the legend goes, JB, you repeatedly turned down millions of dollars to keep [55:54] VLC open source free for everyone without ads. [56:00] So take me through the reasoning behind that decision of [56:03] leaving millions of dollars on the table. [56:06] - Yeah, that's like almost a meme, right, on Reddit or- [56:09] - There literally is a meme on Reddit. [56:11] - 9GAG and yeah, yeah. See, there's- [56:14] - You looking like a wizard in the, in the VLC hat on Reddit. This [56:20] is JB, the creator of VLC media player. He refused [56:24] tens of millions of dollars in order to keep VLC ads free. [56:28] Thanks, Jean-Baptiste Kempf. You can even summon him on Reddit. [56:33] - Yeah. And usually if you see, right, it's usually like people tag me, right? And, [56:37] and then there is me, and then like I say, "Good morning." [56:40] I got twenty-four K upvotes, which is great, right? My karma on Reddit is amazing, [56:44] at least on that account. So the question is, [56:49] needs to be answered first, what is the story about VLC, right? Because [56:54] yes, this is true, I refuse [56:57] dozens of millions of dollars, yes, several times. Yes, I could be a [57:00] multimillionaire and be somewhere on the beach. Um, [57:03] but I did not do it because I thought it was not [57:07] moral and it was not the right thing to do. And this is very important for [57:11] myself, is to be like, I work for the greater [57:14] good, I work for people, and I don't want-- It's not just by myself. [57:18] But the reason is also because I did not feel that I'm [57:22] completely legitimate to do that, and let me explain you why. VLC [57:26] story is a very weird story. In France, [57:30] we have university and we have a type of top colleges [57:34] and those top of excellency schools are engineering schools, [57:38] business schools, and basically lawyers and medical, [57:42] right? But they're outside of university, and in order to enter those, [57:46] you spend two years working like crazy math, physics to [57:50] enter those best engineering schools. [57:53] One of the schools is called the École Centrale Paris. It has changed name since, but it was [57:57] called the École Centrale Paris. And because it was Centrale, they [58:01] had to move it because it was too small after the World War II and, and they moved [58:05] it, they wanted to move it to the center of France in a place called Clermont-Ferrand. [58:09] And the alumni decided that this was not okay, right? It is [58:12] a, the school that Eiffel, right, the, the one who did the Eiffel [58:16] Tower, attended to, right? So they said, "No, no, we are amazing, great school. We cannot [58:20] do that." And so they bought a piece of land south of Paris very near [58:24] Paris. And it was a campus managed by a nonprofit of the alumni, okay? [58:32] Because of that, everything on the campus was managed by students. The [58:36] university did nothing, right? So radio, TV, supermarket, library defining who was [58:44] going into which rooms. Everything was managed by the students. [58:48] - That's amazing. That's an amazing experiment, that it all, all [58:52] didn't go to hell quickly. It somehow flourished. [58:55] - It worked great, and I learned so much in my life [58:59] doing those side activities, right? Because you're twenty-two and you need to run your campus, [59:03] else you don't have electricity, right? [59:05] So you care about that, right? But anyway, in the '80s they [59:09] did a full experiment of deploying a network mostly sponsored [59:13] by IBM and 3Com, which was a token ring network. [59:17] So token ring is something that probably almost no one [59:21] knows about anymore. It's a networking technology where [59:25] you don't have routers, right? [59:26] Everyone is linked. It's type, like really a ring, and when you want to [59:30] send a message, you talk to your neighbor who's going to put the message to the next one, [59:34] who's going to put the things to the next one, [59:37] in terms of ring. The issue with token ring is, of course, is that it's [59:41] very slow because every computer on the network needs to [59:45] open the message, see if it's okay. Is it for me? No, it's not, and [59:48] then send it back, like a token which is traveling around the ring. [59:53] In the '80s, you're doing some Telnet and sending mails as [59:57] university. That's okay, right? But starts the '90s, [60:01] and the '90s and start video games, and when you have high [60:05] latency in video games, basically you die, right? So in nineteen ninety-four, nineteen [60:09] ninety-five, around Doom and Duke Nukem coming around, they want a faster [60:12] network. So the students go and see the university and say, "You know [60:16] what? We want a faster network. We need to work," which, and also play [60:20] video games. And the university tells them that basically, [60:24] "Oh, I'm sorry, we cannot help you because you understand the [60:28] campus is not ours. You manage it, so [60:31] do something. And you should see some basically partners of this [60:35] university and basically go away." And they go, and they [60:39] actually go and see the CIO of [60:43] Bouygues, which is a large French company and who's doing some [60:46] TVs in France. And he says, "Well, you know what? The future [60:50] of video is satellite." Well, today we know it's not, but at least it was [60:54] a good idea. In nineteen ninety-five, the first satellite dish, and he says that [60:59] instead of having like one satellite dish and a big decoder for [61:03] each of the students, which are one thousand and five hundred, what about [61:07] you build, like you put an enormous dish and [61:11] only one decoder, and you send the video directly on the network. [61:15] And that required a very fast network. Today, it's obvious, but at [61:19] the time was, like, the first to do video streaming. So they built this project, [61:23] which was called Network 2000. [61:26] Of course, right, we are in the '90s, right? Everything is- ... futuristic is called 2000, [61:30] like— [61:30] - Yeah, 2000, yeah. [61:32] - And so they do the Network 2000 project. It's completely hacked. [61:36] It crashes after 45 seconds. That's okay. The demo is 40 seconds. [61:40] It leaks memory. That's okay. They put 64 megabytes of RAM instead of the [61:44] 8 or 16 you have, and the demo should have stopped there. And that was the Network [61:48] 2000 project by the students. [61:49] - What was the format of the video that they had to work with? [61:52] - MPEG-2 because satellite is MPEG-2 TS for transport, [61:55] MPEG-2 video, and MPEG-2 audio at that time. [61:59] And the project should have stopped there. Everyone was happy. They had, like, [62:03] amazing ATM network at 155 megabits per [62:07] second. They had probably one of the best network in Europe at that time, and [62:11] they stopped the project. Six months or a year later, two students arrive and say, [62:15] "Well, you know what? Maybe other people care about video streamed on a [62:19] local network," and they create the VideoLAN project, [62:23] VideoLAN. And one of them is called Christophe Massiot, that is a good [62:27] friend of both Kieran and me, and they start the project. It's not [62:30] even open source yet, and they spend around three years [62:34] to get the school to agree to make it open source. Because the [62:38] university wanted to get some-- because of the IP and copyright of [62:42] the students, wanted to basically monetize these MPEG-2 decoders. [62:47] - Just to be clear, so what was the main application, streaming on a local network? [62:50] - It was streaming on a local network. [62:52] - By the way, that's just, like, to state the obvious. This is before YouTube. This is before- [62:56] - Ten years before YouTube. You have a Pentium 60 or 75, right? [63:00] You, the main machine was 486DX at 33 megahertz, right? [63:04] - Bear in mind, television was the main form of video at the time. You could [63:08] get new channels. In the '90s, having even one new channel when you grew up with four [63:12] channels, having a fifth or a sixth was a big deal, and so [63:15] having this satellite service with, you know, [63:19] dozens, even hundreds of channels was so groundbreaking. [63:22] - Especially because this is university where you had a ton of different nationalities, right? [63:26] So there was a ton of people who wanted... So the-- in the end, they had, like, [63:30] several dishes on different types of satellites, right? Because, for example, a lot of people were [63:33] coming from the Maghreb or the Middle East and they, so they went [63:37] to different types of satellites. Anyway, the solution worked great, [63:43] and they started the VideoLAN project. The VideoLAN project has several [63:47] and some are completely crazy solutions, like one how to [63:51] create multicast on a unicast network, but let's not come [63:55] to that. It's too, too complex. But VideoLAN client part is what became VLC. [64:03] Actually, they basically strong-armed the university to force it to open source [64:07] because university did not understand that. And in 2001, it's still early. [64:11] But basically, yes, the university agreed early 2001 [64:15] to make it open source. I joined the project in 2003 because that's [64:18] when I joined the university. So the first thing is [64:22] I'm not the one who created VLC because actually no one did, right? [64:26] - Just kind of naturally emerged from the VideoLAN project. And we should mention that, like, [64:30] again, you said it just, but to make it clear, VideoLAN, [64:35] As what it became was at the time was a set of technologies [64:39] around video, and the VLC, what you called the client, that's the thing [64:43] that most normies, uh- [64:47] - That is correct, and [64:48] - ... think of, like, as the thing, which is, like, the thing that pops up when you click on a [64:52] video and you play it. [64:53] - So I arrive in 2003, and then I will create the [64:57] open source nonprofit organization called VideoLAN, [65:01] and I took everything out of the university to create a nonprofit [65:05] project and something sustainable. It's, yes, it's true that I spent [65:09] more time than anyone on VLC and VideoLAN. That is sure. [65:13] but it's a continuity of a previous project, VideoLAN, the [65:17] student project, which is a continuity of the Network 2000 project, which is a [65:21] continuity of that and that. [65:22] - I'm sure there's moments along the way there you were thinking of, like, [65:26] what is the future of this from an open source perspective? 'Cause as, [65:30] as the internet is blowing up, and there is companies... I [65:34] mean, for people who don't remember, like, there's companies making [65:37] huge amounts of money. [65:39] - And I can tell you that in 2005, the project should have died, [65:44] And I made it to continue the [65:48] project. At some point, we were only two active developers. [65:52] and I thought it was great technology and was useful, [65:56] and it will be useful and I made that my life and my, [66:00] my time. And I made that grow from a [66:04] few hundreds of thousands of users, millions of users to what we have [66:08] now, which is probably billions of version of VLC around the [66:12] world and used everywhere. So [66:15] that's a bit the story of VLC. There is ton of very funny stories around [66:19] that. Many people from around the world working [66:23] on it, like you said, in Syria or middle of nowhere in India. But [66:27] along the way, I got several offers which were either to [66:31] bundle toolbars, right? You remember those horrible toolbars- [66:35] ... which were basically spyware, or changing your web [66:39] browser or your search engine or even, like, [66:43] advertisement inside VLC. And I didn't like that, [66:47] right? I am-- and people don't understand that. It's not-- I'm not [66:51] against money, right? I'm very happy to make money. I created several [66:54] startups and one I hope that is going to work very well. [66:58] It's the fact that I believe that you need to win money [67:02] ethically. There is a right way of doing that, and doing sneaky [67:06] advertisement or stealing data is not the correct way, right? [67:10] For example, if Netflix arrived at some point and say, "Well, we want to put Netflix inside [67:14] VLC," probably the story would have been different, right? But they didn't. The only [67:18] people who came to us were shady ads company. [67:22] And if I do that, right, I would have a ton of money, right? And then three years [67:26] later, project is gone, right? Someone forks it and something else happens. [67:31] - So it's not even necessarily ads or any of that, it's the shadiness of the- [67:35] ... dishonesty of the-- So you had a good radar, you had a good [67:39] threshold of like, "No, this compromises the [67:43] spirit of what this is supposed to represent." [67:46] - But also it's for me, right? I'm like very selfishly, I [67:49] need to go to bed at night and be happy about what I've done, right? Maybe it's [67:53] my upbringing, maybe it's my parents' fault or whatever, right? But [67:58] I believe there is right and wrong, right? [68:01] And this was the right decision at the time. [68:05] It still is. I want to be proud of what I've been doing. [68:08] And like, if I had sold out, I would have [68:12] betrayed so many other people who work here. [68:14] - Yeah, well, I should say me and most of the internet [68:17] thank you for that decision. It's inspiring for others, [68:23] I think that are pushing the open source movement forward, [68:28] that it's okay to do these kinds of huge sacrifices [68:32] if you believe it's right. And I think in that case it was right and it was the reason that VLC [68:36] became as successful as it was, 'cause it's an embodiment, it's a symbol of [68:43] like, you know, freedom and what the open source community can create. [68:46] - Yeah, and be a service for so many people around the world, and this is important. [68:50] - We should emphasize in the 2000s it was really normal to download a program and it [68:54] secretly installs some spyware. It was buried in very faint text or in the [69:00] license text box that nobody reads that at the bottom- [69:03] ... "Oh, I will be installing this toolbar- [69:04] ... and changing all these things," and it was very common to have to, you [69:08] know, you install a program to do something at the time of any sort. [69:11] - To put yourself in the mind of a developer at that time, I think it's [69:15] very easy, to everybody listening to this, it's very easy at that [69:19] time to convince yourself to take a few thousand dollars- [69:24] ... a few thousand dollars to do it. To say no to much more money— [69:30] ... takes guts and takes vision. [69:34] - The last offer I had was obscene, [69:37] and they say, "Yeah, but imagine with all that money you could build [69:41] something new, open source," right? It was like the mind trick was, [69:46] it was difficult. [69:47] But for me it was just like, "No, this doesn't work like that or this is not the right thing, [69:51] so I don't do it." [69:54] and again, right, it's not that I don't like money or whatever. It's just like [69:58] it wasn't right. [70:01] - Well, once again, thank you from me and from the rest of the internet. [70:05] Let me talk a little bit more about the open source movement, about the [70:08] fact that, as you say over and over and over and over, FFmpeg, [70:13] is and many open source projects are built by [70:16] volunteers. So there's a bit of drama recently, [70:20] uh, Kieran, on the interwebs, on Twitter. [70:25] You have a spicy style on Twitter that I think [70:28] articulates and celebrates all the incredible developers and [70:32] development and the code, especially [70:36] assembly that's involved in building some of these codecs and building some of this [70:40] incredible technology. But that brings us to the, a bit of a [70:44] debacle that happened. Tell me the full saga of what happened with [70:48] the Google security engineers. [70:50] - Just to be clear, Google are one of the biggest supporters of open source out there. [70:54] They have been for a long time. It's just I think some things kind [70:57] of went a bit overboard this time. So FFmpeg [71:01] itself, and this is not like a secret, it's on the homepage, you know, the, [71:05] it processes untrusted data. There can be security issues when you parse [71:09] untrusted data. That's very normal. But recently what changed was Google started [71:13] using AI to create security reports on an open source project, FFmpeg. [71:19] Volunteers had to deal with that. They did, they provided very limited [71:22] funding, and they even went to the media first announcing how good their AI was [71:27] before the issues could be fixed. [71:29] - And this is in the public forum. [71:31] - Yeah, this is all public. [71:32] - So report, reporting an issue, using AI to find an issue in the code, [71:36] which is a security vulnerability, and then reporting that publicly before you're able to fix [71:40] it. [71:40] - Yeah. It's announcing how good their AI is, that they provided a standard [71:45] 90-day industry deadline without [71:48] without really understanding the nature of volunteer-driven [71:52] development. In addition, this vulnerability was on an obscure 1990s game codec. [72:00] the way-- And let's look at it from their standpoint to begin with. Let's you know— [72:04] - Yeah. Can you steer me in their case? [72:06] - Yeah, sure. They have substantial resources working on the security of open source [72:10] projects [72:11] that, you know, are ubiquitous, and they've used, you know, a lot of compute to do [72:15] that and very expensive and very capable security researchers, [72:20] to do that. And that's their viewpoint is they are contributing by [72:23] doing that. But I think that's where opinions differ. [72:30] it opened up a lot of interesting fissures I would say. [72:36] it does seem that there's a portion of the security community that [72:41] look at themselves a bit like building architects that never have to go to site. [72:44] You know, going to site is something that is a little bit beneath them, the actual [72:48] day-to-day construction. They're there to do their security things and it's someone else's [72:52] problem. The security industry also kind of has [72:57] a very aggressive tone towards things. The, the language they use is extremely [73:01] aggressive. They use very strong language like, "You will get popped." So and to, [73:05] Joe Public, get popped, [73:07] you know, means something quite bad. For them it means to get hacked. [73:11] The way I would look at it personally is a little bit like the padlock on your home. [73:15] The padlock on your home or, you know, the lock on your home [73:21] is there to protect against the capabilities of what it's there to protect. [73:28] It's not there to protect nuclear secrets. It's not there to protect Fort Knox. And [73:33] it could be looked at that they're using AI at a level of scale to go [73:37] and pick those locks and then say, "Hey, your [73:41] lock's not secure. You need to deal with this." Whereas actually they're the ones with [73:44] resources to be able to [73:48] fix this. But that seems to not be something either they'll contribute to in terms of [73:52] patches or in terms of financially. And the scale of AI is kind of the issue. [73:56] The bug reports are very wordy. They're very, [74:00] very-- It's almost a denial of service by AI-generated bug [74:04] reports on very niche codecs. [74:08] and the other issue the security community has is everything is marked high priority. [74:12] You're going to, you know, "This is the most important thing in the world, and you need to deal with this. [74:15] High, high, high, vulnerable, scary, scary, scary," [74:18] on a game codec used on one disk in 1993. [74:24] And that's where the dichotomy lies. Going around [74:28] telling everyone that their padlock's not safe, well, that's a hobby project of [74:32] somebody. The safety of that codec is consummate to [74:36] what that person thinks. It's their hobby. It's good that they're security analyzing it, but it [74:40] doesn't need a big scary warning, "This is a critical vulnerability." [74:45] We also may recently also see that there was another [74:48] quote-unquote vulnerability. It wasn't at Google in this case, but [74:53] a filter could overflow and have an integer overflow, and one of your [74:57] pixels could be the wrong color. And this was marked high, 7.5 severity in red. [75:05] And at some point, the security industry needs to realize you can't keep crying wolf [75:09] like this because this just leads to people, you know, the equivalent thereof of putting [75:14] password stickers on their PC. You know, you can't just keep crying wolf every day. [75:17] And I appreciate, you know, that's their modus operandi is to create as much [75:23] scare and fear. But from the Google standpoint, at the end of the day, [75:29] they need to contribute either financially or with patches. Google [75:32] uses FFmpeg at a scale probably you or I couldn't even contemplate, [75:38] millions of CPU cores. [75:40] And yes, they contribute in areas mostly regarding their own products, so VP9, AV1. [75:47] But in a wider sense, [75:51] there's a disproportionate level of contribution. Yes, they fund students. Yes, they fund Summer [75:55] of Code. And I think so Alex Strange is a former FFmpeg developer I think [76:01] posting in a personal capacity. [76:02] - So he posted about security engineers on [76:07] Hacker News. His post reads, "The problem with security [76:10] reports in general is security people are rampant [76:13] self-promoters, in parentheses, Linus [76:17] once called them something worse. Imagine you're a humble [76:21] volunteer open source developer. If a security researcher finds a bug in your code, [76:28] they're going to make up a cute name for it, start a website with a logo. [76:33] Google is going to give them a million-dollar bounty. They're going to [76:37] go to DEF CON and get a prize, and I assume go to some [76:41] kind of secret security people orgy where everyone is dressed [76:45] like they're in The Matrix. Nobody is going to do any of [76:49] this for you when you fix it." basically commenting on the sort of the incentives [76:58] for the different people involved and misaligned. [77:01] - The problem here is the disproportion of means on discovery [77:07] compared to patching it, right? And this is the biggest issue, right? [77:12] And after that debacle, Google did some changes. [77:14] - They are now starting to send patches, which is- [77:17] - And they also now have reward tools for fixing issues. So [77:21] it has changed a bit because of that debacle. So it's good, right? [77:25] But we've seen, and we talk about Google, but we have seen like some [77:29] other large companies saying, "Oh, you need to fix this bug because it's critical in our [77:33] product." [77:33] - Can you explain the XZ fiasco? The FFmpeg tweet reads, [77:39] "The XZ fiasco has shown how a dependence on [77:43] unpaid volunteers can cause major problems. Trillion-dollar [77:47] corporations expect free and urgent support from [77:51] volunteers. Microsoft, Microsoft Teams posted on a [77:55] bug tracker full of volunteers that their issue is high priority. [78:02] After politely requesting a support contract from Microsoft for long-term [78:05] maintenance, they offered a one-time payment of a few thousand dollars [78:09] instead. This is unacceptable. We didn't make it up. This is what [78:13] Microsoft Teams actually did." And then you give [78:17] the image and the details and all that kind of stuff, showing that these trillion-dollar [78:21] companies are not giving much money, not giving much support. [78:24] - They think an open source project is a traditional vendor that they have an [78:28] SLA. They think a public bug tracker is actually, [78:32] you know, a third-party vendor's Jira where you can do all of these things. It's [78:36] not. It is there to report bugs. [78:38] I think the thing that made this particularly heinous was [78:42] the name-dropping of Microsoft, the name-dropping that this is a visible product. [78:46] If this was just a general bug report, I think that would have made it a lot [78:50] better. [78:51] - Yeah, so they literally said, like, "This is a big deal because a lot of [78:55] people are using it in Microsoft." I wonder what [78:59] happens psychologically. So I think what happens in these companies, maybe you can [79:03] correct me, is they— You're right. They just think of FFmpeg as like a vendor that [79:10] Microsoft surely is paying a huge amount of money to. [79:14] They kind of assume that in their interaction, and [79:18] nobody anywhere on the stack is going like, "Wait a minute. [79:22] Shouldn't we be giving like millions of dollars to FFmpeg?" [79:25] - And this is a very big problem in large— Like we're talking about some [79:29] companies, but it's the same everywhere, right? A lot of those [79:33] companies. Like the, when we talk to that person, right, [79:37] he was just like a manager on one project in Microsoft Teams, right? He [79:41] had never really discussed with open source community. He [79:45] had no idea, right? It was like—but the problem [79:49] is that usually there is what we call OSPOs, right? Open source program [79:53] offices in those type of companies, and they are the ones who are supposed to [79:57] discuss with open source vendors. Um, or open source communities. But like [80:03] they often don't explain that correctly internally, right? And here it's just like [80:08] we are not your supplier. If you want me to be your supplier, I'm very [80:12] happy, right? I will send you a contract and SLAs. Like I created [80:16] five companies who are doing that around open source projects, so that's okay. [80:19] - We should say that some of the spicy tweets that Kieran, you're [80:23] behind, and some of the debacle produced results. [80:28] - Yes. [80:28] - Positive results. [80:29] - Donations have increased substantially. They're still not enough to [80:33] cover even a single full-time developer, but [80:36] on both a, you know, awareness level and a technical level, [80:41] there's substantially more technical awareness and sort of awareness of the importance [80:45] of FFmpeg as a result, as a result of X and what's [80:48] happened. I can say, you know, it solved its purpose. People [80:52] realize the level of importance FFmpeg has. [80:55] - And on VideoLAN it's the same, right? Like for example, a, a very simple example. [81:01] For more than a year, we couldn't update VLC on [81:05] Android because of a bug on the Play Store, on Android Play [81:09] Store, right? The only way we got someone to answer [81:13] was to put a very spicy, as you say [81:17] tweet saying that we are going to stop distributing VLC [81:21] for Android, right? And we have around 100 million people using that. And now then [81:28] someone from Android actually came and discussed to us, right? We had the [81:32] same issue with Microsoft or, or like saying that we were [81:36] going to stop distributing VLC on the Windows Store. And [81:39] unfortunately, we are so small that the [81:44] only very strong power we have to solve those issues [81:48] is blaming on social network because it snowballs and [81:52] now they listen to us. But so as large companies often have difficulty [81:57] talking to us. Like for example, VLC, right, is probably one of the top [82:01] 10 software used on Windows. I am not part of Microsoft ISV [82:08] programs, right? I don't have a point of contact at Microsoft, right? [82:12] While I'm sure any other software, Adobe, Spotify, has a [82:16] point of contact. I don't have that, right? So [82:20] raising awareness works. It's sometimes very spicy, lot of drama. Well, [82:26] X and Twitter are okay for that, but it's efficient. [82:30] - Uh, so everybody listening to this should go follow FFmpeg on Twitter, on X, follow [82:38] VideoLAN on Twitter, on X. Go donate. Donate ... to FFmpeg. [82:45] - And thank you, Lex. Over the years, several years you've been a supporter of, you [82:49] know, FFmpeg and VideoLAN on X. You know, giving us shout-outs, appreciating, [82:54] you know, what we do. [82:55] - FFmpeg for life. [82:57] - And for example, like Tim Sweeney, Carmack, and a few [83:00] others, like very high-level people have raised also the awareness [83:04] on our X accounts, and that helped a lot also. [83:09] - Karpathy as well. [83:09] - Karpathy, yes. [83:10] - Karpathy as well, yeah. [83:12] - Yeah. I mean, also, you know, outside of the fact that so many people use it, it's [83:15] so impactful on the world, it's also a great representation of a great open [83:19] source project. Like the value of assembly and C and making sure that like you take [83:27] programming seriously for real world systems. [83:30] - It's not just that. We'll talk about assembly later I'm sure, 'cause that's its whole topic in itself, but it's [83:34] also celebrating people like Andreas Rheinhardt who do maintenance. It is, I [83:40] believe unpaid, as I believe as a volunteer. He's doing massive [83:43] refactorings. Uh, Andreas Rheinhardt and Anton Khirnov [83:47] rewriting ffmpeg.c with threading. Celebrating those guys, [83:51] celebrating the untold labor that's gone into this [83:55] that actually doesn't change anything from the user standpoint. The files are exactly the [83:59] same, but wow, the, the, the airplane has been rebuilt whilst it's in the air. [84:03] - Christian Garcia said, "As a teenager running this account," referring to the FFmpeg [84:07] ... account, and you responded, "Teenagers have written more assembly [84:11] in FFmpeg than Google engineers." But also [84:15] just pointing out that there's a lot of incredible contributors who are teenagers. [84:19] - Like JB said, we don't care who you are, where you're from, what you do. [84:24] Teenagers have written thousands of lines of assembly, [84:28] Over the years. Give a shout-out back in the days to Daniel Kang. [84:33] So also highlighting the work of people like Ruikai Peng. This is a 16-year-old, some of his [84:37] first contributions to FFmpeg, [84:40] actually doing and putting some of these quote unquote security researchers to shame by [84:44] by actually finding issues and fixing them and being [84:47] 16. There's no barriers. There's no barriers to you have to [84:51] study on, at college under this person and understand these. It's you can [84:55] learn C, and let's be honest, it's from, it's from the K&R book. Learn C. [85:00] You can learn assembly. We'll talk about that maybe a bit later. [85:03] You can contribute to world-class technologies. [85:06] - In VLC one of the oldest contributors called [85:10] Felix, he's the one doing everything on Mac and iOS. He's starting [85:14] working on VLC. He was 16. We had a guy called [85:18] Edward Wong, who used to be a Google Summer of Code student who [85:22] stayed for three years around VideoLAN. He was 14, [85:26] right? And, and part of Google Summer of Code and Google [85:29] Code-in, which were programs where basically we have students or high school, [85:34] We wrote a ton of assembly for x264 and for VLC [85:38] and for FFmpeg, right? So everyone can contribute. [85:41] - And he also did a good job because he didn't play the alarmist CVE [85:45] heist, create a CVE, which is like, a [85:49] public exposure of security and do these big scary [85:53] red 7.5 high priority. He just fixed an issue in Git [85:57] after three days and just fixed it. He didn't need to go and play a big [86:00] security drama about it. And I think [86:03] I posted, you know, the kids are all right. Whereas- there's, you know, [86:07] there is a por- I'm not saying all security people do this, but there is a portion of the [86:10] security community, as Alex said, that likes to hype themselves up by [86:14] creating drama. They would have happily raised, "This is a high [86:18] priority CVE 8.0" or whatever on a [86:22] issue that actually was in Git. It wasn't even in a release, it was in development, and three [86:26] days later was fixed. [86:27] - Well, I just want to put a little bit of love out there, even to the bigger [86:34] much love and respect to Google engineers. Like you said, they're [86:39] Some of the, the best software engineers in the world, and they do contribute a lot- [86:43] ... even on the security front. And also, you know, I'm a big fan of [86:47] Theo. Much love to Theo. He was part of this, [86:50] Debacle and drama a little bit. I think when you just zoom [86:53] out on the grand arc of human history, [86:57] the drama contributed positively to everybody involved. Donations went up. [87:02] It brought more attention to the topic, allowed [87:06] everybody to bicker in a way that ultimately got them to figure out [87:10] what FFmpeg is all about. [87:11] - So the way we looked at this is like it's a rap battle at the end of the day, you know? [87:16] No, but it is. We say stuff, we say stuff- [87:19] ... but we can, we can leave it on. X is a perfect place for, you know, international rap [87:23] battle. You say stuff. I say stuff about your mama, but it doesn't mean, you know, I have [87:27] an actual personal issue with her. [87:29] Uh, and that's what it looks like. The Theo situation, you know, JB can maybe expand, went a [87:33] little bit too far and there was a little... But, you know, it's just a bit of fun. [87:37] It's just a bit of rap battle. It's a bit, it's WWE. You know, [87:40] everyone's having a bit of fun on X. [87:42] It doesn't need to be taken seriously. You know, the teenagers thing, you know, [87:46] that... So that guy was a Google employee saying, "Hey, you know, there are other ways to run [87:50] an open source business." You know, go and there's like, oh, man, just have a bit of fun, you know? [87:54] That's what the point of this account is. And, and furthermore, if you can teach people [87:58] about the ways of open source projects, assembly, et cetera, by doing [88:02] that, I think there's a lot to be offered here. It's not dunking on people for dunking's sake. [88:06] It's showing actually the story that I think X learnt is these are not big [88:10] corporate open source projects. This is not Kubernetes where there's, [88:14] you know, hundreds, maybe thousands of people- [88:15] ... paid to develop this stuff. These are just people in their basements in their spare time, [88:20] and if you can address that topic in a fun and entertaining way- [88:23] ... I think that's the good thing and that's, that's the value of X and [88:27] then the reach we have. [88:28] - And to be honest, right, like even at Google, Google is [88:34] one entity, but so many different people, right? And there is a ton [88:38] of Google engineers we work with [88:42] all the time, and even like Google from YouTube to Chrome to [88:46] Chrome Media to the rest of Google, those are very different types of entities. But [88:50] what we do is efficient. And, for example for, [88:54] for Theo, right? It went a bit too far. I had him... Like I calmed everyone [88:58] down. I had him on the phone. We said, "Okay, like this goes too far," and so on. [89:02] But in the end yeah, it's a rap battle, but it's positive for the [89:06] project. It, like the awareness we have on open source and, [89:10] and I mean true open source from communities right now is [89:14] increased dramatically in the last two years, and this is useful. [89:19] - Uh, what do you think motivates all the incredible contributors that we've been [89:23] talking about? Like, what's the engine? It's so interesting to see. [89:26] - So- [89:26] - Like you said, they're sitting in the basement. What's the driver? What's the engine there? [89:29] - There are many drivers, but weirdly the main one [89:33] is that what we do in multimedia plays videos, and video is [89:37] cool, right? And, and for example, we have so many [89:41] people in the community who arrive because they loved watching [89:45] anime, right? And this is like the advice when people ask me, [89:50] "What should I work on in open source? How do I start?" And my answer is always the same: [89:54] work on something you love. [89:56] I am working on VLC because I love movies, right? And I love [90:00] watching the same movies over and over, even if my wife hates me when I do [90:04] that, right? But because it's interesting, right? Because it's a topic that [90:08] you like, right? The first, that's the first thing where people come to [90:12] usually to VLC and FFmpeg. The second thing is that [90:15] technically we, because we search for excellence, this is [90:19] the best school ever, right? This is the best school [90:23] ever of programming. If you're good in C, in [90:27] FFmpeg, if you know how to write assembly, I assure you you're going to [90:31] be one of the best programmers ever, even if you're working on writing [90:34] TypeScript, because this is the most amazing thing to [90:38] do. And you will, like, have to get reviews by some of [90:42] the most seasoned programmers ever who are going to look at [90:46] every part of your code and tell you why it's not great. It's like we are the [90:50] best teachers that you've ever had in programming, right? [90:52] - Andrew Kelley started Zig. He was an FFmpeg developer and started Zig after [90:56] his FFmpeg school. I mean, [90:59] it's the place to learn so many aspects of programming in the [91:03] real world, in a thing used by billions of people. You have nowhere to [91:07] hide. You have to be open and honest about your flaws and how you can learn [91:11] and be better. [91:12] - And what is also interesting in multimedia is [91:16] that you have 16 milliseconds to display a frame. It's not like [91:19] a game engine where you can basically slow down and wait a frame. [91:23] Like, so it's, you need to be good, right? There is no choice, else you don't have your [91:27] video. And because of how codecs, if you miss a frame, you're going [91:31] to destroy the look of the video, right? So you need to be good. [91:35] You need to be perfect to have the right thing. But also [91:39] is that it's not just pure programming in the mathematical sense, right? [91:43] A lot of people don't understand, but [91:47] in order to program correctly on the open source multimedia community, [91:51] you need to understand how computers work. And when you write assembly, [91:55] you need to understand about CPU pipelining, right? You need to [91:59] understand how SIMD works, how the ALU works, right? You need to [92:03] understand what, how IO works, right? And this is what I [92:07] think that is missing to a lot of engineers and software engineers today, is [92:11] understanding what we call computer architecture. And, [92:15] like, seriously, like some of the debates is like, should we use this assembly call or this [92:19] one? And people say, "Well, no, it's going to be like three cycles on this [92:23] type of CPU and this one," and has massive impact on the output, right? [92:27] - We should expand. FFmpeg is probably one of the biggest CPU users in the world. There's [92:31] it's probably running— ... as we speak [92:35] easily 100 mil- order of magnitude 100 million, maybe even a billion CPUs [92:39] as we speak. So every instruction matters. There's not... [92:45] The impact, at least in terms of CPU, is massive for everything that we do. [92:51] - So first you come because it's an interesting subject, then you stay because it's [92:55] excellent, and in the end you're very proud of it because [92:59] it's in the hands of everyone. Like so many [93:02] people like, "Oh, I'm working for whatever consulting company and I'm doing some [93:10] portal to download invoices for your PG&E." Wow, [93:14] great. Like, so many jobs are like that. You're not going to, [93:18] to tell that to your grandma. But if you go to see your grandma and say, "I do [93:21] this so that you can play video on your laptop," they understand. And this is very [93:25] important, right? Because you're working on VLC, FFmpeg, H.264. It's [93:29] in the hands of hundreds of millions of people and you have an impact. [93:33] And so you can be proud of yourself. And so I think that [93:37] in addition to doing a great resume, all those things are why people contribute. [93:42] - Yeah, those are side effects. My favorite quote on this topic is John Collison. [93:45] He said, "The world is a museum of passion projects." You know, everything out [93:49] there is a passion project. And open source multimedia and open source [93:53] in general, you can just do that so much faster. There's such a [93:57] faster network effect, you know? [93:59] I can open a cafe and that can be my passion project, but I have to get building codes, I have to build a [94:03] building, I have to find a location, I have to do all the, you know, all [94:07] sorts of things. Well, in the software world, that passion project can be, [94:11] can move quickly, it can be amplified by the network effect, [94:16] and that amplification can be more than the sum of the [94:19] parts. You know, you can be, you can find people [94:22] interested in extremely obscure things [94:26] and have a network effect and make something that is truly amazing. [94:31] - And on that topic of passion projects Tim Sweeney actually said in a reply [94:37] to a tweet that was complimenting JB. He said, [94:41] quote, "Many things in the world only happen because an awesome [94:45] person decides to do it. This is the case with VLC." And [94:49] that speaks to something interesting to me, that it does seem that a small number of [94:53] people, sometimes one person, can create [94:58] something incredible in the software world. Like you said this over and over and [95:01] over. I think JavaScript is an incredible thing created by, [95:05] Initially a single person. Some of the programming languages like [95:09] Python and C and Java, like just one person has this [95:13] vision, has this design, and brings it sometimes over a weekend is the [95:17] initial spark. [95:18] - Yes, Linus built Git in two weeks. Wow. [95:23] - It changed the world, Git. I mean, it really changed the world. [95:25] - Linus' passion project. "Hey, I'm uploading this tarball to an FTP, like deal with it." [95:29] - But for me, it's not just in software, right? And I believe in, [95:34] in individuals that are going to change the world, right? And it's with [95:38] a good, as you said, vision, right? I want to do that. It [95:42] is useful, it will be useful. And whether it's going to like build [95:45] train or cars or rockets or something like, I believe [95:49] people who believe in themself and have a vision can have a huge [95:53] impact for humanity. [95:56] - Let's actually zoom out before we zoom back in. We'll just keep [95:59] going up and down the stack. So you know, we've been [96:03] talking back and forth VLC and FFmpeg. Kieran, you said that [96:07] FFmpeg and VideoLAN, VLC coexist, [96:11] and there's no central point of importance. It's a kind of [96:15] what you call the binary star system. [96:18] Uh, they succeed because of each other. Can you explain the difference, how [96:22] they interact? What is the- ... are they competitors? [96:26] - I don't think they're competitors. I think the simple answer is, the [96:30] short answer before I go into detail is VLC is to FFmpeg as Android is to Linux. [96:36] So they depend on each other, but they coexist because of each other. So they [96:40] are a binary star system is the analogy I used. [96:43] - By the way, I feel horrible that I just recently learned that Alpha [96:46] Centauri, the closest star system to us, is a triple star system. [96:50] - And when you start doing the physics, it's a nightmare, right? But, but, but like- [96:55] - Hence the three body problem. But anyway. So a lot of [96:58] FFmpeg pipelines involve the x264 project, which is [97:02] a VideoLAN project. I would put a finger in the air and say [97:06] 80-plus percent of those pipelines are dependent on a VideoLAN project. [97:11] VLC, obviously, as we've discussed, a VideoLAN project, uses [97:14] FFmpeg, gives it reach, exposure to weird files, [97:19] Historically used some donation money to fund FFmpeg [97:22] development, and we'll talk a bit maybe about some of the reverse engineering later. [97:27] So it's a binary star system. They work and feed off each other. Many of the developers are [97:31] shared. There's no central location. It's a virtuous cycle working together. [97:36] - And we should mention that x264 is the encoder for H.264 video standard. So [97:43] H.264 is the standard. X264- [97:46] - Is the open source implementation of the standard [97:49] - ... that's used by basically everybody- ... for everything. [97:52] It's, that is the main driver of this. When you think of an MP4 [97:56] file that has H.264 codec in it- [97:59] - If it came from a software environment, like a data center or somewhere, [98:03] the chances are it was created with x264. [98:06] - And that's under the flag of VideoLAN. [98:09] - That's a VideoLAN project. So in the VideoLAN graphic, it sits in the VideoLAN [98:13] world. [98:14] - And VideoLAN has a, says a bunch of stuff in it. Go to the [98:17] VideoLAN website, there's a bunch of icons. [98:21] - Like if you look, there is so many libraries, right? [98:24] - libdvdcss- ... libdvdnav, libdvdpsi, libvlc of course, vlc-unity, libblu- [98:35] Blu-ray. Uh, yeah, there's many more. [98:39] - And there is so many more, right? [98:40] Lately, lately the dav1d project that we might talk about is the [98:44] last project from VideoLAN. It's everywhere, right? And we do, [98:48] we have a libspatialaudio lately that we announced. We have a- [98:51] - checkasm. [98:52] - checkasm- [98:53] - We'll talk about that later [98:53] - ... which is like an insane project- ... but amazing. So and x264 is one of [98:59] those VideoLAN projects. And my opinion, for example, is that [99:04] x264 was, is the most amazing encoder ever designed, [99:09] and this helped the adoption of FFmpeg. A lot of people and large [99:12] companies went through FFmpeg because they wanted to use [99:16] x264, and x264 increased the popularity of FFmpeg. But [99:20] also VLC had its popularity because [99:24] it's played so many files that were done by FFmpeg, right? So it's, [99:28] it's many projects that are intertwined and work together. [99:32] - Yeah. Unfortunately, there's a, there's a thing on X where VLC is mentioned and [99:36] there's people, "A quick reminder that it's FFmpeg inside doing the actual work." And that, and [99:40] that's like I said, it's not, that's not the case. We work together. [99:46] - And to give you an idea, right? When I compiled VLC for Windows, I compiled [99:50] around 16 million lines of code, right? One [99:54] million of those are inside the VLC repository, and FFmpeg in total is probably two, [100:00] around two, right? But so it means that so many dependencies are outside. And [100:04] if you also look at FFmpeg per se, FFmpeg also is [100:08] integrating third-party libraries like x264, but [100:11] LibOpus and so many others, right? So we all depend on each other. [100:15] - Uh, yeah, that's why I was hoping to do this episode as we are doing that [100:19] just kind of joins FFmpeg and VLC- [100:23] ... because it's really, it's, it's really two of the same, like you said, [100:27] binary star system and we're all just orbiting it. Can we give a shout-out [100:31] to some of the people along the way? We didn't really quite talk [100:34] about the history of FFmpeg, so- [100:38] maybe can you tell me about Fabrice? Can you tell me about [100:42] Michael Niedermayer? Can you tell me about some of the key figures here? [100:46] - Let's just talk about the eras of FFmpeg, because there's [100:49] key eras and key people that made this [100:52] possible. Uh, Fabrice Bellard, as you mentioned, creating the [100:56] concept, and then probably in the 2000 era, I would call the era, [101:00] eras tour of FFmpeg is the 2000 era was Michael Niedermayer. So [101:05] key things he got done was exhaustive support for DivX and [101:09] Xvid at the time, and all sorts of weird variants of what's known [101:13] as MPEG-4 Part 2. So this predates the [101:17] MPEG-4 Part 10 that we're used to. So this was 2000 era video [101:21] codecs where there were, oh, flavor after flavor of weird decoders. [101:27] At the time in the 2000s, you needed a new player to play every different type of file [101:31] format. So there was Windows Media Player to play Windows Media formats. There [101:34] was RealPlayer to play RealMedia formats. And those were the other, the other [101:38] key thing in FFmpeg at the time were native decoders for those. I actually do remember [101:42] being a teenager, I must have been, [101:46] figuring out there was this one player that could play, [101:49] could decode these files without having separate bloated players. Because [101:53] at the time when you downloaded RealPlayer, there was a ton of other stuff in there, a ton of ads, a [101:57] ton of other things, and just having a simple library [102:00] that was fast led to that. And then I think 2008 was a, 2008 onwards [102:08] was a big change because that's when H.264 got its maturity and I think [102:14] something hopefully we'll talk about a bit more. This was the beginning of high definition video. [102:19] So H.264 was the key decoder of that. [102:22] So I'd call that the late 2000s and 2010s, and that's when the big reverse [102:26] engineers came along and really did astonishing work. [102:30] The beginning was a single player that could play Xvid, DivX, [102:35] Windows Media, and RealPlayer was already a massive achievement in itself without codec [102:39] packs, without weird stuff you had to download that had weird ads and weird spyware. [102:43] - VLC 1.0 was out on those times, 2000, 2009, 2010. [102:49] And this is like where it exploded. [102:53] - Yeah, without codec packs, it just works- ... across all these different- [102:57] - It, de facto, it's just like all the codec packs are FFmpeg inside VLC, [103:01] plus we have other modules for all the type of codecs. [103:03] - But back at the time that wasn't, is there were weird, in the 2000s, there were weird [103:07] codec packs with DLLs coming from this place, DLLs coming that- [103:11] - With a lot of spyware [103:12] - ... with spyware, with you know what. It wasn't reliable, you didn't know, and having a single [103:16] player that was open source or single playback module/player that [103:20] could do this that was open source. But I think the thing to emphasize is [103:24] this task in the 2000s that Michael did was Sisyphean. It was really, [103:28] the number of edge cases are poor beyond comprehension in terms of [103:32] you could have a Chinese CCTV system that did one weird variant [103:36] of MPEG-4 Part 2, what's known as MPEG-4 ASP, [103:40] and that was a weird variant, and you had to fix that without breaking everybody else- [103:44] ... times a million. [103:45] - So that's, so you said that's where a lot of the reverse engineering was happening. [103:49] - It started in the 2000s with the Windows Media stuff because that was- [103:52] ... proprietary. It started with the RealMedia, so with Benjamin Larsson. [103:56] - Kostya Shishkov. [103:57] - Kostya Shishkov, that era. Those were the key, that was the key [103:59] groundwork. And then in the 2010s was kind of [104:04] the Paul Mahol, Kostya era building, doing some of the [104:08] most difficult codecs. JB maybe can talk about GoToMeeting 4 and [104:11] GoToMeeting 5, and- [104:13] - What? What's the GoToMeeting? [104:15] - So, like, let's talk about this [104:18] amazing Ukrainian guy called Kostya, who was at that time living in Germany, and who [104:25] was in love with Sweden, right? He— And the guy was the most... He's like, [104:33] like a lot of the people in the community are very clever. He's [104:38] one of those who are, like, borderline geniuses, right? [104:41] He was able to reverse engineer extremely complex codecs, [104:45] And he does that, and we do a bit of engineers with [104:48] Kieran, but clearly not at this level. [104:50] - No, no, yeah. [104:51] - Um, he reverse engineered binary blobs, which are 20 megabytes? [104:56] - Yeah, so just for reference, one megabyte binary blob to reverse [105:00] engineer is probably order of magnitude a month of work, and this guy is doing [105:04] 20, 30 megabyte blobs. Maybe we'll talk about that in a minute, about the [105:08] subtleties of how you do that. But this guy is doing it for very difficult and very obscure [105:12] codecs. [105:13] - And did that for fun, right? And so GoToMeeting was a big problem with VLC [105:19] because that was like the number one [105:23] feature request for a long time, so I put a bounty. And the guy at some point said, [105:27] "Okay, JB, I'm going to do it." And in a matter of two [105:31] months, and then he explained how he did it. He was just like, "Oh, I looked at the code, [105:35] like this looked like a DCTs that I used to see on WMV [105:39] and so on." He did that, and the funniest part is that [105:43] the code he's written is a ton of jokes. And [105:47] there is, there is a ton of JB, right, my name, and, and [105:51] Kempf and Kempf and Kostya jokes inside the code. The code is beautiful, right? [105:57] - So one of the things I wanna comment is I've gotten a chance to [106:01] speak to some of the developers, some of the assembly language level [106:05] People, and they all always make everything sound like it's kinda easy. [106:10] There's a kind of humility because, maybe [106:15] just the level of what's required to do this stuff is so [106:18] high that everything else seems easy, I guess is the lesson to take away from that. [106:23] - So in the community, like some of the most impressive people are the ones doing [106:27] reverse engineering- ... and the other ones doing the assembly folds, right? [106:31] Um, and both of those type of people are [106:35] amazing. x264, for example, became amazing because, of a guy called Loren Merritt— [106:42] ... who is, was from University of Washington, I think. [106:44] - At the time, yeah. [106:45] - And who was, like, who made everything great and fast doing a ton of assembly. [106:51] Um yeah. So this is like the [106:55] the golden era, I guess, where so many things got done. [106:57] - So, yeah, if you look at Kostya, for example, he looked at the world as a binary specification. [107:01] He didn't need documentation or anything. It's, "I have a binary and I can [107:05] figure all of this out." And he regularly used the phrase binary [107:08] specification. "Ah, you know, it's not a problem." And he would go away, and he would come back, and [107:13] he would do interesting stuff. [107:15] - Can you actually speak to the details or add color and [107:18] texture to what it takes to reverse engineer a blob? [107:22] - Yeah. So let's look at GoToMeeting, for example, is a good one because, [107:26] um, I record a meeting on GoToMeeting, for example. [107:30] How do I play it back without needing this GoToMeeting [107:33] player? There may not even be a player. I may, I may need to send a recording of a meeting to [107:37] someone that doesn't have a player or whatever. [107:40] So first of all, there's a ton of other stuff there. There's an actual video [107:44] conferencing client. You need to go and find, it may be easy, it may not be easy to find the actual [107:48] module doing the decompression. You need a way to [107:52] actually dump the YUV data from the module. So often it [107:56] involves opening in a disassembler, trying to guess where the hooks are [108:01] to incorporate that module and run that module natively [108:05] to decode a sample file. So figure out where this module is doing [108:09] the decoding process and find a way to hook in and [108:13] output the raw YUV data, 'cause you will need that- [108:16] ... as a point of comparison for when you actually do the reverse engineering, 'cause you'll need to be bit [108:20] exact or in some cases close to bit exact. And then you open up your disassembler, [108:27] use a lot of intuition to go and figure out, you know, where the DCT is, [108:31] where's entropy coding. There, there is a kind of, [108:35] not a rule book, but there's always a pattern of some sort. For example, [108:39] GoToMeeting, you know it will be a s- a lot of screen codec tools. [108:43] There's also different variants, so often I think there's, what, GoToMeeting 4, 5- [108:46] - Well, 2 or 3, 4, I think. [108:48] - 2, 3, 4. [108:49] - So as you mentioned here, going to Perplexity, GoToMeeting uses its own [108:52] proprietary codec for older s- recorded sessions historically stored in [108:57] WMV files that require a special decoder to play properly on [109:01] Windows. Without this decoder installed, Windows Media [109:05] Player and some editors cannot decode the video track, so you [109:09] may only hear audio or see a black screen. Boy, do I remember [109:12] that. But this is reverse engineering that. [109:16] - This is key, right? Because the GoToMeeting is something that not many people know [109:20] anymore, right? Well, you know about Zoom and, and Teams and so on. But like, now [109:24] let's fast-forward 10 years, 15 years, and like this is a [109:28] GoToMeeting.exe for Windows 32 bits, right? Which is like, oh [109:32] yeah, but I'm on Android, I'm on an iPad, I'm somewhere else, right? How are you [109:36] going to do that? I'm going to be on RISC-V, on Arm. Those are blocked, [109:40] but there are tons of files we need support for the future. [109:44] And this is why those type of work are— exceptionally useful for humanity. [109:51] - I just have to say, though, that reverse engineering process is mind-blowing. [109:56] It's crazy. It's like, [109:58] it's a kinda like, you know, I've been reading a lot and interview [110:02] archeologists. I mean, you just have so little signal. [110:05] Yes, yes, you know over time you get so much experience, you [110:09] understand the structure of the original code, so you can kinda start inferring [110:13] basics. But you're like-And we like archaeologists with a [110:18] little brush trying to reconstruct the entire human civilization [110:22] - Kieran is too humble, but Kieran has done some reverse engineering also. [110:24] - Of CineForm, yeah, at the time, um- [110:26] - CineForm, nice [110:27] - ... yeah, at the time before actually led to the open sourcing of that work. Um, [110:32] so in parallel to doing the binary side, you obviously have samples. In many [110:36] cases, you don't have many samples so you have to figure out what all the [110:40] different flavors are, and you may have a s- So CineForm, for example, is actually a [110:43] collection of different approaches and toolkits within that codec 'cause often [110:47] it grows naturally. And the hard part is finding a sample that gets you kind of [110:53] somewhere to start without having to implement 10 different other things. So [110:58] start there. [110:59] I think thankfully at the time I found a sample by pure chance that had a lot of flat blocks. [111:03] It was animation, so that really helped a lot because [111:06] it wasn't using particularly complex coding tools, et cetera, and you could kind of get [111:10] somewhere and then, and then build up and build up until you figure, "Hey, here's a few [111:14] bits here. I missed this. I missed this, this if branch that it does," and go, "Oh." So when we [111:18] say samples, you mean sample videos- [111:20] ... and then, and then you're tracking, trying to infer, like, what is this [111:24] codec doing- ... by observing the sample and then looking at what, at [111:29] the lo- at, at the machine lo- [111:30] - The machine code saying- [111:31] - At the machine code [111:31] - ... "Ah, I have byte, this byte is six. Take this branch." And in a different [111:35] sample, oh, it's- [111:36] - That's nuts, man. [111:37] - And- [111:38] - That is nuts [111:39] - ... so you see, this is nuts. Then you go to things like GoToMeeting. [111:42] - Yeah, yeah. [111:42] - It's like- [111:43] - Mine was easy, right? [111:44] - ... imagine- [111:45] - Yeah, right [111:45] - ... two order of magnitude of more complexity. A guy alone [111:51] somewhere in Germany doing that. [111:53] And for a long time, you work, you're in a black box because a decoder, [111:57] for a long time, because there is so many steps from the entropy decoding, the intra [112:02] prediction, the motion prediction, the IDCT, and so on. For a long [112:06] time, you don't see anything, right? So you're debugging purely in memory. [112:10] - Debugging guesswork. [112:10] - And you may have the buffer that the coefficients are stored in completely wrong, [112:14] and so you may be going down a complete rabbit hole thinking it's this and then, [112:18] oh damn, that's not, that's, that's something else, and- [112:21] - And you're doing that on binaries that are tens of megabytes, [112:25] millions of instructions, right? [112:27] - So you're stepping through the debugger, [112:29] like one by one, you know, instruction by instruction going, "Hey, this instruction changes this. [112:33] This does this." [112:34] - Pausing the program on the CPU level. Like it's- [112:36] - Pausing it, yeah, on the CPU level, watching what's going on, trying to figure out- [112:39] - Sometimes you need to, like, be in a VM, so yeah, that you can pause the VM. [112:43] - Yeah, pause the VM, dump the memory, 'cause there could, some of the codecs could have encryption. [112:46] There could be like a DRM on there. So you need to dump the [112:50] memory from a virtual machine. [112:52] - Like when I joined École Centrale Paris in 2003, Jon Lech [112:56] Johansen basically broke the DVD specification and created [112:59] DeCSS, showed us how he was breaking a [113:03] DRM, which was MP4 FairPlay from Apple. What he did [113:07] on his laptop, and I was young, I was 21, was just like mind-blowing [113:11] because he was basically debugging Windows inside a type of VM [113:15] with ex- Like, wow. It's incredible. It's [113:18] mind-blowing and inspiring. Does it get, like from your [113:22] experience and from what you've seen in the community, does it get discouraging? [113:25] Does it get- [113:26] - People help you. People send you samples. People are keen. Sometimes you don't [113:30] have access to an encoder, so [113:32] this is even more difficult because you just, you just ask and you have to [113:36] ask for samples. I remember VideoLAN used to tweet for samples at one [113:40] stage. "Hey, I need this obscure sample," and- [113:42] - For a long time I was, "Oh, I need this codec, and I need this codec." [113:45] - And if you were really lucky, you would find like... If you were unlucky, you'd get [113:49] like one or- you'd get nothing or you'd get one or two, and then they would... Sometimes you'd find a [113:52] goldmine. It's like, "Yeah, my company has 100,000 of these files 'cause we're dependent on it for [113:56] some reas-" Um, and so those are the, those are kind of the best if, if... [114:00] Because then they can test bit exactness across the huge range of coding tools. [114:06] - Can you explain bit exactness? [114:07] - Bit exactness, so most but not all video [114:11] codecs, certainly from about the 2000s onwards, [114:16] have a bit exact definition, so every implementation must produce [114:20] exactly the same bits, bit for bit, in exactly [114:24] the same data that comes out of a decoder. [114:26] - For like a large number of samples? [114:28] - For a given sample. So Lex's implementation, JB's implementation, and my [114:32] implementation of H.264 must match bit exactly. That [114:36] wasn't the case in the '90s of MPEG-2, probably fair to say [114:40] one of the biggest mistakes the video industry made, and I think people who were in the [114:44] room in '92, I don't think, most or both of us were in diapers, I suspect- ... but [114:48] have acknowledged I would give a shout-out to Yuri Reznik. [114:53] He's acknowledged that was one of the big mistakes of the era. [114:56] - And you're saying the encoders needed to be able to run tests and then the bit exactness. [115:00] the bit exactness. I mean, that's a nice thing to guarantee. [115:04] Like there's a parallel sort of development here on the way the web [115:08] browser works, which is a, you know, takes HTML and displays [115:12] it, and there's no bit exactness there across the different engines. [115:15] - I would point out actually FFmpeg is unique in the sense that it's, [115:20] it has been a winner-takes-all scenario. You have... Browsers is a good analogy [115:24] because it has to parse a lot of different content and render it in a particular way, like a [115:27] decoder. [115:29] But there still are multiple browser engines. There's Firefox's one, there's Chrome's one, there's a [115:32] few Japanese ones that are pretty decent. [115:35] That's not been the case in multimedia in general across a [115:39] wide range of codecs. FFmpeg has kind of [115:42] won it all, I suppose, in a sense because of, because of the fact that you can get [115:46] every new codec added is actually worth more than the value of that codec itself [115:50] because it makes the whole thing better. [115:52] - Man, this is really cool. Going to Perplexity. Yuriy Reznik is a [115:56] multimedia and signal processing researcher, got his PhD in computer science [116:00] from Kyiv University with over 150 papers and more than 80 [116:04] granted US patents, contributor to major multimedia standards [116:08] including H.264, MPEG-4, AVC-H.265 MPEG-4 ALS, G.718, and- [116:17] - G.71 is telco stuff. Telco. [116:19] - Oh. And so he was more connected to companies. [116:21] - RealAudio, RealVideo, right? That was- [116:23] - Oh, yeah [116:24] - ... very important at that time [116:25] - ... Zencoder, Brightcove, Contex. This, man, I need to hang out with [116:29] Yuriy. He's legit. And he's like one of the most nice person- [116:33] - Slack guy, yeah [116:33] - ... ever, right? Like for example, for my for my [116:37] startup that I'm doing right now called Kyber, right? I met Yuriy because [116:41] I met him every year at the Mile High Video Conference, which is in Denver. [116:45] And he gave me like so much good ideas [116:49] and good things. He's like really amazing person. [116:52] - He tells us how, how, you know- How great it is to be, you know, [116:56] even know us. And then we just like, you know, you look at that and it's, I think it's the other [117:00] way around, Yuriy [117:01] - That reminds me of a thing that you mentioned to me about FATE testing and, [117:05] like, the insanely rigorous process that's used to [117:09] test everything that's incorporated into into FFmpeg. Can you take [117:13] me through the testing process? [117:14] - Yeah. So FFmpeg has a system called FATE, FFmpeg Automated Testing Environment. [117:19] Because FFmpeg runs on so many different OSs and can be compiled with so [117:23] many different compilers, [117:25] there's been a crazy number of configurations. So you can see the [117:30] absurd combination of compiler variants, operating system variants, [117:36] instruction sets. You can see at the top macOS has tons of different [117:40] variants because it has iOS, it has tvOS. [117:42] - Well, I'm looking at a page fate.ffmpeg.org. [117:47] 81 minutes ago, 76 minutes ago, looking at the [117:51] different architectures, the operating systems, the different compilers, Apple Clang [117:55] version... [117:57] - Combinations are crazy. [117:58] - ... the combination is insane. RISC- [118:00] - So these are all run by volunteers, so these are all volunteer systems. The ones at the top, [118:04] for example, the Macs I host in my office, for example, [118:07] Host all sorts of different stuff. Other people host other things. So [118:12] it's really there to make sure... because FFmpeg does quite complex C [118:15] code, for example, you do have miscompilations. So the compiler [118:19] will sometimes compile C code incorrectly. For example, this [118:23] happens once in a while. [118:25] - Oh, there's like, there's a log of all the compilations. [118:28] - Yeah, log of all the compilations, all the tests. I think one of the other ones will show all the tests [118:32] passing. [118:33] - If you click, you can see all the tests- ... back. All tests successful. [118:37] - In logs test, yeah. So you see all those tests are passing [118:43] of all the different codecs, all the different filter transformations, all the- [118:47] The level of scale is quite crazy. [118:50] - Oh, that's nuts. [118:50] - On all the combinations. It's not just a matrix at this point. It's like a pivot table [118:54] of different combinations. [118:56] - That's nuts. [118:57] - And it's a key part of what we do because you may be able to test something [119:01] locally, you make a change, but actually that breaks [119:05] GCC version 11 on Mac or something like that, and you're [119:09] able to then fix that. We also have miscompilation, so the C code, [119:12] sometimes the compiler can have a bug in it where it creates the wrong output, [119:16] and that can have quite a big effect sometimes on a video because of the [119:20] way frames have dependencies. Even a small change in the output [119:24] can cascade to actually quite big glitches. [119:27] - You see PowerPC, you see RISC, you see ARM. [119:31] - There was PowerPC, there was RISC, there was weird stuff in the past like DEC Alpha. [119:34] There was— [119:35] - You see Visual Studio, different versions of Clang or GCC. [119:37] - Visual Studio, Intel compiler, Apple Clang, you name it. [119:40] - What are some of the pain points? Like maybe do you have [119:44] emotional triggers maybe nightmares about a [119:48] particular operating system, a particular container, codec combination of- [119:53] - I mean, for me, it's really easy because I have a day job. My company [119:57] builds... The company I started builds equipment for broadcasting sports matches [120:02] between TV stadiums and studios, for example. [120:05] We have to work with 10-bit video, and 10-bit video has a set of challenges that [120:11] you can't process 10-bit data natively on a CPU. So that means you [120:15] have to stick it in 16 bits. So that means you [120:19] have six wasted bits. So there's different packing formats to actually pack [120:23] the data more efficiently because when you send that over a network, [120:27] you lose... 'Cause you need to save that 40%. For example, on PCI [120:31] Express, you may only have bus bandwidth to do that. And so I think internally we [120:35] have about... Some are industry ones and some are internal to our [120:39] own hardware that we build. We have a, I think a 5 by 5 or 6 by 6 matrix of [120:45] every single format to every single other format conversion. In fact, one, [120:49] one of them I sent you, and they're all written in handwritten assembly, and they're all [120:52] written, and they all support different CPU generations. [120:57] So this is really traumatic, handling all these different combinations [121:01] times a million. [121:02] - By the way, the company you're talking about is Open Broadcast Systems. [121:04] - Yeah, so no, no relation to the free OBS streaming service. [121:08] But JB and I have started companies [121:11] broadly speaking around the FFmpeg VLC ethos, so that's really [121:15] low-level work. So in most companies, this wouldn't be written in [121:19] assembly. It would be accepted that C is fast. [121:23] As you can see from that, C is not fast. [121:26] - So here it says 62 times faster than C. [121:31] - Yeah. So it's taking those, the ethos of doing [121:34] low-level programming, real-time programming, [121:39] and using that for commercial applications, and JB and I have started companies around [121:43] that, in many cases hiring developers from the open source community [121:48] to use that ethos. And so that's a great example [121:52] of some of the things we're doing. In most companies, it would be, say, "Oh, [121:56] I'll write this in C and it's fast and we're done," but actually you can get a lot better. [121:59] - For me, like, some of the headaches we have is around some OS that are difficult to [122:07] support, right? Because if you look at VLC and thanks to FATE and FFmpeg, we run [122:14] on... The last version of VLC runs on Windows XP [122:19] and still run there and runs on Windows 11. We work on macOS 10.7 to the [122:26] latest macOS, whatever it is, right, 26. [122:30] We work on iOS since iOS 9, well, we are actually [122:34] iOS 26, right? We support many types of Linuxes, BSD, Solaris. The last [122:42] version still runs on OS/2, right? Like there is maybe 10 [122:46] users of OS/2 in the world, and one of them is maintaining VLC. Then you [122:50] realize that this very small team around VLC and using FFmpeg [122:54] codecs and all the other ones support more [122:58] OSs than Microsoft or Google or Apple, and they have infinite amount of power and [123:06] resources. But for example, the worst is iOS. [123:11] For in order to build on iOS 9, we need to do some [123:15] very clever mixing of several versions of the Xcode IDE and SDK from Apple, from [123:22] several versions, and do a type of Frankenstein version of that [123:26] so that we can still support iOS 9, which is not supported at all by the [123:30] compiler of Apple in order to still run on Arm32 on iOS 9. And [123:38] you've seen on FATE that it was still supporting iOS 9, right? So, [123:42] My headaches are mostly related to the support [123:46] of so many OSs. And it's important because, like, we [123:50] receive so many people saying, "Hey, thank you. I still have my iPad 2 to [123:54] watch movies," and it still works on iOS 9, right? [123:58] And it's also an impact of, like, not forcing people to buy new [124:02] hardware when it works fine if you optimize it correctly. Which brings us [124:06] to what we were saying about assembly. It's also fighting, like, [124:10] the fact that you need to buy something new nonstop while you could optimize [124:14] more, which is a lost art. [124:18] - You gotta tell me about this lost art or this, uh- [124:23] ... the carriers of the flame of assembly. What [124:28] what is, what is assembly? Why is it beautiful? [124:31] Why is it challenging? How does it work? [124:34] - So when you write assembly code, you write this using the instructions the [124:38] actual processor is using directly. So most of the time you would [124:42] write in a language, let's take C as a good example. The compiler would use that to [124:46] create assembly language and machine code [124:49] instructions for you based off your C code. [124:53] And there's a specific flavor of assembly that we use in FFmpeg that's called [124:57] SIMD, SIMD, single instruction, multiple data. So this means, for example, [125:03] say I want to add five to a number in [125:07] scalar assembly, so this is what's known as you work on an individual element. [125:11] So I wanna have a number of-- I have the number ten and I want to add five. I use the add [125:14] instruction, and I add five to ten, and I get 15. With SIMD, with [125:20] SIMD, I can have a whole vector of 16 different numbers. They could all be different. [125:25] If I want to add five to that, I can run one instruction, [125:29] and that one instruction sums all 16 elements. And that, as you can [125:35] imagine, lends itself very well to video. Video [125:39] is, you know, pixel grid, so I can perform operations on [125:43] multiple pixels at the same time. The key thing that we do differently in [125:46] FFmpeg is we don't use any abstractions or any major [125:50] abstractions on top of that. So there's a part of the world that uses what's [125:54] known as intrinsics. So these are C functions that behave very [125:57] similarly but not quite the same to writing assembly by hand. [126:02] So the registers that data is stored in [126:06] on the CPU, the compiler allocates those for you. [126:10] And so the key thing to understand was when we write SIMD is we have [126:13] a 10x, and not percentage, 10x to 50x speed improvement. That, that function is [126:20] 62x, um— [126:21] - That's nuts. [126:22] - ... on the FFmpeg account, as you know, posts and tweets a lot about that to try and say, [126:26] "Hey, we are doing this stuff." [126:29] - You are a person who sees the beauty in assembly, but it's also [126:32] extremely useful for these kinds of application to actually- [126:35] ... significantly outperform even C, which is crazy. [126:40] - It is necessary. Right? Because, like, one of the projects that we need to talk [126:45] about is called dav1d, right? So dav1d is a decoder for the format that was done by [126:53] Alliance for Open Media which is an, a video decoder called AV1. [126:58] - So if, for people who don't know, we've been talking about H.264. [127:02] AV1 is another hugely popular standard and codec that is increasingly taking [127:10] over the internet. [127:12] - And when this format was launched many people said, especially [127:16] even from the Alliance for Open Media, right, which is Google, Netflix, Amazon, [127:20] uh, Mozilla, say, "Well, this format is so complex, it must be [127:24] done in hardware to do decoding," right? And well, I [127:28] arrived with a few other people mostly [127:32] Ronald Henrik, and Martin, and we said, "We need to have an [127:36] extremely good software decoder because it's going to take [127:40] time to have hardware." And so we wrote this project, which is beyond insane. We are [127:47] talking about 30,000 line of C, but 240,000 lines of handwritten Assembly, right? [127:56] - Handwritten Assembly, 240,000 lines. [128:02] That's incredible. That mean-- I mean, some of the stuff we're talking about is probably the [128:05] biggest Assembly code bases. [128:09] - To give you an idea, and Kieran can correct me, but I think the FFmpeg has [128:14] 100,000 lines of Assembly for all the codecs. [128:17] - For all codecs. Mm-hmm. [128:17] - And just this one has 240,000. It's a VideoLAN project, of course. And [128:24] it is optimized at the maximum because the [128:28] motto when we're starting the project is every cycle [128:32] matters, right? Every cycle matters because David is [128:36] used in VLC and in some software AV1 playback stacks. [128:40] We are talking about probably 3 billion devices which [128:44] are going to decode video nonstop because, for example, 30% of the [128:48] video from Netflix are now in AV1, 50% of YouTube, [128:52] right? So, and you often don't have a hardware decoder because not [128:56] many devices have a hardware decoder. And with dav1d, we realized that [129:00] with one or two cores you were able to decode 720p correctly. So it is—literally— [129:08] - Yeah, that's dav1d [129:08] - ... incredible, right? [129:09] - That's dav1d. Look at that Lex. [129:10] - Uh, yeah, so this is another spicy tweet from [129:14] you. This is what peak video codec should look like, 79.9% assembly— [129:20] - That's almost [129:20] - ... 19.6% C and 0.5% other. [129:25] - And what's incredible is with those tweets, which is factual, people get [129:31] crazy. They are unhappy, right? They say- [129:33] - For a year, for the last two years they go crazy, "No, intrinsics is fine. The compiler is..." Oh, they [129:37] go, "I have never-" [129:37] - "You can optimize your compiler, auto-vectorization, it's your fault, you don't [129:41] understand." And we've tried that forever, right? [129:44] - For two years, and two years later, showing hundreds of examples [129:49] of handwritten assembly. "No, no, no, you're doing it wrong. The compiler can do this." [129:52] - So we should actually just articulate a little clearer. So the intuition there from the [129:56] software engineering folks, when you have code like... Okay, let's just take an [130:00] example, C++. There's a compiler that's doing a lot of the optimization. [130:05] - Yes. [130:06] - And the presumption is if you have a good enough compiler, if you continue to [130:10] improve the compiler, you're going to generate code- [130:13] ... that can perform like optimal performance. You cannot possibly beat it. [130:18] - Yeah. Yes. [130:18] - And you're consistently challenging that thought that if you do- [130:21] - By orders of magnitude [130:22] - ... by orders of magnitude- ... handcrafted assembly can outperform C. [130:27] - The two things that they tell us is, yeah, but modern compilers [130:31] have auto-vectorization, right? Because SIMD that we're doing is [130:34] vectorization. And like it's not even close, right? [130:38] It's not even close, right? It's not like 5%, 10% slower. [130:41] It's multiple times slower. [130:43] - So can we... I don't know if you can say something philosophically, because there's a lot of great software engineers, great engineers, great machine learning [130:47] lot of great software engineers, great engineers, great machine learning [130:51] people. Karpathy will listen to this and say, "What's the intuition he's [130:55] supposed to get from this? What are we supposed to..." [130:57] - Karpathy learnt assembly because of the tweets by the way. I just... He start- He went, he's like, "Oh, I [131:01] think this is a movement." [131:01] - He's like, "Let me figure out what's happening here." [131:02] - No, no, he, and you know the way he documents his work and so. [131:05] - Philosophically, what's important to realize is that we [131:09] passed the time where hardware was going so much faster, right? [131:13] We are at the end of Moore's law. We have limitation for for AI, [131:17] for memory. You need to go down in the stack and optimize [131:21] more to get more power from what you have, because our [131:25] request for power, CPU power, GPU power [131:29] are exploding while the hardware is not exploding in speed, right? [131:35] So you- what people do is that they add more cores, right? But that's basically [131:39] like at some point you can add 250 cores, right? So what we [131:43] do is to take every inch of the machine. [131:46] - Not just that, not just that. We abuse the machine. We go and use, we [131:50] use the machine in ways that the, [131:53] that the creator didn't expect. Sometimes we use an instruction that's completely unrelated to [131:57] what we do. We use a cryptography instruction in video processing to do nothing [132:01] related. [132:01] - And one of other things that we do, for example in dav1d, which is a bit [132:05] crazy, is that we don't use the function [132:10] calling convention from the operating system. [132:13] - We should explain that. [132:14] - That is extremely- [132:15] ... complex. But basically, usually when you do move from one function in, [132:19] in code to another, there is a way to save the registers, the [132:23] state of the CPU to enter another function. And this is like standard. [132:28] - It's a bit complex. I would simplify this a bit. So, [132:31] so dav1d does things to abuse the calling convention. You could define the calling [132:35] convention as [132:36] I've written a function and I want to call another function. How is the data [132:40] shared between the functions? Because there's a convention, what's known as a calling [132:44] convention, and what dav1d does for optimal reasons is [132:49] create its own calling convention sometimes. So if I wanna call Lex [132:53] Fridman's library, we got, we've got to agree on a convention so that I can share [132:57] data with you in the assembly language space. And one of the challenges [133:01] in assembly is every operating s- well, not every operating system, but there are, well, at least [133:05] four that I can think of on x86, Linux 32-bit, Windows [133:08] 32-bit, Windows 64, Linux 64. They all have their own calling [133:12] conventions. And so one of the amazing things Loren Merritt did, who we talked about [133:16] before, was create a very lightweight abstraction layer, so you could write your [133:20] assembly code once and it handled all the calling convention stuff for you, [133:25] which was always a problem because you had to manage four different variants. But [133:29] dav1d takes this even further, for speed reasons it does its own calling convention, [133:34] within itself to bypass the kind of rules, the rules [133:38] of, the rules of sort of functions and say, "Okay, actually I'm gonna call a function this [133:42] way because I know it's within my library." [133:45] - Does it have to be special to every single operating system? [133:48] - Well, if it's custom, no. But the, the, the challenge is [133:51] in general, yes, and in terms of, in terms of each instruction set. So the [133:55] thing to also emphasize is we do this on every instruction set. So every [133:59] instruction set has its own handwritten assembly, which is even more crazy. And [134:03] that, that, that matrix has got bigger in recent years because [134:07] of RISC-V, because of ARM64, because of the new SVE. There's SME. x86 has [134:14] AVX-512, AVX. So we do runtime [134:18] processor detection. We see what the machine FFmpeg is running on or dav1d's [134:22] running on is capable of, because you could be on a laptop from 2008 [134:26] where this isn't there. Runtime detection, we set function pointers accordingly. [134:31] And then from then on, off you go. [134:34] - Or you could be on a machine with RISC-V. [134:36] - Yes. And in all that, we don't even respect the calling [134:39] convention of the operating system in order to be faster, because we know that [134:44] we are going to be called from within our binary, so we can share data [134:47] without saving all the registers in the common way, because that can lead [134:51] to loading and saving registers on the L1 and [134:55] L2 CPU and gets us faster. So that's why I said that understanding CPU architecture, [135:03] computer architecture is key. And this is also why it's handwritten. I don't know [135:07] anyone, I've never heard any other project than dav1d doing that. This [135:11] is why Kieran calls it, calls it an art, right? It is an art. [135:14] - I think in a mass world, there isn't something on billions of devices. [135:18] I know there are some specialist industries. I know in high-frequency trading, they take this [135:22] really seriously, where they're receiving feeds from a market, and they need to react [135:26] within X number of microseconds, and so the instructions matter. But that's not [135:30] a mass, you know, a mass-produced thing that's on a billion devices. That's [135:33] hyper-specialized, running on hyper-specialized hardware. We're running on [135:37] all hardware from- [135:39] - Sorry to linger on it, but, like, that's a really [135:42] counterintuitive, almost, like, revolutionary [135:46] idea here, that there's a huge amount of value to assembly. [135:50] Like, what are we supposed to take away from that? Like, what... You know, there's a bunch of people listening [135:54] to this, they're basically like, sorry, for myself [135:58] included, you know, I programmed for many, many years in C/C++, [136:02] going up the standards of C++, fell in love with C++, even meta programming and so [136:06] on, and then transitioned more and more because of machine learning about 15 [136:10] years ago to Python. And so, like, for me in this Python world, JavaScript [136:15] world, now vibe coding, where I'm just using natural [136:19] language, sitting in my jacuzzi, drinking a drink... and just talking to the computer, [136:24] re- like, like record stops. Why is the value [136:28] to go back all the way down to the low level? Like, what's the intuition? [136:31] - Because you can get more power per dollar [136:35] invested, right? And sometimes it's going to be a [136:38] problem that is limited by your hardware. [136:41] A good analogy is what you see in quantization [136:45] in LLMs, right? And people are doing, "Oh, I'm going to do that in FP8 [136:49] or FP4 or some crazy things like Microsoft Phi, who did it in 1.5," [136:55] because you're constrained by memory, because you're constrained by the machine you [136:59] can run. Because at some point we are doing real time, [137:03] and I believe this is going to happen on AI inference also, is that at [137:07] some point you need to get faster, and you cannot always get harder, [137:11] More powerful hardware, right? So you need to [137:14] analyze code and see where, like, where is the [137:18] mission critical, where is the things that are called nonstops. And [137:22] for example, dav1d is a good example. It's going to be run [137:26] billions of hours per day. [137:28] That makes sense. It doesn't make sense to be on the glue of FFmpeg- [137:33] - No [137:33] - ... uh, CLI. It makes sense over there. [137:35] - Yeah, and this has to do, also we'll talk about it more, but your new effort, your [137:39] new company, Kyber, is doing that kind of thing for [137:43] ultra-low latency, so the slogan being, "Every millisecond counts." And [137:47] when you actually extremely highly constrained in some dimension- [137:53] - We are also arriving at a point where we've done [137:57] so many great things, but the hardware is getting back to us, right? Because [138:01] cost is increasing, because we need more power, and so you're [138:05] limited by either your CPU, your RAM, or your [138:09] networking, and you need to optimize, and this is where [138:13] value is going to be. Especially because, like, doing AI is going to help [138:17] do the programming of, like, business, right? And so [138:21] the core thing that you will not be able to vibe [138:25] code are optimization for the hardware to be as fast as is possible. [138:31] - I'd love to talk to you about who and how [138:34] should learn assembly, but first, I think we need a bathroom break. [138:39] Quick ten-second thank you to our sponsors. Check them out in the description. It really [138:43] is the best way to support this podcast. Go to lexfridman.com/sponsors. And now back [138:49] to the episode. All right, and we're back. There's this nice repo with the [138:55] assembly lessons. First of all, do you think developers should [138:59] learn how to program in assembly, and how would you go about learning it? What is [139:05] this asm-lessons? [139:08] - So I personally wasn't happy with the way [139:11] assembly is taught in books and online, 'cause it's very [139:15] grammar-focused, and you don't, in general, learn a language from [139:20] learning the grammar and the structure. You learn a language by [139:24] asking someone what their name is, and you start from there, and you go and solve real problems, [139:28] That you have when you want to communicate. You don't learn sentence [139:32] structure, and this is the interrogative and the adverb, and all, all the assembly books seem to be [139:35] doing like that, going through every instruction, even ones that aren't really relevant, [139:40] explaining what they all do and how they... It, it actually doesn't really change much. [139:43] So, [139:45] and the other problem that we have in our community is assembly is taught sort of hand to hand, [139:49] like person to person, like blacksmithing one by one. That's, that's the only logical [139:53] sort of analogy, and that doesn't really scale online. It doesn't [139:57] do other things. So this... I've started a set of assembly [140:00] lessons in the, in the way it's done in FFmpeg, which is a little bit different to the [140:04] way assembly in general [140:08] for... I don't know. I'm trying to think the other good big use case of assembly is in [140:11] embedded devices, in really low power, cheap devices, and that's [140:15] completely different to what we're doing here. [140:17] I think it would be good if you could highlight the requirements, which are quite simple. [140:20] It's high school mathematics and C. And actually not even C, really, really [140:24] it's pointers. To emphasize, yes, we've talked about how brilliant this stuff is, [140:28] but high schoolers like Daniel Kang have written [140:33] assembly in FFmpeg. I think there's been contributions because of these lessons. [140:38] So it's really about [140:40] trying to get this dying art to continue, because we've shown it's [140:43] possible with dav1d to produce something amazing. [140:46] There's still a lot of codecs in FFmpeg that are only maybe partially [140:49] assembly optimized. [140:53] And so it really, it really starts with basics and continues, explains a lot of the jargon, a lot of [140:57] the syntax. It doesn't really try and explain to you, you [141:01] know interrupt handlers and interrupt instructions and all of these [141:05] different jump targets actually makes this really vector focused. [141:08] - And describes all kinds of registers: general purpose registers, vector registers, [141:15] Really nice examples. Oh, this is cool. [141:18] - It's a classic, yeah, it's a classic example of FFmpeg. But [141:21] some of this assembly language is really beautiful, and I think it's beautiful [141:25] because it's kind of like flying a Spitfire. It's [141:27] really aviation at its purest, but also [141:31] pushing the aircraft beyond what the designer thought was [141:35] possible. So we're abusing, for example, sometimes cryptography instructions to [141:39] do certain things, and there's a level of beauty and art where [141:43] it's really you and the processor. There's nothing in between. It's you and the [141:50] joystick of the cockpit, and you move that joystick, and it's physically connected to the [141:53] ailerons, and you can push that plane beyond what it can normally do, and there's a [141:58] level of, yeah, beauty and amazingness to go that. But I don't think [142:05] the sort of person-by-person assembly that is... someone taught me, and I've taught multiple people, [142:09] is gonna work long run [142:13] just because of the particular flavor and the way that we do it. [142:16] - It's literally no, I should... I was gonna say [142:20] wizards handing it down. Um, I realize I look like a wizard- [142:23] ... wearing this hat. But you're basically just like the sages, the wise [142:27] sages handing- ... down the craft. Can I ask you about LLMs? Like- [142:31] ... can they help? [142:32] - They had more of an understanding than I expected, but they are still... [142:36] I've asked it questions, and it still goes and [142:39] starts hallucinat- not hallucinating, but making modifications, and then I go, [142:43] "Is it bit exact?" "No." "Fix it." And then it just goes and does the same thing, and [142:47] it's going, it... There isn't the corpus of information like Stack [142:51] Overflow to work on. [142:52] - There is not enough data to train on. [142:55] And this is the biggest issue. Um, I started my career [142:59] actually doing some assembly for Itanium, [143:03] right? So the Itanium is a dead processor type, right, which was done [143:07] by Intel and HP a long time ago when they wanted to do 64 bits. [143:11] Well, they lost, and then we got AMD, who did it, AMD [143:15] 64, which became x86-64. But Itanium was [143:19] extremely interesting in the sense that those were [143:23] processors who had a ton of computing power to do floats, [143:27] FMAs, which is similar to what we need now for, for LLMs, right? [143:31] And you could pack three operations per [143:35] line that could be loaded. So basically, you had an output [143:38] of basically six billion of operation per second, [143:42] but the bus, the memory bus only allowed [143:46] 1.5, right? So your, your CPU was four times faster, so [143:50] you had to do crazy things to, to pack things in memory or [143:54] reuse the registers, and those type of semantics, no language [143:58] could do that, right? So like I have [144:02] the Itanium programming book because Intel did amazing books, [144:06] but that's exactly what Kieran says. If you don't know what you're, [144:10] you're going to do, it's impossible to read, right? It's a ton of jargon and [144:14] so on. While those lessons [144:17] are amazing because they are targeted to a real problem, and you can do it yourself. [144:21] - And people have. People have. There are patches, and they said, "Oh, I studied your lessons, and here's my first [144:25] changes." [144:26] - That's amazing. [144:27] - And part of that in the lessons is a framework called [144:31] x86inc, written by Loren when, when he was working [144:35] on x264, and it allows you to do more things about [144:39] that to create a type of like not caring too much about [144:42] different calling convention. And we had a lot of students who, [144:48] Gave code to x264 using that a long time ago, right? [144:52] So it's really doable, and I believe it's [144:56] necessary to understand assembly language, even if [145:00] you don't do it much, to understand what's going on inside your computer, [145:04] and that will make you a better programmer. And I assure you that [145:08] because doing that, you will understand some of the architecture of the memory [145:12] inside your computer, right? Understanding register, L1, L2, [145:16] L3, RAM, SSD, disk, and so on, [145:20] which are very important because then you have a good [145:24] programming culture that will make you a better programmer. [145:27] - Uh, what do you think about the Rust programming language? 'Cause that's a bit of a meme. [145:31] - We have very different opinions with Kieran. [145:33] - I think it's valuable what they're doing in terms of memory safety as a concept. [145:37] - Can it achieve some of the speed up that assembly achieves? [145:42] - Oh, not assembly by hand, no. I think that that's a given. C potentially, [145:46] but I see it very... It has a very big Esperanto vibe [145:49] about it. It's like we're gonna solve this, and we're doing this in a particular [145:53] way. [145:54] - Meaning it's a bit too utopian? [145:56] - There's a lot of focus on the self-importance rather than solving real-world problems. [146:00] It reminds me of the Sinclair C5. Sir Clive Sinclair of Sinclair [146:04] Computers built a car, and he said, "Oh, everyone will be traveling around [146:08] in one of these electric cars." And it was... Rust reminds me of that, where [146:15] I think the community doesn't quite understand that [146:19] in order to get people to move, you have to build something that's as good as, if not [146:22] better than what you have now. Yes, people are doing Rust rewrites, but if they're, [146:29] if they only do 85, 90% of the feature set [146:33] of what we need, like things like coreutils, that last [146:36] 1% takes 99% of the time. To use [146:40] Elon's famous quote, "Prototypes are easy." Like this kind of stuff is easy. But this, to get [146:44] a real electric car, you have to make a car as good as, if not better than what we have now, and [146:48] Rust isn't in that stage yet. I don't think anyone would [146:52] object to seeing Rust code in FFmpeg, [146:56] but it needs to work as well and support the same unit testing as everything [147:00] else. It needs to be flawless. It can't just randomly break. They can't just randomly [147:04] break ABI when they want to. It needs to have, I think, [147:08] more-- I think it still has only one compiler implementation. [147:12] So it, it's got to be as good as, if not better, and saying, "Hey, here's [147:17] my utopia of memory safety," isn't enough, even though we [147:21] probably all agree that that's the goal. [147:24] - So I've done a ton of Rust, and the two major [147:28] topics I had was adding Rust modules inside VLC. [147:32] One of the reasons VLC got popular and which was one of the main [147:36] architectural decision, is that VLC is a very small core and a [147:40] ton of modules, right? And so you can write modules in C, in [147:45] C++, in Objective-C, and anything that is basically interoperable [147:49] with C. And so we did some Rust [147:53] modules, and so I have experience on that, and I wrote some of it. And also, [147:57] like, my new startup called Kyber, is an open source project [148:01] mainly done in Rust. What Rust is extremely good in, in the [148:07] sense that it's a better C++ that cares about memory and allows you to do [148:13] things about memory ownership that no one else can do so far. [148:19] However, it's great when you start a new project from scratch, and you [148:23] do everything in Rust. But it's very not good when [148:27] you interop with existing part. And some part of the Rust [148:31] community believes that they need to rewrite everything, and everything will be better with [148:35] Rust. And the answer is like, no. Like, I'm almost always, in all my [148:41] years of being engineer, manager, CTO of startup and so on, don't rewrite, right? [148:47] - Is that-- That's the initial instinct for a lot of people when they [148:51] show up to a code base probably before LLMs, is [148:55] like probably because they don't understand [148:59] the wisdom of the way things have been done in the past. They say, "Well, we need to [149:03] rewrite it." Hence why there's a thousand JavaScript frameworks. [149:06] - But the reason is the following, [149:09] and this is very important to understand. It is an order of [149:12] magnitude easier to write code than read code. [149:17] And you see that also with LLM. They can write code, but analyzing is a lot- [149:22] ... more difficult. And so when you arrive and when [149:26] you arrive to a very complex piece of code, right? You don't understand [149:30] it, right? Because it's so much more effort to understand the code [149:34] from someone else because you don't have the thought process. Um, [149:38] And often I joke about some languages [149:42] mostly Perl, for example which has very complex [149:46] syntax. And imagine I am at my maximum [149:49] intellectual efficiency in programming, right? [149:53] And I write the best code ever. I will not be able to understand [149:56] myself six months later, right? Because reading code is more difficult. So [150:02] very often you arrive, you don't understand all the wisdom, all the business logic, [150:06] the reasons that were done that is maybe not documented. And you say, "Well, [150:10] I'm going to write it." And the thing is, no, you don't, right? [150:14] Because that's, as Kieran said, right? I'm going to rewrite coreutils in Rust. And [150:18] then, of course, you arrive very quickly at eighty percent [150:22] then ninety percent, takes a bit more time, and then you got the last ones, right? [150:27] On the other side, right? So for new projects, it's great. Everything related [150:31] to parsing files networking because of the [150:35] memory checker, boundary checker, it's amazing, and there is nothing else. [150:40] To answer a bit differently for us, [150:44] imagine I take a piece of software like dav1d or [150:48] x264, right? Which has a ton of runtime in assembly, right? [150:52] Um, I rewrite the C part in Rust, right? So it's more secure. [150:56] Yes. But then you arrive into the assembly, and you can jump [151:00] anywhere in the memory because we are doing handwritten assembly. So [151:04] even if I rewrite the C part in Rust, for security [151:08] reason, you break all the security when you [151:12] you write handwritten assembly because we can jump anywhere. [151:16] So in my opinion, we need to do something that is [151:20] secure assembly, right? So which is compile time, check the [151:24] assembly, which is similar to the checkasm projects that we're [151:28] doing on dav1d and x264 with VideoLAN, is to start instrumenting your [151:35] assembly at compile time to check that it's not jumping anywhere in the memory. [151:39] Because else you might rewrite a part of C in Rust, but if [151:43] you want to have the same performances, you're going to have inline assembly, and so you destroy your whole [151:47] security model. So that's a bit what I think about Rust. [151:51] - No, I just wanna... I would say on a personal level, I'm so in awe [151:55] about assembly. I actually-- [151:58] Once in a... It never gets old, the speed improvements to show sixty-two [152:02] x. So there are months, on a personal level, I run [152:05] our internal test suite at work and just see I'm still in awe at the gains we have. [152:10] - Well, there's a source of joy and happiness with programming for different reasons. [152:15] But I think one of the greatest happinesses is in the optimization of code. [152:21] And it sounds like you're, like, at the cutting edge of that. [152:24] - I was like, "Whoa, that was cool." [152:25] - And in the community, I want to speak about two people who are [152:30] wizards of assembly, right? The two of them are actually working [152:34] living in north of Europe Sweden and Finland. And [152:42] Henrik Gramner knows so much about Intel x86 assembly that when we ask questions at [152:50] Intel about things, they tell, like, "Why are you asking us, Intel? [152:54] You have Henrik. Henrik knows better." He knows all the cycles [152:58] of almost all the SIMD instruction by all the CPU [153:03] generation. "Oh, yes, this is a P4, this is a Nehalem, this is a Core 2," et cetera. [153:08] That person is, like, the best person on assembly in the world. [153:12] And he's the nicest person that you've seen, [153:16] like, very... He arrives, you don't see he's [153:20] amazing. And the other one is called Martin, [153:24] Martin Storsjö, and he's-- they're doing mostly the same [153:28] on Arm, right? So Neon, right? And iPhones and Androids and so on. And he codes in [153:35] assembly on his phone, editing it with the crappy keyboard, [153:43] like virtual keyboard you have while watching his kids [153:47] play in the playground, right? Like, like this is just like [153:51] wizard level. So those two people are like- [153:55] - Yes. So when you're programming assembly at [153:59] that high level, a part of that is knowing the architecture that you're programming on. [154:03] - Yes. On Arm in particular, yes [154:04] - ... Arm in particular. But x86, I mean, these are complicated architectures, right? [154:09] - Yeah. But Arm in some ways is more com... x86 with, [154:13] Out of order execution is not so bad. Arm, you really need to understand all the [154:17] different generations of Arm processor because they're all different. There's A72, [154:22] ... et cetera, et cetera. And there's the Apple variant, there's this variant, there's that, and you need to write code that [154:26] works [154:27] efficiently on all of them. x86, well, broadly speaking, you have Intel, AMD, and you have [154:31] sub-variants, but generally speaking, there's... [154:36] Something fast is gonna remain fast on all of the variants, whereas in Arm it's a [154:39] completely much more complicated ballgame. [154:43] - We're taking a nonlinear journey through history here, but we're [154:46] talking about Michael Niedermayer. And I wanted to ask [154:52] about this. For a time there was a split in FFmpeg and Libav. [155:00] - Yes. So in open source projects sometimes you disagree, right? Um- [155:10] - You have such a nice way of putting it, yeah. [155:12] - And the good thing is because of the license, you're allowed to basically do your [155:16] own, right? Um, and this is normal, and this has happened all the time, right? At a [155:21] point there was a GCC at the time of GCC 2 and EGCS which became then GCC 3, right? [155:29] There is what we told KHTML with WebKit, with Blink. Um, [155:34] it is a same process. And also, like when I want to do a new [155:37] feature today in VLC, I fork, I do my thing on my own, and then I merge [155:41] back to the community. So there was a split in the open source community on [155:45] FFmpeg, which become Libav and FFmpeg. And after a few years, [155:49] well, the community merged back and people moved on. It's a bit, um, [155:54] drama that is normal in open source community, but forks [155:58] are even... They're important because they change the, [156:02] the status quo of a community. Um [156:06] not talking about FFmpeg and Libav here, but the, or the GCC [156:10] fork made GCC a ton better because the, some [156:13] people wanted to change the architecture fundamentally to make it [156:17] faster. And of course, it's always question of people and [156:21] so on, but in the end you realize that FFmpeg today is [156:25] better than it was before the fork. And [156:29] now, well, we're back all together, right? And I spent a [156:33] lot of time, and, and Kieran can say in the, in, in the community. [156:37] It's not often, to be honest, very [156:40] well explained because a ton of the reasons are not very public. [156:45] But I think that's, that's normal and that's good. [156:49] - Yeah. I mean, you're making it sound really nice, but there is battle, there's pretty heated [156:53] battles inside open source projects. I mean, it is a very passionate community and [156:57] you're kind of in a distributed way have to define the direction of things. [157:01] So here looking at Perplexity, "FFmpeg and Libav [157:05] split in 2011 mainly over project governance, leadership style, and [157:09] development processes, not because of a fundamental technical disagreement. [157:13] Uh, FFmpeg effectively absorbed Libav's work [157:17] while Libav withered and most distributions and developers moved back to [157:21] FFmpeg." Yeah, that was a, that was a weird experience 'cause, you know, I'm a Linux user, [157:24] perspective, that was a weird experience 'cause, you know, I'm a Linux user, [157:29] so, you know, whether it's Ubuntu and so on, all of a sudden, I think for, for a [157:33] for, for a little bit, Ubuntu, I feel like, am I remembering [157:37] correctly, switched to Libav and- [157:39] - 12, 14, something like that. Yes. Something like that. [157:41] - And then they switched back to FFmpeg. I was like, "What is happening?" [157:45] So on the sort of you get to feel the ripple effects [157:49] of the different internal debates that are happening. [157:53] - To be fair, on Apple, when you type GCC, you get Clang. Like they, they did [157:57] something like that as well, so. [157:58] - Yeah. So, so to me it's like the fork was like heated drama, but most of the [158:05] development from Libav was merged back into FFmpeg, right? [158:08] So de facto FFmpeg got a a superset around [158:12] Libav, and so that gave the user, because in the end we work the user, for the [158:16] users, a, a larger set of features and a ton of things that were, um, discussed. [158:22] For example, the debate on reviews, on, on how we push are [158:26] something that now is completely settled in FFmpeg and is following what mostly what [158:33] everyone in the community agrees, right? So de facto, everyone who [158:37] was active on Libav came back in work on [158:40] FFmpeg because the disagreements were fixed, and [158:44] in the end, FFmpeg is stronger than it, it was before, right? And- [158:49] ... I know people love drama, but, um- [158:52] - Well, my main concern, I understand, and I think [158:56] looking at the, the long history, it's all for the good. [159:02] But I do... I am concerned because there's so few humans [159:06] that are critical to the success of open source projects that I have seen it, [159:11] Be a psychological toll on folks [159:16] and, you know, sometimes leads to burnout. So you have these incredible people that are at [159:20] the core of open source projects. There is a moment that happens [159:24] 'cause, like, what is the motivation of doing it? Ultimately, it's because you're passionate about it [159:28] and it makes you happy. Then at a certain point, you wake up and it's like, "This's been a [159:32] bit too much heat from the drama. So, like, at the, at [159:36] the project level, the project continues and often flourishes. [159:40] But sometimes there's these individual humans that are just like- [159:44] - But- [159:44] - ... I've had enough. [159:45] - Yeah, but it's not just about forks, right? So it's a g- very, uh- [159:48] ... what, what you, what you are referring to is [159:52] one of the most challenging and most interesting part of open source [159:55] today is maintainers burnout, right? [160:00] And AI is a problem because of that. And Daniel [160:04] Stenberg, which is the maintainer of curl who's probably one [160:08] of the best promoter of open source in the world. [160:12] He's, by the way, a member of the European Open Source Academy with me, so I'm [160:16] very, like, humbled to be on the same community as him, right? He's against what [160:20] he call AI slop, right? Because it gives a ton of, um fake reports or- [160:27] ... bad reports, bad patches, and then a lot of maintainers have [160:31] a lot of burden to maintain the software. And this is straining the [160:39] mine of open source developers much more than forks. [160:42] Uh, and for example, the XZ fiasco was because there was [160:46] one guy maintaining it, and he got basically hammered by two [160:50] attackers who were asking him questions nonstop at weird times at [160:54] night to block him, and at some point he got fed up and says, "Okay, I can't do [160:58] that," and gave the commit access to the attacker. [161:01] Um, so burnout in open source community is [161:05] something that exists but mostly it's about maintaining things, right? [161:11] - No, for sure. But I wonder how do we help that, 'cause those people are so [161:15] important. The- [161:16] ... the human beings are so important to the core of these pro- projects. [161:19] - So, so for example, now I am maintaining a ton of multimedia and non-multimedia [161:23] library- ... as maintainer because the maintainers [161:27] got fed up, right? Some on VideoLAN, some outside of VideoLAN, [161:31] Because it's sometimes you need a tough [161:35] skin, right? Because you get, like, it's not really attacks, but oh, this [161:39] is not working, this is not working, and you feel it personally. And this is [161:43] also why resources or the, the Google fiasco is, [161:48] was a problem, right? They don't realize that in the end you have, [161:52] You know, it's like the same graph where you see, like, everything and it's just like [161:56] one random open source project that is maintaining the whole- [161:59] - The Nebraska thing, yeah [162:00] - ... internet. You see the one, right? The- [162:01] - Yeah, this is the meme. I mean, it applies to, to a lot of open source projects. [162:05] But this is [162:07] the all modern digital multimedia infrastructure, and then that thing at the very [162:11] bottom that everything relies on is FFmpeg. It's [162:15] true. And then there's usually, you know, a handful of folks that are maintaining [162:19] that. [162:19] - And FFmpeg or VLC, right, you have a community of 10, 15 [162:23] core developers, are not the worst open source project. XZ, [162:27] which is even in more installations, is one person, right? There is one guy- [162:32] - libxml is, uh- [162:34] - Yeah, libxml, right? There was a big stop. No one is maintaining- [162:37] ... libxml anymore, which is like parser, the only library that is able to parse XML [162:41] everywhere. [162:41] - All the crazy edge cases of XML under ridiculous circumstances, and they [162:45] get attacked by security researchers because there's one other [162:49] crazy edge case that they haven't thought of, and it's like, yeah, but the body of [162:53] knowledge to actually resolve that is massive. [162:56] - There is one guy maintaining all the time zones for everyone who is in the [163:00] middle of, I think, was it Nebraska or- [163:02] - Yeah, it could be, yeah [163:02] - ... South Dakota? Like, the mental health [163:06] of the open source maintainers is something that large [163:09] corporations don't care or don't see, right? It's just like, "Oh, yeah, I'm just [163:13] doing an open source report," and so on. [163:16] - Mm. Some of it is financial, but some of it, and [163:20] people should definitely support open source financially— ... all across the board. [163:24] But some of it is also, like, spiritual on a basic human level. There's something that [163:28] happens, [163:29] like, with this image of F- FFmpeg and so much of the internet depending on it, [163:34] where people almost, like, talk down to the [163:38] folks who are carrying these projects forward and maintaining it. [163:40] - In the security community, they certainly did. That was one of, that was one of the things I think that [163:44] argument came out is [163:47] there was, there was a portion of the security community who's like, "No, these guys write crap code. [163:51] They need to fix their crap code." I'm like, "No, no, no, no. This is a guy's hobby project. [163:55] You've, you've have a security bot that's gone and found some AI-generated stuff. [163:59] That guy didn't write crap code. It's just [164:02] an edge case to the 99.99999 percentile he [164:06] didn't think about because it's his hobby project decoding Star Wars games." [164:10] - Forget the hobby project aspect of it. It's, it's just hard work, and it's [164:14] it's beautiful, and it's like the, the right approach there is to celebrate people- [164:18] ... for doing incredible, incredible work. It's, it's just [164:22] incredible that humans step up- [164:25] ... not getting really paid at, at first or maybe ever, and then they're [164:29] doing it out of the love of it, and we need to, like, human [164:32] civilization runs on people like that. We need to celebrate them. [164:36] - To, to give you an idea, I received death threats on VideoLAN, right? And, um— [164:40] - You mentioned that to me. Like, what, what is, what is behind that? [164:43] - So that must be, what, 2009, 2010, right? Um Apple is moving from PowerPC to Core [164:50] Duo, um, that probably in 2006, and by [164:54] 2009 or 2010, I decide that we are not going to do new versions of VLC for PowerPC. [165:02] At that time, like VLC, we were close to the number 1.0 [165:06] release. We were four of us, right? Like, just like, "No, this is not [165:10] possible." So I receive a death threat with some powder in it, [165:14] right? It-- Remember there was some- ... anth- anthrax threats- [165:17] ... at that time, right? And it was because I had [165:22] taken the decision to not maintain the PowerPC port [165:25] anymore. And of course, it wasn't anthrax, of course. It was some type of [165:29] flour and so on. But I received that as a, with a letter of like, [165:33] "You, you piece of shit, you should die, PowerPC forever," and so on. And it was 2009 or [165:41] 2010, right? I was, I was young. I was just like, "Why? What did I do?" [165:47] - Yeah, that can break your spirit. It's like, why- [165:49] - My mother freaked out, right? We had to go to see the police and so on. And now, [165:53] like, I'm going to say that I'm quite happy that this happened [165:57] at that time. It forged me a lot, right? I am... [166:01] I can see, I can take a lot of hate on me. I'm okay with it, right? [166:07] - It sucks that that's part of reality, 'cause all the people that love VLC, [166:11] all the people that love FFmpeg, like me, you know, I legitimately [166:19] hundreds—probably thousands of times in my life had a smile on my [166:23] face because FFmpeg made me happy, period. And how [166:27] many times did I get a chance to say that? Zero. Until I [166:30] realized there's a Twitter account. And every once in a while I'm, like, [166:34] messaging it. [166:35] - One of the things I like on the Reddit meme about me, which I don't like this meme [166:39] for a lot of reasons, but... And someone says, "Oh, JB is on, is on [166:43] Reddit," which I am, right? And I say, and say hello, right? And then I got so many [166:47] people who say, "Oh, thank you for VLC." And, like, I take pictures, and then [166:51] I share that to the Signal, to IRC. Uh, yes, we use IRC on different- [166:58] - I saw as a quick tangent, you mentioned IRC is like Slack for old [167:01] people. So you still use IRC? [167:03] - Of course. [167:03] - Yeah. I have it on my phone as well. [167:05] - Of course. [167:05] - Every day. [167:06] - Works fine. [167:07] - Wow. It works fine, huh? [167:08] - Works fine, yes. [167:08] - You have to power with a crank, I guess. [167:10] - No, but there's no- [167:12] - There's AOL. There's AOL as your social media. [167:13] - There's no ads, there's no tracking, there's nothing. Like, it's, uh- [167:16] - The biggest issue, to be honest, right, compared to Slack is that it doesn't have [167:20] threads. [167:20] That's annoying. It doesn't have emojis for reaction. Sometimes it, it would be [167:24] nice. [167:24] - IRCv3 has. [167:25] - Yes, v3, but no one does it, and you cannot edit your messages. [167:29] Right? And the rest, it works perfectly fine forever. [167:31] - But how do you communicate without emojis? [167:33] - Well, that's, that's why I said it's for old people. [167:35] - Old people. All right. [167:37] - And we do emojis with like- ... you know, the colons and dash and- [167:41] ... parentheses, right? So. [167:43] - Old school. So anyway, you communicate on IRC. What were you even talking about? [167:46] - Yeah, we are talking about death threats and- [167:48] - Oh, damn [167:49] - ... but having people thanking you, and sometimes- [167:51] ... they got people who send me a message and, and, "Oh, thank you for VLC." [167:55] And I always answer because I want to [167:59] validate the fact that you need to thank the open source community. [168:03] - Yeah, please, everybody listening to this, celebrate, [168:07] celebrate FFmpeg, celebrate VLC, celebrate all the [168:11] incredible open source projects, Linux, everything. [168:15] There's so many, there's so many... And you know what? I mean, even outside of [168:19] open source, just celebrate companies that [168:22] create software that you use a lot and love. [168:26] - Celebrate human endeavor. Celebrate the human effort to not just build something [168:30] that's okay- ... build something that is damn good. [168:34] - Yes, this is important, right? Like, because as we said, right, we work for [168:38] technol- we do something very complex for [168:42] the normal people. Like, we want our excellence in tech to be useful for everyone. [168:48] And this is why, like, this is why we work, right? This is why I wake up in the [168:52] morning is because I want people to use our stuff- [168:56] ... Because it's making everyone's life easier. [168:58] - Want to solve hard problems. Work on something interesting, work on some interesting [169:02] technical challenges. [169:02] - As we are engineers, we love to build things, right? When I was young, like very early, I knew I [169:06] wanted to build, to be an engineer. I wanted to do cars, right? Maybe at some point I [169:10] will go back to cars, right? But this is like we [169:14] want to build things that are cool and useful. And they need to be [169:18] challenging, right? Because you want your brain to turn on. [169:21] - When did the two of you first fall in love with programming, with building, [169:25] with engineering? [169:27] - When is the first time you programmed, Kieran? [169:29] - Microsoft QBasic. As I was on Windows 3.1 and Windows 95 Microsoft QBasic. [169:34] - Oh, wow. Wow. What'd you build? [169:36] - Uh, like a multiplication, just counting loops like 10, 20, 30, 40. [169:40] - Nice. [169:41] - Then I thought I could do everything after that. I wanted... I jumped from doing that to I want [169:45] to create a soccer, no, a football, soccer video game. [169:48] And I drew all the, I drew everything out. I was like, "I'm gonna do it." And I didn't quite grasp that [169:52] actually, didn't grasp actually it's a massive piece of work to jump from BASIC and [169:57] drawing some pictures to a video game, but there we go. [170:00] - Yeah. I think I did also BASIC and then, uh, Turbo Pascal when I was, [170:08] yeah end of elementary school. But [170:12] mostly the first time I actually did some serious programming was the [170:15] first year of you call that middle school when you're 11? [170:22] Um, I was I lived in Italy for a year in Florence and it was [170:27] amazing year. And like the maths teacher [170:31] told us to, to work in a programming language called Logo, where you had a [170:35] turtle that was designing things- [170:38] ... On the screen, and you would turn left and right. And in the end, we used that to [170:41] do a very complex programming because of course you could do things. [170:45] And, and this changed, like, as I knew I wanted to [170:49] do things with computers and program. [170:51] - I don't think we quite talked about E- H.264 properly. We talked about David. [170:56] Can we return- [170:57] - Sure [170:57] - ... backtrack a little bit to H.264, this thing that powers [171:01] basically all of the video on the internet? So, uh can you tell me the story of [171:08] H.264? And Kieran, you're actually a contributor- [171:11] - Yeah to H.264. So, so H.264 is a video encoder for the H.264 [171:16] video standard. It dominates internet video, but also other areas [171:20] such as Blu-ray discs. And Blu-ray discs are interesting because the people that make them really want the [171:24] highest quality, [171:25] and there's some really cool high-end films that have been encoded broadcasting and all [171:29] sorts of other areas. H.264 was a big step change [171:34] 'cause it kinda happened at the right time as well. A lot of the development took place [171:38] when HD video was coming out. Intel Core 2 and Nehalem [171:42] CPUs were getting fast. You could do real-time video. But the most [171:46] important thing was a key sort of focus on visual [171:51] metrics. So industry and academia for 20 years [171:55] before, was obsessed with mathematical [171:59] metrics or what's known as peak signal-to- noise ratio. So mean squared error, [172:03] logarithm of mean squared error, and that led to tons of issues because mean squared [172:07] error leads to blurring because you actually want to, you want to minimize-- You want to [172:11] add a little bit of error to everything to, to reduce the mean squared error as opposed to having a big [172:15] error, and that led to loads and loads of blurring. So but hobbyists bucked that trend. [172:19] It was for their own personal videos, mostly anime. [172:22] So there were two, there were two things they did differently, and there was a big iterative feedback loop with the [172:26] community. They did some stuff differently. Two, two big things, psychovisual [172:31] rate distortion, so using block energy, trying to [172:35] compensate for human perception when making decisions. [172:38] - So the psychovisual distortion, that's the critical- [172:42] ... thing. That's the thing. I mean, it's kind of revolutionary, like, [172:47] that we can, like, rethink. [172:50] Don't, don't make it, like, this kind of theoretic thing- [172:53] ... of compression. Make it all about- [172:56] - Being pleasing visually to the eye. [172:57] - Yeah, yeah. So compressing in a way that loses the least amount of information [173:02] for the stuff that matters for us humans. [173:04] - Yes, exactly. As opposed to what industry-- Some parts of industry are still [173:08] obsessed by this, which is mathematical numbers that don't look [173:12] good in reality. And then adaptive quantization was the other big one [173:15] where it was biasing bits against, [173:19] complex areas and redistributing them to less complex areas like [173:23] grass. Grass has some high frequencies, but it's kind of-- [173:27] it's less complex overall compared to more complicated things. And this came around by [173:32] ParkJoy. So ParkJoy was really the canonical sample that [173:35] was... Is the running around in the park. [173:37] - This one. [173:38] - Yeah. So this guy was really the-- [173:42] So this was created by Swedish television in the beginning of [173:46] HD, and it was done on film, and it was no expense spared in terms [173:50] of production quality, and it was given away for free. This was really-- [173:54] And this is the sample really that sorts the men from the boys in terms of it has [173:58] so many challenges with the trees, with the water, with the [174:02] grass, with the motion, with the... I don't think there's still [174:06] been any public test sequence as good as that these days. [174:11] - So for people who are just listening, we're looking at a bunch of humans running- [174:16] ... along a river, as you have the reflection, a lot of really high [174:21] information textures everywhere, the leaves and the lighting playing with [174:25] the leaves and all of this. [174:27] - You could show clearly that encoders with high PSNR- [174:30] - Will blur everything [174:31] - ... will blur everything, and you could see actually I could turn on psychovisual stuff, I could turn on [174:35] adaptive quantization, and [174:37] it would just look so much better. But your metrics-- And these metrics are at [174:41] the time were considered so holy. [174:44] These are the holy metrics that are untouchable. PSNR is the most important thing. [174:48] - Uh, can you speak to how do you measure psychovisual stuff? [174:52] Like, how do you turn how pleasing a compression is for a human eye- [174:57] ... into a number? Is that even possible? [174:59] - That's what, that's what Netflix have been trying to do with VMAF. They said they've used a machine [175:03] learning model. [175:04] - That's a more recent thing. But back in when x264 was being [175:08] developed, that's by eye you're basically- [175:10] - It was by eye. It was developers on their laptops. So it's not like even with [175:14] big companies with professional screens or anything, it's- [175:17] And that was actually one of the goals, which was I don't-- The developers at the time, [175:20] Loren Merritt in particular, said, "I don't wanna test this on a thirty thousand dollar screen. [175:24] It's-- I want this to look good on someone's laptop at home." [175:27] - Yeah. Brilliant. [175:28] - Um, there is another sample which is-- [175:31] ... A sample that is Planet Earth's killer sample that I absolutely love. [175:37] And you are going to see why, right? [175:39] - Yeah, you're going to love this. [175:39] - Uh, it's a ton of birds, right, flying, and the more it goes, the more there are [175:47] birds, and at the end, right, it's almost like [175:51] you have millions of birds. It's the most complex thing [175:55] ever to encode, right? And well, you're watching it on YouTube, and you see how [175:59] bad the YouTube encoding is actually, right? Um, [176:03] and this is, like, phenomenal to, to optimize and [176:07] get perfect quality in a constant bit rate. [176:11] There was a lot of optimization, mostly by Loren also, um- [176:15] ... on anime, right? For a long time, anime was very badly [176:19] encoded because there was a ton of banding, right? And so you see those [176:23] those issues, and there was a ton of things. So [176:27] x264 is, like-- And today it's still the reference [176:31] to any encoder, new encoder, AV1, AV2, VVC, HEVC, everyone compares to x264. [176:38] - One of my favorite films, Cinema Paradiso, I know the engineer [176:42] who created the Blu-ray, and he showed me the comparisons of x264 versus [176:46] others, and the... it's completely different. And I think [176:50] a bunch of guys in the Blu-ray world started using x264. Um, I think the big [176:54] one was Chris Henderson from Warner Brothers. He did the whole Fringe box set with that. [176:58] So quite, like, a thing a person on the street actually watches and wants to look good. [177:02] And so they kind of took a risk in their jobs doing that because they're in a big [177:05] company. That big company can buy whatever they want. And they said, "No, no, no, I want to use this [177:09] free and open source thing so that things look good for my, my [177:13] customers and build the best." And to this day, I personally still try and [177:17] avoid watching the most cinematic films on streaming [177:20] services and buy the physical discs because they look, [177:24] they look good without even having to buy an expensive TV. I think that's the key thing. [177:28] - And x264 is yet another example of open source project. It was started by [177:33] Laurent Ehrsam when he was at École Centrale Paris, where VLC was born. [177:37] And then you got a generation of people like Loren, like Jason, like, [177:41] uh, Måns, like so many- [177:43] - Henrik from- [177:44] - Henrik, uh- [177:45] - Anton, uh [177:45] - ... and this is-- Anton, and this is where the assembly [177:49] thing that we use now on FFmpeg, dav1d and so on, was born, right? So [177:53] x264 is, like, amazing project with people who were really all over the [177:57] world, and I think most of them never met each other. [177:59] - But all of them, according to Kieran, or large percentage, love [178:03] anime. There's several things I've never got into, and one of them [178:07] is anime, and I need to [178:09] - I watch anime so much, especially at the time. Like, at the time it was like [178:17] a lot of anime content doesn't exist commercially, right? We [178:21] are before Crunchyroll, right? So what happens is usually people who [178:25] love anime, who take some things, some DVDs in [178:28] Japan and rip them because there is no commercial offering. And- [178:33] ... some of the people who are, what we call fan subbers, are basically [178:37] translating themselves to make subtitles, right? And at that time, [178:41] you download completely illegally. It was the only way to do that, right? [178:45] And so all of that was handcrafted, and it fits the open [178:49] source community, right? Because they needed tools to encode, to do fan subbing, [178:53] right? One of the most amazing open source projects for subtitles is called [178:57] Aegisub, and it's a subtitle... It's [179:01] done for anime, for, for South Asian and Japanese languages. [179:06] - There are weird textures in anime that I don't think you get [179:10] in real life content. I think that was a key one, which was optimizing these weird [179:14] textures that you get- ... because anime is not done in a normal fashion. [179:17] - Yeah. The way you produce it is not-- You, you mostly produce it, like, [179:21] on screens, right? Since a bit of time, and you have all those gradients, [179:26] right? In colors, because they are very easy to produce digitally, very complex [179:30] to produce in real life. And the [179:33] subtitles also are very complex because you need to have often the [179:37] Japanese and then you need to have the diacritics, right? [179:41] The what we call the rubi, right? Which is the hiragana and the katakana for the [179:45] kanji. And then because of course you, so that you have the official [179:49] subtitling, but you also need the English subtitles or the French subtitles because you [179:53] want to learn that, right? And there is so many things crazy on, on [179:57] subtitles and we had like crazy samples on, on subtitles that [180:00] we've seen all around. So this is an important [180:04] part of the, the culture, but also because there was no [180:08] official offering. There was no way of doing that. [180:11] - Uh, can you speak to the difference between H.264 and AV1 and then [180:16] x264 and dav1d? This is this big [180:20] step. Can you help people understand, are some of the streaming sites [180:24] moving more towards that direction of AV1? [180:26] - Let's be honest, all of those codecs since MPEG-2 [180:32] video are the same concepts. The same concept about inverse transform, about intra [180:39] prediction, motion compensation, entropy coding, all of them. [180:44] However, each generation gives you a bump between twenty-five and fifty percent [180:51] more compression for the same quality. [180:54] And so you had the MPEG-2, you had the DivX era, you have H.264, which was, like, [181:02] changing, right? H.264 improved so much. And then you had more, right? [181:06] You had HEVC, you had VP9 at the same time of HEVC. VP9 is a bit similar to HEVC in [181:14] terms of quality compression, but it's royalty-free. [181:17] Because in multimedia there is ton of patents and the [181:21] licensing after H.264 became out of hand, right? [181:24] And could cost hundreds of millions of dollars per year. So it made no [181:28] sense. So Google did this VP9 and the Alliance for [181:32] Open Media did this new codec called AV1. So you can imagine that AV1 saves [181:39] between forty and sixty percent less bandwidth than H.264- [181:45] ... for the same quality, visual quality. [181:48] - At a given bitrate. [181:49] - At a given bitrate, right? So that's really like you increase [181:53] the quali- either you set the bitrate and you increase the quality, or you set the [181:56] quality and you decrease your bitrate. But because now you move from, from [182:01] SD to HD and HD to 4K and 4K to 4K HDR, like you increasing the size by like two, [182:09] factor two, three, four, right? So you need to have better compression to [182:13] keep it in terms of something that is manageable. [182:16] - It's more coding tools, bigger blocks, lots more [182:20] sub-partitions in each block. It's just exponentially more complex. [182:23] - It's more complex because the encoder needs to search [182:27] more possibilities, right? So you, for example [182:31] one of the things that is easy to understand is to predict a block, [182:35] a color block to another, you have directions, right? So you can go [182:39] left, right, bottom, up, and then in terms of [182:43] Like the other quadrants, right? What I call north, [182:46] northeast, northwest, and so on, right? But that's eight directions. Then you can do [182:50] more direction. You can do sixteen or sixty-nine or one hundred and [182:54] twenty-eight, right? You can-- And every time your encoder is going to spend [182:58] more time to see, oh, well, this block is exactly this one [183:02] and those type of tools that you can bring, and the encoder needs to [183:06] check which of the tools are going to compress you better. And so [183:11] I guess that AV1 encoding is two order of [183:15] magnitude more than H.264 in terms of CPU cycle, right? Order of magnitude, right? [183:21] - Yeah. And as we discussed, CPUs are not getting faster. You're just throwing more cores at the [183:25] problem. [183:25] - But also it's a fact that you encode once and you have hundreds of millions [183:29] of users, right? [183:30] So for example, YouTube, a very good example. YouTube encodes almost [183:34] everything in H.264, but the popular video [183:38] gets re-encoded in AV1 because it costs more, of [183:42] course, to encode, but you encode once and you send that to millions, right? [183:46] So it's a trade-off between encoding time and complexity- [183:50] ... and CPU usage on the server side and on the client side. [183:54] Because at the end, if you're distributing a video to hundreds of thousands of [183:58] people and the size is half of the other, then [184:02] it's better. It's better for your battery, it's better for your modem, et cetera, et cetera. [184:07] - So we can lay out, let's say, the top five codec container combos [184:14] would be H.264 inside MP4 containers, AV1 inside MP4 WebM containers, [184:23] ProRes for nonlinear editing, [184:27] Inside MOV containers. So for people who don't know, I guess ProRes is [184:32] - It's Apple's codec for editing, originally for Final Cut Pro, and [184:36] it's designed to be fast to decode, fast to seek, because an editor will need to move [184:40] very quickly. So it's a different use case to the distribution element. [184:45] - There's no or very minimal temporal compression in the- [184:48] - There's none, yeah. There's none in ProRes. So you can cut, so you can do cuts. [184:52] - This is what we call intra-only codecs, right? So [184:56] I'm going to explain quickly what is IPB frames. [184:59] - Yes, please. [185:00] - Um, so I-frames, often key frames, but [185:06] is complete frames. It's like an image. It's a JPEG, right? You have... You can [185:10] start, you see everything, right? And then [185:14] you, the next image can be a P frame, which is a predicted frame. So you take [185:20] some part of the previous image saying, "Well, I need the block five and seven [185:24] and forty-two," [185:25] and you replace it, and then you just give the extra information, right? [185:29] But that means that in order to decode this P frame, you need to have access to a [185:33] previous I frame, right? [185:35] And then, of course, you have more complex one, which are B frames, [185:39] which are B-predicted frames, which can depend on [185:44] different type of frames, some in the past, some in the future. [185:49] And so ProRes is an intra-only codec. For the people who can see, this is- [185:54] - Yeah, that's a good one [185:54] - ... a very good one, right? So I-frames are complete frames. Um [185:58] P-frame basically depend only on I-frame, and B-frames can depend on in front. [186:04] - And this GOP, group of pictures, I think the default for actually FFmpeg [186:11] for H.264 is like two hundred and fifty frames, something like this. [186:19] - Yes. [186:19] - And to me, it's just, it's like magic, that you could predict [186:23] that you could have a complete frame every- [186:26] - Several seconds, that means [186:28] - ... several seconds, and then you could still, you could have this chain of [186:31] predictions you make, and the fact that you can-- The fact that somebody like me [186:36] can can use FFmpeg to compress something and not notice [186:39] that the result still plays back smoothly is like magic. [186:43] - You can even have, and we use that in tons on Kyber, is what we call intra-refresh, [186:50] where basically it's there is no I-frames present. [186:54] - You have no I... You have one at the beginning. [186:56] And you never send an I-frame. You get a- [186:57] - How does that work? What is it? [186:58] - You build up an I-frame gradually across as the stream continues, so- [187:02] - Ah. So you refresh certain parts- ... of the image. [187:05] - But so you never have an I-frame. Like this is intra-refresh that we use, right? [187:09] - That's even smarter. [187:10] - But for me, for me the biggest mind-blown when I started was the B-frames. [187:15] B-frames means B-predicted frames can depend [187:19] on frames that are coming in the future. [187:23] That means that in order to decode this B-frame, you need to wait for the next [187:28] frame that is dependent- [187:31] - Yeah [187:31] - ... buffer that, decode that one, so that you can decode the B-frame, [187:35] right? So the way you decode the frame, the [187:39] decoding order is not the same as the display order. [187:42] Right? That means the encoder needs to be very clever and decide [187:46] that, "Well, you know, I'm going to depend on things like in the future." [187:50] So this is like- [187:51] - It's incredible [187:51] - ... mind-blowing. [187:53] - The fact it works so smoothly every day is kind of miraculous in some ways. It, [187:57] it works so... You can have a stream that works [188:01] across the world on their decoder versus one in the US versus one here of [188:05] different manufacturers, and they produce bit for bit exactly the same material. [188:10] That's quite remarkable, and do quite complex things, and getting more and more complex [188:14] and still be bit exact. There's a lot of work that goes into that. [188:18] - There's a lot of knobs you can control in this whole process. There's a lot of really fascinating [188:22] parameters that I've gotten to know more and more over the years that [188:26] FFmpeg gives you complete access to. Maybe you could speak to some of them. So first of [188:30] all, like obviously, we can lower the resolution, we can lower the frame rate, [188:34] we can use different kinds of codecs, as we mentioned, from H.264 to AV1. [188:39] There's ways to tune the trade-off between bitrate and [188:43] quality, as we've kind of spoken to. You know, you could do constant bitrate, [188:47] you can do constant quality, say RCQ, QP. You can do the longer or shorter group of [188:55] pictures, GOP, that we mentioned. I mean, all that kind of stuff. It's crazy. [188:59] Number of B-frames. [189:00] - Yeah. What is crazy is that a ton of people's job [189:06] is to optimize those parameters, right? [189:09] A ton of people that you see at YouTube, at Netflix, at Meta, and so on, [189:13] they're not writing codecs. They're just like finding the right [189:17] parameters for the file they have, for the format they [189:21] have, right? Because like something that is for a movie or something that [189:25] is user-generated content from your phone or a screen recording [189:29] or something that you're going to video edit, you don't want the same things. [189:33] And there are thousands of people whose job is just to optimize all that. [189:38] - Yeah. They're wizards. Hats off to them. [189:42] Uh, YouTube like to deliver, all the streaming sites actually, to deliver [189:46] at scale. And like YouTube is really magical because it's [189:50] not just doing like what Netflix does, which is one way [189:53] broadcasting type thing. It also has to upload videos from [190:01] all the places. So they're also doing encoding at scale- [190:04] ... for videos that are gonna be watched by like five people. [190:07] And it still has to deliver them re- like on a moment's [190:10] notice. No, no delay, nothing. No... I mean, very minimal latency. [190:15] And also serve it in all [190:19] different resolutions. Like YouTube is basically the web version of VLC. [190:25] - Yeah. Well, actually, it's funny because, like Google Video, which [190:29] was something they did before they acquired YouTube [190:33] was actually using the VLC plugin so that you could run VLC [190:37] inside the web browser using the ActiveX [190:40] plugin. And so it worked in Internet Explorer, [190:44] and you were actually running VLC inside your browser. [190:49] Which is funny because today we have the opposite, where we have VLC WebAssembly, where [190:53] we compile all VLC and FFmpeg to decode, to run VLC in [190:57] inside the JavaScript virtual machine with WebAssembly. [191:04] - Okay, there's this legendary story that you pointed me to [191:09] that it was discovered via [191:12] WikiLeaks release of all seven documents. The CIA was using a modified [191:16] version of VLC to basically try and trick people, what? To steal their data? [191:23] - Yes, exactly. [191:23] - So can you explain what the heck happened? What... [191:27] - So, so this was a surprise, right? Because at some point, WikiLeaks, [191:31] uh mentioned some documents. There were a few ones with something related [191:35] to Blu-rays and VLC, but the, the most interesting one was the [191:39] CIA Vault 7, which, if I understand correctly, [191:43] Was the CIA had, like, a custom version of [191:46] VLC where they had a specific plugin. Yeah, [191:50] exactly. This is-- Like, we had to, to write a press release on that. [191:53] - Uh, VideoLAN wrote a press release saying the only safe source for getting [191:57] VLC media player is the official VideoLAN website. I mean, I [192:01] suppose that's a security vulnerability for [192:04] basically any piece of open source software. Somebody can trick you. [192:09] - To download in a fake website- [192:12] ... or targeted advertisement, right? That was a targeted advertisement, to watch a [192:16] specific file you need to watch with this custom [192:20] version of VLC. And it was the normal binaries of VLC, except they [192:23] added one DLL, I think it was psapi.dll- ... which was basically reading your, [192:31] your document folder, encrypting that, and sending that. And [192:35] the thing is, this is very clever, to be honest because once you're [192:39] watching a movie, right, you're going to do that for two hours, and you're not going to touch your [192:43] computer. And sometimes it's normal because it's HD that your, your [192:47] fans are going up and say, "Vroom," and there is ton of CPU usage because you're using [192:50] VLC, right? That's normal. But the thing is, what you don't see is that actually [192:54] a powered version of VLC that is used by CIA. We had exactly the same problem [193:02] with Chinese hackers that were targeting [193:06] Indian people, and that got VLC banned from India [193:10] until I had to, to fight in courts in India, the [193:14] Indian government, to unban VLC. They didn't use VLC. [193:18] They took just one DLL, because we signed the DLL correctly, [193:22] And they used that DLL to do another program. [193:26] So you had the vlc.exe and was calling libVLC, but it was [193:30] calling it into a fake one. And they used that to target. Um, [193:35] there is not much we can do actually to block those type of hacks. [193:39] - Yeah, and I think people should, for all open source software, for all [193:43] software in general, people should pay attention where they download the thing. [193:46] - Yes, because that means that they were not downloading it from our website. [193:50] - Do the search engines help you? [193:52] - No, they don't. [193:53] - Just to clarify, 'cause you can, you know, to prevent threats from [193:57] people manipulating SEO to get up there on the links and try to- [194:00] - Absolutely not, right? We have a big issue for, like, more than ten years, [194:04] is that there is a fake version of VLC in [194:07] Germany that was reported for now for 12 years, [194:12] and Google basically decides to not-- They know what's in it, [194:16] but the binary is too big for their virus analyzer to [194:19] analyze it. And so while if you're in Germany, you can go to a [194:23] website that is a fake version of VLC with a custom installer, [194:27] and it's very popular in Germany because their website is in German, and Google [194:32] mentioned that before VideoLAN. And the weirdest thing is that it doesn't do [194:36] anything on your machine for three weeks. [194:38] Because that's how they do the detection. And after three weeks, there is a small [194:42] program that is a service that install at the same time that wakes up after three weeks, [194:46] and it start downloading spyware and adware. And Google knows about it. [194:50] They've decided not to do anything. The guys use dark SEO in [194:54] Germany to do that at some point. And this is very damaging, right? [195:01] Because one of the things that they are downloading is actually something that is replacing [195:05] your ads inside your machine, right? [195:08] - It's actually quite surprisingly effective. [195:11] Whoever is doing it with Twitter and X. With X, [195:16] I'll get emails about, "Your X account has been hacked." And [195:19] however they phrase it, it gets me to, like, at least [195:23] click on the email, not to follow the thing, and then you're like, [195:27] "Man, whatever they're doing with the psychology to try to trick you, [195:31] they're quite good." [195:32] - There is a security v-version of VLC, right? You received an email saying, "Hey, there [195:36] is a security version update on VLC. Think about updating right now because- [195:40] ... it can hack your computer." You come. It's a website that looks decent, and, uh- [195:45] ... and you download. It's a new version of VLC. Great. You don't know. A month later, you're hacked. [195:49] You have no idea. You're part of a botnet. [195:51] - Yeah. So make sure wherever you're downloading stuff, it's legitimate. [195:56] I'm part of the botnet. Speaking of which, so you've mentioned [196:00] that VLC sandboxing is some- is something you're working on, [196:05] and it's actually something quite challenging. Why is it important? Why is it hard? [196:09] - So VLC is a core with around [196:13] 500 plugins, right? One of them is FFmpeg, but we have, we [196:17] support so many other formats. We support new [196:20] protocols, we support new filters, we support weird architectures. [196:24] And in this release of VLC, you have [196:28] modules that are going to call your drivers, right? [196:31] Uh, mostly the hardware decoders, which are going to call [196:35] your Intel, your NVIDIA, your AMD driver. And all calling FFmpeg, right? And [196:43] there might be a security issue. There might be a security issue in the shader, [196:47] there might be a security issue in VLC, in FFmpeg that is going to [196:51] basically crash. The issue is that you running [196:54] VLC like every, every other program, like Adobe, right? You're running it [197:00] on your machine, and it has access to all your documents, right? So [197:04] the idea is to be sure that you do a sandbox so that we [197:07] can protect from ourselves, because inside the [197:11] VLC process is running some code that is not even ours. Either it's open [197:15] source for other projects that we integrate in VLC, or it's your [197:19] GPU driver or something that is provided by someone else [197:23] inside. And so when we crash, we want to not [197:26] allow people to do bad things, right? Because one of the common way of hacking [197:30] people is to crash a program, very often done with a [197:34] web browser, very often done with PDF files, [197:37] less often with multimedia, but that could happen. And when you crash, you launch something [197:41] on the machine of the person. Could be a ransomware, could [197:45] be a botnet, right? So security of desktop application is important. [197:50] On mobile, it's a bit different because most of the mobile application are running on [197:54] inside their own sandbox. But for VLC, we [197:58] could run it inside one sandbox, but the problem is that we need access to [198:02] so many things that it's basically we would do-- we [198:06] would have all the permissions, right? And so if you have a sandbox [198:10] and you put some holes everywhere, it defeats the purpose, right? [198:14] So what we are trying to do, and we're actually doing, is splitting [198:17] VLC into several processes. One is decoding, one [198:21] is demuxing, one is filters, and all of them run into their own sandbox so that the [198:29] whole VLC, a part of VLC crash, like Chrome crashes [198:33] on some tab, right? It crashes, but it does [198:37] not crash the whole program. And this is what we're trying to do. And it's [198:41] difficult because it's a sandbox that needs to sustain gigabits per second- [198:45] ... of mem copies. Now, it's not a website which is five megabytes or [198:49] 10 megabytes. We're talking about hundreds of megabits per second. So this is [198:53] why it is quite challenging. And this is a research topic that [198:57] we are working on in order to have multimedia player that is secure. [199:03] - This is all the kind of stuff you have to think about when millions of people are using. [199:06] You've mentioned something somewhere where like all the different [199:10] features of VLC, when you have that many people using it, somebody will use [199:16] every single feature, and they will tell you about it. [199:20] - Best feature in VLC is called the puzzle filter, right? [199:24] So you click the puzzle filter, and it transforms your [199:28] video into a jigsaw puzzle, right? [199:30] - Nice. [199:30] - And you can click and move the pieces, right? [199:34] It's very, very useful when you're watching a French movie, right? You're bored, [199:39] ... because it's like very long things or a love triangle, right? We've seen [199:43] that so many times, right? But, but you need to watch it because someone, your wife [199:47] or-- ... told you to do that- [199:49] - To catch up [199:49] - ... or your boyfriend told you to do that. So you're doing that, right? [199:52] And you can click and move the pieces around. [199:55] It's absolutely useless, right? Like, who cares about that? [199:59] First, it was done by a math teacher in high school [200:04] in south of France to teach his students about Bézier curves, which [200:08] is something that- [200:08] ... everyone should know about, right? It's very useful. But the code was clean, [200:12] so it got in VLC. It was merged in 2010. Five years later, I [200:16] receive an email saying, "Hello, JB. I have a problem with VLC. [200:20] The puzzle is too simple." And I was just like, "What?" [200:24] And yes, the puzzle was in the UI maximums by 16 [200:29] by 16, right? Only 256 pieces. And he says, "I'm [200:33] sorry, but in a movie I love puzzled, this is too simple," right? [200:37] So there is a commit of me, you can check it online which is JB changing [200:41] that the dimensions are 256 by 256. [200:45] - Right. [200:45] - But my point is, so many unused features [200:50] are used by a few people, right? There is a way to watch VLC [200:54] movies in command line without any UI, right? It's- [200:57] - I saw that. You can do ASCII. [200:59] - ASCII art. Is it useful? Very useful. Imagine you're [201:03] debugging... imagine you're debugging a multicast network, right? [201:07] You have thousands very complex, [201:09] very complex networking stack, right? You can SSH to all of the [201:13] routers and put VLC on it with no UI, and you're going to see [201:17] whether it's black or it's not black, right? So you see if or it's all green or not [201:21] all green, right? So you can see- [201:22] - Amazing. [201:23] - Yeah, right. [201:23] - This is fun. [201:24] - People don't realize there is so many things in VLC, that are useful and they are-- [201:32] they have users, because once you have hundreds of millions of users, [201:36] you have people who use every feature. [201:40] - I would love to sort of [201:42] zoom in and talk a little bit more about the distinction between kind of [201:47] downloading a file and watching it offline versus [201:50] streaming. So the complexities, the challenges of streaming. [201:55] Is there something we could say about what it takes to, [201:58] stream files? Because we've been talking about codecs- [202:01] ... and I think a lot of that implies encoding and decoding without the [202:09] having to communicate- ... over the network. [202:11] - Sure. [202:11] - Sure. So can you elaborate, like, what's required to do over network stuff? [202:16] - Yeah, but it is less complex than it seems compared to everything that we've [202:20] talked about. Especially because the most complex thing [202:24] is not about streaming in terms of, uh, [202:28] streaming services, but it was what was done to [202:32] actually broadcast through satellites. Because in, in [202:37] most of the modern, uh, broadcasting services, you can [202:40] pause and you can go on. But when you're sending live streaming, whether it's [202:44] broadcast or live for streaming services which are live, this is [202:48] much more difficult because you need to encode in real time. You-- [202:52] When you go on a satellite, you have a specific size of the link, right? You [202:56] cannot have a burst- [202:58] ... of bandwidth even for a second, right? Because you don't have the [203:01] space for that in your total file. However, there is [203:05] different types of challenges, which are interesting challenges, but [203:09] I think they are less complex than the one we've seen with late [203:13] '90s and early 2000s about broadcasting and streaming through satellite. [203:18] - They're different. They are control systems challenges, whereas, whereas some are more mathematical. [203:22] I think that's the difference. [203:23] - In the streaming world, what you have is called, what we call adaptive streaming, because the [203:27] difficulty-- And it's not really a video problem, it's mostly a CDN [203:31] problem, is that you might have too many people watching the same thing at the same time, [203:34] and it's a congestion of the network, right? So your [203:38] player has difficulty downloading things fast enough to play [203:42] them. So what happens is that locally, the player is going to read a lower [203:49] resolution- [203:50] ... of it. But there are some very clever algorithms to do that, but most of it [203:57] is quite basic, to be honest. [203:59] - Even on the buffering side, it's pretty basic. [204:02] - Yeah, you start to download a segment, what we call a segment, [204:06] and then you time, right? And if it takes more than 50% [204:10] of the time to download a segment, you go down to... Right. And the difficulty [204:14] is more about when do you go up in bandwidth, in quality. [204:18] But this is not very complex to do. When you encode, you're going to [204:22] encode seven resolutions, right? And, and you're going to give the [204:26] bitrate. The difficulty is to have your encoder gives the same bitrate, [204:30] but it's not as strict as it, it used to be. So- [204:34] - Uh, probably YouTube has to figure out how the human [204:38] psychology side of that, like how pissed off do you get [204:43] when it's like very low bitrate and, [204:48] How long should it wait before it increases the bitrate even though the connection [204:52] is better? Because maybe the, the changes in the bitrate is what, like, [204:57] affects you psychologically. [204:59] - No, I think actually the interesting one is the audio. [205:01] - That's true. [205:01] - The-- You can kind of notice when they move from [205:05] full fat AAC to the there are compressed versions of [205:09] AAC that use Spectral Band Replication. You can kind of see it goes a bit tinny, and [205:13] that up and down is very jarring. The video side [205:16] is a lot smoother, and there's less notice. It's really the audio you can definitely, you can [205:20] definitely feel it from when it's moved you from a different audio profile to one or the other. [205:24] I don't know. We're surprisingly tolerant at skipping audio glitches. I, [205:28] I'm surprised people I know who are not video engineers, how [205:32] tolerant they are, how tolerant they are to watching sports at [205:36] 30 FPS, for example, whereas it should really be 60. [205:39] The world is a lot more tolerant to that, but audio people are very-- There's-- It's an immediate [205:43] feedback mechanism of, "Oh, something's changed." [205:44] - If you hear a glitch, you realize it directly. [205:48] I get to fully realize that, I suppose. One of the things I'm afraid of when I listen to [205:52] audio more and more, that I get to notice every single tiny detail, [205:56] and that you can over-obsess when people, people in general are able to kinda, [206:02] kinda blur their consumption. They can, they can look past certain imperfections. [206:08] - But then when you combine like [206:11] an event that is, for example, a sport event that is probably going through [206:16] satellite or- [206:17] ... somewhere else and goes to a central place for encoding, and then you [206:21] need to encode this older resolution in real time. You don't have [206:25] time for QA. You need to push that to CDNs. You need to add [206:29] probably DRM for protection. You need to have that [206:32] over a ton of different devices. Then yes, it is complex. But-- [206:40] And also, like you're in the web browser or in very much different [206:44] devices that you use for television, where you had like a, [206:47] defined set-top box or cable box that, that you know where you [206:51] control end-to-end. So it's a challenge, but it's less... [206:55] I think the networking part while you [206:59] agree to have 10, 20 seconds of latency, I don't think this is very difficult. [207:05] - Speaking of networking and latency, so your new [207:09] effort, as we mentioned, is Kyber, which is aimed at ultra-low latency. As you say, [207:17] every millisecond counts, and you're applying that to [207:21] remote control machines like robots, drones, computers. Can you tell me about it? [207:25] - Sure. If you start from where we used to be, right? You used [207:29] to use FFmpeg to encode files, right? And then we used FFmpeg and VLC to encode in [207:37] streaming services, right? And then you need to go [207:41] lower and lower. And the question was where up to where can we go? [207:46] And this question is very important because there are many use cases [207:49] where you need to be fast, and it's when you have [207:53] feedback interaction, right? We are not just listening to something, you're actually [207:57] controlling it, right? Because-- And that's the biggest difference that compared to [208:01] what we've done so far, is that I need video [208:05] to have a feedback on something that is happening live, whether it's a drone flying, [208:10] whether it's controlling a humanoid robot from [208:13] distance, whether it's controlling a hover, whether it's playing a [208:17] video game in the cloud gaming, because this is, [208:20] What I did on a previous job, right? I was CTO of a [208:24] cloud gaming startup. And this is a very interesting [208:28] topic because you push to the limit the network. [208:32] You need to be-- to care not about the [208:36] quality like we've done on video, and we've talked about with H.264. You care about [208:42] latency, because a milliseconds is [208:45] meaningful when you're controlling a car, right? For-- Well, you've-- you've [208:49] seen, you've used Waymos, right? When Waymos don't work, and [208:53] that happens even if one percent of the time, there is someone that is basically [208:57] remote controlling that. And this is exactly [209:01] the stuff that we're building. It's really an SDK platform to do end-to-end [209:09] control of machines. [209:10] - So the-- this comes up quite a lot in a lot of different contexts in robotics. [209:14] So obviously, teleoperation, teleop is becoming more and more [209:17] important including for training, Robots via machine learning. [209:25] - Yes. And what we do is a bit different from everyone else, is that [209:30] we take only one socket, one connection, which is a QUIC [209:33] protocol based on UDP which is interesting because [209:37] it's done for low latency. It doesn't have two of the, what we call the TCP head of line [209:41] problem and the HTTP head of line problem. It's ciphered by default, but on [209:45] the same wire, we send multiple streams, like multiple tracks. We send [209:49] audio, we send video, but we also send the commands, right? Uh, [209:53] mouse, keyboard, gamepad, and so on. And we do [209:57] that while maintaining coherence, right? Synchronization. Because [210:01] what people don't realize is that all the clocks [210:05] actually drift. And when you're controlling a robot, a robot is [210:09] going to have, like, two cameras, five cameras, ten cameras, a ton of captors, [210:13] GPS, and so on. And if you want to train correctly your [210:16] robotic AI model, you need to have all those that are in [210:20] sync and current. And what we've done, and it's all the stuff that we learn [210:24] on VLC in broadcast in real time, and MPEG-TS that Kiran's know [210:28] well, is that we account for clock drifting. And so when I record a Kyber stream, [210:36] a robot, I am sure that it's going to be predictive in the way you [210:40] play it back. And so when you're going to do recording and training [210:44] of your AI model, you need to be sure that every time you [210:47] retrain based on the data, it-- the data is going to stay [210:50] coherent. And clocks actually drift. Like, [210:54] the existing solution works with one camera. Once you're going to a five or [210:58] six, it's more complex. [211:00] - Uh, so you wanna make sure that the visual [211:04] snapshot perfectly matches the time it actually happened. [211:08] - Exactly. And also, if you're going to control, right, I do something on robot, I [211:12] need to be sure that it is actually happening at that precise time, [211:16] right? And so we have on the, the server, which would be a robot, a [211:20] time of, like, re-timestamping mechanism accounting for clock drift for [211:24] that, right? So that's one of the use cases of Kyber to, to [211:27] control robots. Um, I also think, like, remote [211:31] drones, remote whether it's defense or non-defense, remote [211:35] cars, remote submarines. There is many places in industry or remote surgery where [211:43] the expert cannot go everywhere the machine is because it's either [211:47] dangerous or it's too costly, right? So you, you allow people to have machines [211:53] next to you, right? The goal of Kyber is to make distance disappear [211:57] because it's either projection of skills or projection of power, [212:01] right? So imagine we are all like— you've seen the Meta [212:05] Ray-Ban and, and everyone else, right? You need to stream there, right? Because you're not [212:09] going to run anything over there, right? So you need GPU power whether it's on a cloud, on a [212:13] phone to stream that. And so all of these use cases needs to [212:16] be not about extremely low latency, but real-time latency for [212:20] video. And so that means you need-- [212:24] we're toying with the encoders so that the encoders encode a frame in [212:28] four milliseconds. And, and, and Kieran with his company also goes [212:32] under those type of latency, because you need to optimize [212:38] at max the local latency, right? Because it's the decoder, the [212:44] encoder and so on. Because this time is going to be added to your networking time. [212:52] So-- And it's not just about low latency, it's also about, like, reliability. We do [212:57] clever things like forward error correction, right? So forward error [213:01] correction is you over-transmit a bit of data, right, a [213:05] few percent and while over-transmit, you're allowed to [213:09] lose some packets. Because all of that is very difficult [213:13] over an internet network where you're going to do things very far away. And [213:20] if you check that all packets are delivered, you add a ton of latency. If you [213:24] don't want latency, what we do is that we over-transmit [213:28] some data that you can retrans-- reconstruct on the client [213:31] side when there is things that are broken, right? So [213:35] And we a few, a few days, weeks ago, we were doing the [213:39] demo around Las Vegas for the CES about we had a, [213:43] a, a rover that is fully 3D printed. It's very simple. It's a car, right? It's a [213:47] small car with a telescopic arm, and it [213:51] was actually controlled from France, right? And the, the video [213:55] was with a webcam and a very small server, right? A small, a [213:59] small PCB was basically running and send that to [214:03] someone that is on the other side of the planet. [214:06] And so there is so many use cases. You can also think about [214:09] having AI who are going to control many drones and so on. And [214:13] technically, we need to be amazing in video, we need to be amazing [214:17] at networking, we need to care about any milliseconds in [214:21] networking, in encoding time, in decoding time, and also you need to [214:25] integrate very low level. [214:26] - So sync everything together well. But how-- Like, what kind of latency [214:30] can you get to? Like, why-- When you say milliseconds, what, what's the goal? [214:34] - So my goal is four milliseconds glass-to-glass latency. [214:38] - What's glass-to-glass mean? [214:40] - So it's easy, right? You have a computer which is running a program, right? Probably a [214:43] video game, and this one is actually running, right? It could be-- [214:47] it's an example of a robot, right? And you have the replicate that is- [214:53] ... done through the network. And, and you want, [214:56] if you take a, a one thousand hertz camera, you can take a [215:00] picture, and you want that to be at four milliseconds. Four milliseconds means two [215:03] hundred and forty hertz, right? [215:06] - Yes. Nuts. [215:07] - So far we achieve seven milliseconds [215:11] from a Windows to Windows or Windows to Mac. And [215:15] if you look in the timing, most there is around three point [215:19] five milliseconds inside the NVIDIA [215:23] hardware encoder and around two milliseconds on the Intel [215:27] decoder, right? So, like, the encoder plus the decoder is already six [215:31] milliseconds, right? So in order to go down, we need either to [215:34] have some other type of codecs, or some better encoder that are faster. [215:41] But four milliseconds is, would be the grail. [215:44] - That's pretty nuts. I love it, though. I don't think anyone's ever achieved that, [215:48] right? That's fast. [215:49] - You can achieve that with custom hardware- [215:53] ... with SDI, with professional hardware. [215:56] But I want that to work over the internet. I want [215:59] that to work with any robots where you're going to have a small Jetson [216:03] Nano in it or a N150, right? I want [216:07] that because there is going to be millions of robots or- [216:11] ... drones are just rolling robots or flying robots or, or swimming robots, [216:15] right? It's just you, a machine that you control. And in order... [216:20] Either you need to teleoperate them [216:23] or when everything will be fully autonomous, you need to [216:26] teleobserve them, right? You need to check what's happening. [216:29] And in my view, in the future, like, all those remote [216:33] cars will be teleobserved by an AI [216:37] model, which is just going to say, "Well, everything is good." And when it's not good, [216:41] say, "Hey, there is a problem," and then you have an operator, right? And this is [216:45] going to be about safety, right? When you have your humanoid taking care of your grandma or my [216:49] grandma, [216:50] I want to be sure that everything goes well, and I'm not in those type of horrible [216:54] scenarios where the robot is dangerous. Or when I'm driving, I, [216:58] I want, like, the car to stop when it should stop, and if [217:02] needed, someone takes care of that, right? And so there is so many ca- [217:06] cases, scenarios about real time, and so the goal of Kyber is to make real time [217:12] control of machine. Distances appear. [217:16] - It's incredible. And some of the same technology, some of the same ideas that we're talking about [217:19] is connected to what you're doing. [217:22] - And for me, it's amazingly challenging, right? Because I would say that on [217:25] video I'm doing okay, but networking I have so much [217:29] more to learn, right? It's um, about, like, congestion [217:33] protocols, bitrate adaptation in real time. But it's, [217:37] it's quite funny. And so I created this project and [217:41] we have fundraised in the US, of course. But it's open source, right? [217:45] This is important, right? Like, we've not said that, right? But everything on Kyber [217:49] is open source. [217:50] - So how do you make money? [217:51] - It's a dual license, commercial and AGPL, right? You remember what you said- [217:55] ... about licenses. Basically, if you [217:59] want to use Kyber in your product, you must have your full product open source. [218:04] If you want to use this amazing technology but not open source, [218:08] you pay the commercial license, right? So the small people or the [218:12] hobbyist and the very small guys who want to do that, they can use [218:16] the technology. They build something that is open source and cool. [218:18] - That's awesome. [218:19] - And if you're a large company, you're going to have the support, all the IP, [218:23] the right modification, and so on. So yeah, [218:26] it's really cool and also I'm building robots, and I love that, right? [218:32] Like we have-- Like the rover we have is 3D printed. We are [218:36] finishing a demo where it's an actual wing, right? Like a type of drone wings [218:39] that is also fully 3D printed. We are trying to do a sailboat that is 3D printed. [218:47] And we'll work on some humanoids. Of course, they are not going to [218:51] be very good robots, right? It's not our job, but we're here [218:55] for everyone to make robots. Cool. [218:57] - Ah, you're talking to the right guy. I love robots. There's a bunch of them [219:00] upstairs. And teleop is gonna be really, really important, [219:04] especially as the number of robots scales across the world. So 100%. Let's talk [219:08] about the future of multimedia. FFmpeg, VLC, but some of the [219:14] codecs, we didn't really mention AV2. So can we just [219:18] lay out what is AV2? What is the hope for it? What is H.265, H.266? [219:26] - So AV1 is this codec that is done by the Alliance for Open [219:29] Media, right? Where there is Google, Netflix, Amazon, [219:33] Apple VideoLAN, where we try to make a [219:36] royalty-free very good codec, right? And now it's being deployed. [219:40] But actually, the codec was finished in 2018, but a codec takes years to be [219:48] used in wide scenarios, right? So, [219:52] AV2 is the next generation of this codec. It's 30% better, right? So [219:57] if you keep the same quality, you get 30% bandwidth reduction compared to AV1. [220:03] - What's the connection with the dav1d and AV2? [220:06] - We are going to do a dav1d 2, right? That I call Devid because [220:11] de is two in French. Um- [220:13] - Ah, well done [220:13] - ... And you have to know that dav1d is an actual what we call recursive acronym, [220:17] right? Because it means D, dav1d, is an AV1 decoder, right? So [220:23] - Oh, nice. Nice. I didn't even think of that. And people should [220:27] know that dav1d is spelled with a one. [220:29] - Yes. It's... And so dav1d 2- [220:32] - It's gonna be spelled with a two. Please tell us [220:33] - ... is going to be D-A-V-2-D. Sorry, I don't know how you pronounce that. [220:38] And again, we did a demo at the CES of VLC running the first demo of AV2. [220:44] - So can you clarify to me the specification of AV2? And then the encoding and the [220:51] decoding [220:52] - Sure. So the specification is like the [220:56] document which explains how the codec is supposed to work, right? [221:00] - And that's really AV2. [221:02] - That is AV2, like H.264. [221:04] Right? Then you have an encoder. The current encoder is called AVM, [221:09] and there will probably be other encoders, probably one called [221:12] SVT-AV2, and those are the encoder. The same way [221:16] x264 is an encoder to H.264, the same way that x265 is an encoder for the H.265 [221:23] codec. And the decoders for AV1 is dav1d. The decoder for [221:31] AV2 is dav1d2. The decoder for H.264 is [221:37] ffh264 inside FFmpeg. The decoder for HEVC is ffhevc inside [221:45] FFmpeg. And there is a next generation codec [221:50] from the MPEG world after H.264, H.265. There is one that is called H.266, [221:56] also known as VVC. [221:59] - So HEVC is H.265. VVC is H.266. Why is H.266 super sexy- ... and so much better? [222:10] - So the question often we have is why are there two names? [222:13] Because most of the time it is a conjunct work from the ISO world [222:19] and the ITU, which is the International Telecommunication Union. [222:24] - These are these two regulatory bodies. [222:26] - No, one is a private entity and one is the United Nations. [222:28] - Which one is the private? [222:29] - ISO is private. [222:31] - In theory, H.264 is MPEG-4 Part 10, H.264/AVC. And this is the full name. [222:44] - So it's the concatenation of the ISO name and the ITU name- [222:48] ... even though they work together. This is, this is politics, historical, you know- [222:53] - And for HEVC, it's MPEG-H, H.265, HEVC. [222:59] And there is H.266, which is also named VVC. [223:03] - Is there a high-level thing to say about the improvement of- [223:05] - 30% each generation is a best summary. [223:09] - This is true both for the AV- ... codecs and the- ... H.264, 5, 6. [223:15] - So the professionals who are listening to us are going to kill us because they say, "No, it's [223:19] 35%, 25%- [223:20] - "No, it's 50, 60" [223:21] - ... it's 50," blah, blah, blah. But globally, you need to know that HEVC is [223:24] 30% better than H.264. H.266 is 30% [223:28] better than H.265 because there are so many cases and so many scenarios. [223:32] For example, there are cases, especially for screen recording, where the gains are [223:39] humongous because you arrive, you have the right tool that is done for that. [223:43] And so for a specific video, a new generation is going to give you [223:47] 70% gain or 80% gain. [223:50] Right? But there used to be a ton more codecs, but now the two [223:53] main codecs for transmission are the H.264, H.265, H.266, and the other is AV1, AV2. [224:00] - And I guess the major difference would be the cost of encoding. [224:05] - Yes, and the royalty of the patents. [224:08] And this is the reasons why you see the AV version of codecs, is [224:11] because they try to be as royalty-free, [224:15] which means no cost for the patents as much as possible. [224:19] Because what you need to know, and we've not talked about that so far, is [224:23] that multimedia is what we call a patent minefield. [224:27] There is two places where you have the most patents. It's [224:30] everything related to 3G, 4G, 5G, RF, and multimedia. [224:36] Um, because it's very mathematical, and you can get great gains and so on. So [224:42] Google and Meta and Netflix wanted something where it was [224:45] royalty-free. There are people who said that they have patents outside, but they are [224:49] fringe patents, right? So it's mostly true that it's patent-free. [224:53] - Oh, you should extend. Patent checking was done as part of the [224:57] standardization process in AV1, AV2, whereas patents are not even discussed [225:04] in the MPEG world. Patents are off-topic completely. [225:08] - Can you educate me at the patents side? [225:10] - So usually, so MPEG does a format, right? And then there is, [225:14] Everyone comes around and say, "Well, I have all those patents for the format," [225:18] and they do usually a union called what's called [225:22] MPEG LA, MPEG Licensing Association. And you [225:26] put all your patents in, and then you ask everyone who's [225:30] using this format to pay for it. [225:31] - Wait, can you elaborate? What does it mean to have a patent of a codec? Why is there many patents? [225:36] - Uh, imagine I'm doing something where I'm going to [225:40] instead of doing blocks which are square, I'm going to do rectangles, right? [225:44] - Oh, so every idea- ... somebody patents it. Oh, man. [225:52] Oh, man. People and their... How many lawyers are- [225:56] - I mean, it pays for a lot of lawyers, right? Like- [225:58] - The biggest issue is not the following, right? Because at time of H.264, the patents were, [226:02] let's call it, like, sane. But there was so much money in that- [226:09] ... that for HEVC, a lot- [226:12] there were a ton of things that were pushed inside the specification, which are not [226:15] useful in 99.9% of the time, but so [226:19] just one could add a patent on it. And so it became [226:23] that for HEVC licensing, there was MPEG LA [226:27] plus another patent pool called HEVC Advance. Plus- [226:31] - That one [226:31] - ... um, I think Nokia was outside of the patent pool. [226:34] - Yeah, a few of them are outside, and some other one that's- [226:36] - And so it was impossible to license, right? And I think that several months [226:40] ago, HP decided that they were going to remove support from [226:44] HEVC in their Windows laptops because the cost was increasing of those patents. [226:52] And it arrived- ... where a point where-- And there was uncapped patent. And so for [226:57] YouTube or Netflix, we could talk about hundreds of [227:00] millions of dollars of licensing for patents per [227:04] year. And they said, "You know what? At hundred million [227:08] per year, you know, I could create my own codec," and this is what they did. And so that's why we [227:12] have the Open Media alliance, Alliance for Open Media, where we [227:16] are part of, that is, that created AV1 and creates AV2. We create also audio codecs. [227:23] But yes. So the main difference would be that, and because you need [227:26] to work around the patents or go do some things that are not patented, [227:32] a lot of things are different, right? The basic things that were done in [227:36] MPEG-2 thirty years ago are, of course, out of patents. But [227:40] so for example, there is things like a golden frame, a S [227:43] frame or, or different type of- [227:46] - These are all patented ideas. [227:48] - Yeah, no, it's I can't believe it's not butter. I can't believe it's not a B frame. [227:52] It's, I mean, it's kind of what it is. In some ways, it's like a- [227:55] - Oh, so it's a different variant of a B frame. [227:57] - Yeah, that's to try and sidestep. Things like that. [228:00] - And so you need to have double creativity, right? Creativity in terms of [228:04] being more efficient, but creativity of being sure that you don't [228:08] infringe existing patents. And so, for example, VVC [228:12] is, has all the patents of HEVC plus new ones, right? It's [228:16] why AV2 tries to be as royalty-free as possible. [228:20] - To what degree does FFmpeg and VLC have to think about this kind of stuff? [228:24] - We don't, and one of the reasons why VLC was in France is [228:28] that France rejects software patents. So most of those patents [228:33] are illegal in France because I once made the calculus that [228:40] if I had to pay all the licensing fee for VLC, I needed to pay more [228:44] than two hundred euros per user, right? It's the same in dollars. [228:48] But most of those patents are invalid in Europe [228:52] because those are called, it's basically mathematical patents or idea [228:56] patents, and they are not valid in Europe. [229:00] - Uh, let me just at a high level, just out of curiosity. So the [229:03] meme online and the interwebs on X and Twitter and so [229:07] on, and my own, I have friends in Europe, [229:13] this, the sense is that Europe is not friendly to entrepreneurship. [229:17] They over-regulate, there's too much bureaucracy, and so [229:20] on. Is there anything positive to say? Is there hope for entrepreneurship- [229:26] - Yes [229:26] - ... in the future of Europe? Is Europe over from a tech perspective? [229:32] - Just look at the two of us, right? It's notable that there's two people from [229:36] the European continent on this podcast talking about video. It's fair to [229:40] say the community is weighted heavily. [229:42] - What you probably don't see yet is that there is a [229:46] new generation of entrepreneurs in Europe and [229:50] mostly in France. UK has done it since a long time because, [229:54] well, it's more, it's more Anglo-Saxon type of [229:58] business, look at business. But especially like what happened in [230:02] in France, and of course, sometimes a bit overdone with everything called French tech, [230:06] but today, most of the people who come on the market want to create [230:10] startups. Fifteen years ago, it wasn't the case. Everyone [230:14] wanted to work on big companies because when you failed in [230:18] in France, for example, twenty years ago, fifteen years ago, [230:22] and you destroy your company, which is normal for startup, right? You, [230:26] you were not allowed to create a new company, right? There was a lot of stigma. The [230:29] stigma is gone. [230:32] ... there is so many things happening on AI in France and so on, right? [230:35] So there is sure, over-regulations. I, [230:40] I know that, right? I'm an entrepreneur. But it has some good things also. [230:45] - I mean, is there some paralyzing aspects? You know, if I look at [230:49] the case of somebody I've become close with, Pavel Durov, [230:54] you know, he was blamed directly by the French government for the kind of things [231:02] his, quote, "platform" was hosting. [231:05] I could see the same kind of stuff basically, just as an example, [231:09] VLC being blamed for the kind of videos that people are watching. [231:13] - But they tried, right? Like we had, we had issues. Like- [231:19] - I mean, is that, that's the pressure that people worry about because if you have to think about [231:23] that kind of stuff when you're kind of just obsessed about- [231:25] - No, you don't think about it- ... and that's, that's okay, right? Like- [231:29] - But what if they come in? When, what if they show up and- [231:31] - There is no office. VideoLAN doesn't have an office. [231:33] - I mean, this is what happened with Pavel. They arrested him, right? So arrested him for [231:37] particular videos or, or a particular content that's being shared on the platform. [231:42] - Sure. I don't have any platform. Everything is on the client side. [231:45] - Yeah, but they're, they can still arrest you. [231:47] - On what ground? I'm not sharing anything. I'm not-- The content doesn't go through [231:51] my stuff. [231:52] - For sure, but it's still lawyer fees. That's the problem. [231:55] - Yes, that's correct. [231:55] - It's paperwork. So like, actually, if you had infinite trillions of dollars, [232:02] You would win easily because you're on the right side. But the thing is, there [232:08] is a degree to which they suffocate you with paperwork. That's the downside [232:12] of bureaucracy, through paperwork, through process. [232:15] You know, it's the Kafkaesque thing. [232:17] - You have to realize that one of the good things, for example in [232:21] in France or most of Europe, is that the-- [232:27] Answering to a court order does not make you bankrupt, right? It's not like [232:31] in the US, where it can actually bankrupt you, right? There is-- The [232:35] way the law system works is that, like [232:39] I receive lawyers' letters every week, right? And I can tell you that the cost [232:46] of lawyer fees for VideoLAN is less than ten thousand dollars per year, right? [232:52] Right? So that's not really scary. [232:53] - I mean, similar with Pavel. The intelligence agencies tried to like [232:57] say, "Can you put a backdoor in VLC?" [232:59] - Yes. Two of them. [233:01] - What, what do you say? [233:02] - No. Well, I was a lot less polite. [233:05] - I see you... Yeah, yeah. You're basically saying, "Hell no." [233:09] - Like, if we had to compromise our software, we would shut it down. This is clear. [233:13] - And what's the definition of co- compromise? Like allowing a [233:17] government to do a backdoor- [233:19] - There is no code that gets into VLC that we don't control, [233:23] and the way we compile VLC, you would call me completely paranoid. [233:28] Like, we compile on boxes that are offline, [233:32] where we start by compiling the compiler. We do everything [233:36] offline on places that have never been connected to the [233:39] internet. We-- The way we do signing, there is double [233:43] signature. And especially because, for example, we've [233:47] seen, and we believe it's a governmental agency that is [233:51] not from the Western world who tried to push a fake [233:54] binary into our own servers and that scared us a [233:58] lot. And VideoLAN is open source. How can you kill it? [234:03] Like, I move to where, right? I move to Malta. I move to I don't know, [234:07] Cayman Islands, and I change the domain name, and I [234:11] start again, right? Like, VLC is a tool. It's a tool that is going to help [234:18] people doing things. We are not a platform. [234:22] And for patents, well, I'm sorry, but most of the [234:26] patents... Like, you shouldn't be able to patent math and [234:29] matrices. Like, this is wrong. [234:32] - So does VLC ever, like, censor the kind of videos it [234:35] can play and not based on the content of the video? [234:39] - No, never. We never do that. [234:42] Because, like, VLC is completely offline. It doesn't talk to [234:46] any server, so we don't know anything that you're using the software for. [234:49] - So again, there's no government that can say, you know, [234:54] like the French government come in and say, "We don't want, [234:57] uh... I think anime is destructive to society. We don't want any [235:01] anime not allowed to be..." [235:02] - No, they cannot, they cannot do that. And also what they tried is to say, "Hey, [235:06] I want to know if that person watched that type of video." And the answer is like, [235:10] "No idea." [235:11] - So no on that too. So for surveillance, no. [235:14] - No, no, because the only infrastructure we have is a downloading infrastructure. [235:18] There is no telemetry in VLC, right? [235:20] - It, it would be difficult 'cause of the international nature. [235:23] It would be difficult for you to incorporate that code because there would be someone in the UK and [235:27] someone in Germany and someone in the US as part of VideoLAN who'd [235:31] be able to see that. It would be extremely difficult. [235:33] - The only thing that we can do, which happened, is like we had the issue-- We had the [235:37] case with some police in the US who said, "We have a murder case," right? [235:41] "Uh, and the file is destructed or doesn't play in that version of VLC. Could [235:45] you help us?" Right? We never have access to the video. It's like a normal support, right? [235:50] - Oh, it's really about playing the file? [235:51] - Yes. And, like, I remember in the middle of the Afghan War, right? I [235:55] received an email from someone in the army, right? I [235:59] don't remember the grades, right? It was just like, "We have a big issue with the latest [236:03] version of VLC because it doesn't play [236:07] correctly the file on an RTSP server that we have where there is [236:11] all the movies." And he says VLC is very important for the [236:14] morale on the troop on the ground, right? Because at night I think it might be boring, right? [236:18] So they have a collection of videos to watch or movies over there, right? So and, [236:22] and of course I did an update, and I broke some support of RTSP, [236:26] right? So I gave them another version just for them, right? Because it was [236:30] important. And because VLC is completely open source, I [236:34] think it is allowed on the US Army laptops, right? Because I [236:38] guess someone in, in the, in the US military actually looked [236:42] at it and say, "Well, okay, this is okay," right? And the way we document how we [236:46] process, that was okay, right? So the only way we work with [236:50] authorities is to help them doing support. [236:53] - That's amazing. That's an amazing story. Yeah. [236:56] - We don't see anything happening on how people use VLC, and this is strong. [237:00] - Do you feel the stress of this? So first of all, millions of people [237:04] using it. Second of all, the military using it. [237:09] Maybe sometimes pressure from governments. [237:11] - Yes. [237:11] - Does that, does that... That's a, that's a small team, right? [237:16] - Yeah, but- [237:16] - How big is VLC- like the core contribu- how many? [237:21] - Six, eight. But everything legally is only me. Everything that is legal is only me. [237:28] - You're not stressed about this? [237:30] - I used to stress about that a lot. [237:32] But the thing is, we're doing what we can for everyone, for the greater good. [237:38] We work that we make some extremely complex technology [237:41] easy for everyone. We're a tool, and every [237:45] tool is going to be used for great things and for bad things, right? You, you [237:49] cannot blame a tool, I think. And this [237:53] is, like, very important for us. Um, I used to [237:57] be a lot, in a lot of stress. I'm not anymore, right? [238:00] - What's the secret to your zen? I mean- Over and over in the chats I've had [238:05] with you in the conversation today about every [238:09] even tense topic, you're very zen. What's the source of zen? [238:14] - I have a way of thinking about what is the worst case scenario, always, right? [238:22] And the answer is, at the end, if I take like a, like a chess player, right? [238:27] In the end, am I dead? Yes or no? Right? And, and I do that nonstop, right? [238:33] And that's also how I do my, my startups, right? Is that [238:37] I'm here to get something right. What is the worst case? It goes [238:40] bankrupt. That's life. A company lives, a company dies. That's [238:44] okay, right? Like, and so my moral way is always like, [238:48] am I dying in the end? Am I hurting someone? If the answer [238:52] is no, then too bad, right? Like, oh, some lawyers are going to be [238:56] unhappy. What are they going to do? Take all the money of VideoLAN? [238:59] Wow. They're going to have 50 grand. Amazing, right? What are they [239:03] going to do with that? The source code is out there. It's not [239:07] stoppable. Also because what we do is good and it's done for everyone. [239:14] - That's beautiful. Uh, Kieran, you said that there's an active archiving [239:19] preservation community? [239:20] I think that's super fascinating. You wrote that they're stretched in budget, but they [239:24] see the extreme importance of FFmpeg as a Rosetta [239:27] Stone so that multimedia can be played a thousand [239:31] years from now. I mean, that's a beautiful way to see FFmpeg, [239:35] VLC as a tool for preserving visual knowledge. [239:41] - Yes, that's right. One of the coolest communities in open source [239:45] multimedia, mainly led by someone called Dave Rice, I'll give him a shout-out, I think [239:49] from City University of New York, is the archiving community. They've done [239:53] so much stuff. They value the open source, one, because yes, they lack [239:57] budgets, but two, they see the fact that archiving video is important for the world, [240:04] and but being able to play that is a big problem. Famously in the UK, [240:08] there was something called the New Domesday Book, and they archived [240:12] lots of stuff on BBC microcomputers. Within 10 to 15 years, no one [240:16] had the right software to play that. I think it was 20 years or something like [240:20] that, and someone had to go and reverse engineer this, and that was like 20 years. [240:23] Imagine that in a thousand years. [240:26] I think one of the great things about FFmpeg is it's written in C. C is [240:30] the closest to mathematics you're probably gonna get. The closest to logic is- [240:34] - Do you think in 1,000 years we'd still have C compilers? [240:37] - Yes. We have languages that exist that haven't changed too much. We have [240:40] mathematical notation that exists. It will be like Latin. C [240:44] will be like Latin. It will be a thing that you learn [240:47] from the past, but it will still be usable in certain contexts. [240:51] So the archiving community are really great practically. They, again, limited [240:55] funds. They funded the development of the FFV1 codec, so that's a [240:58] lossless codec. So the archiving community is really scared about the [241:03] act of compression losing things, and this could-- They have a fair point in this, you [241:06] know. If they compress too hard, it could change the [241:12] view of the material. There could be something slightly different here and there, so they're really [241:16] concerned about things need to be not just compressed well, but lossless [241:20] and be fast. And so they worked with FFmpeg to develop a whole new codec [241:24] designed for fast software-based encoding. [241:27] They're really concerned about resilience, so if they're storing on [241:31] tapes or other hard disks, I lose some bits, I need to [241:35] recover quickly. I can't lose a whole GOP because I've lost a bit- [241:39] ... something like that. [241:41] So they're a really great bunch of people. They funded GPU encoding in FFmpeg [241:45] to make FFV1 encode faster. And it's really about [241:48] preserving the world's multimedia heritage in a way [241:53] that's usable, and there's a lot of great teams and a lot of archival [241:57] groups across the world who've, who've chosen FFmpeg and [242:00] FFV1 as their archiving solution. And they can [242:04] really provide us also super specialist advice. They can- [242:08] ... explain, "Ah, in the 1950s, colorimetry was done like this on this [242:13] certain type of tape, and so there is [242:17] this special case that you need to handle, and you'll never get this anywhere else." [242:20] - You see, they know things on video that we don't. [242:24] Like, every time I talk to, was it Dave- [242:26] - Dave Rice [242:27] - ... or the people from the British, uh- [242:29] - British Film, uh [242:30] - ... Film, it's just like every time I just learn something new, and I've been doing video for 20 [242:34] years. They have, especially on colorimetry and colors. [242:39] - Storage, these other things. [242:40] - I mean, they have a deep, deep appreciation of the content itself, of the [242:44] video itself. And like, especially when you're thinking of lossless, [242:49] they're terrified of losing something essential- [242:52] ... about the thing, and in so doing, they're deeply understanding the [242:56] thing that is to be preserved, which you sometimes might not be thinking about when [243:00] you're- ... obsessing about the actual technology of the encoding and so on. [243:03] - And when you enter the rabbit hole of film scanners, right? So you take those- [243:10] ... those things to make to digital, like, it's like- [243:13] ... a huge topic that, like, would take another five hours of podcast- [243:18] ... just on that topic. [243:18] - On film, and there's a lot of film that needs to be archived. Film is degrading. It's maybe not stored [243:22] in the right environment. The other thing is they can... What they also do is, because it's open [243:26] source, they give this away, their workflows, to countries who can't afford [243:30] to have archiving institutions, where archiving is done by volunteers, it's done [243:34] by other things. They go and teach, you know, in India, they teach children to do, to [243:38] do FFmpeg commands. They're really great. They're really, [243:43] They're really the model community, the model ethos of what we're trying to achieve. [243:46] They are [243:48] such a great bunch of people, so interested in participating and being part of [243:52] something much bigger because they realize the work they're doing in a thousand [243:56] years is gonna tell a lot. [243:59] You know, in a thousand years we may be drowning in AI slop. This [244:03] stuff needs to be important and, you know, archived well. What was life like? [244:07] - Yeah, it feels like capturing the 20th century and the 21st century is [244:10] essential because it feels like a transition point, where we went from scarcity [244:17] of data to slop- ... oceans of slop, and that transition point is good to archive. [244:24] - It's important, yeah. [244:25] - But people don't realize we are losing today a ton of films. [244:29] There is a ton of things from the '30s, from the '40s, and the '50s that where [244:36] there is no value- [244:37] - And tape. '70s and '80s, there's tape, and there's not enough tape heads in the [244:41] world- [244:41] - To read all the tapes [244:42] - ... left to redo, so they have to decide what they want to archive and throw away the rest of the [244:45] tapes. There's huge moral hazard, I guess for want of a better phrase, around this [244:50] topic because [244:52] this is a digital record of human history and they have to make decisions [244:57] that... And there's digital stewardship, I suppose, for want of... I made that phrase up. That's not a real phrase. [245:01] Um, to make sure the world can have this information in something that's playable [245:06] by everybody, not- [245:08] ... playable on some device that, well, it doesn't exist anymore. [245:13] - And then there's like, realistically speaking, there's a needle in a haystack [245:16] where there's a lot of value in archiving all that [245:20] footage, and then over time finding the gems- ... that we don't know are there. [245:25] - Hey, there was something in that corner that we just didn't- [245:28] - Yeah. Uh-huh. [245:28] - And that, that would've been compressed away because it was some little thing. Oh, wow, there's [245:32] something there. [245:33] - That's it. [245:33] - And, and that's... They've made sure that it's lossless. They can prove [245:36] mathematically that it's lossless. They can run different [245:40] trade-offs for if there's bit fro- if they lose a bit, a single bit flips, I [245:44] can make sure that I only lose a portion of a given frame. We can do error [245:48] they can do error recovery on previous frames. They can do all sorts of different things. [245:52] - Do you think VLC and FFmpeg will be here 100 years from now? [245:57] - FFmpeg, yes. [245:58] - Yep, FFmpeg, yes. [245:59] - VLC, maybe. [246:01] - What's the future of... Where is FFmpeg going? [246:05] Where is VLC going? Like in the next... If you think about, like, five [246:08] years, 10 years, 20 years. [246:10] - Five years, 10 years is easy. The question is after that, right? The question is- [246:16] ... do we arrive at something called holograms, right? [246:19] - Yeah, so will VLC and FFmpeg expand- ... to whatever- [246:26] - Multimedia [246:27] - ... multimedia- [246:27] ... so multimedia might become, I'm sorry for the pothead expansion of [246:31] topic, but, you know if you look at something [246:35] like Neuralink with brain computer interfaces, it's very possible that [246:39] we start to consume what multimedia means is whatever [246:45] codec, whatever data that our brain wants to consume through the [246:48] brain computer interfaces. That's one. Then virtual reality, of course. [246:52] - You will have VLC for Neuralink. [246:55] - Yep, and you'll have FFmpeg -i input format human brain. [246:58] - Yeah. There's gonna be codecs for the brain. [247:01] - Sure, 100%. [247:03] - Of course. [247:03] - Yeah, to compress neural information, yeah. [247:05] - I mean, today there is like, there are new codecs for- [247:07] - Whoa [247:07] - ... for example, what we call point cloud, right? Or volumetric videos, right? [247:11] There is a ton of research on what we call RGBD, right? So codecs for [247:15] depths that is useful for robotics and for 3D things. [247:19] - Nice. [247:19] - There is a ton of codecs for compression of 3D elements. [247:22] - Compression for astronomy. [247:23] - Uh, for example, on VLC, we also have already a VR and XR version of [247:27] VLC. And also on Kyber, right? We talk about Kyber. On Kyber, we also [247:31] like do streaming of XR content on for the glasses who cannot [247:35] have enough power or inside the Apple Vision or the Quest. [247:39] So we already work on streaming 3D, XR, interactive, low latency. There is [247:47] something called volumetric video, point cloud videos, so it's [247:51] not stopping. And yes, at some point it will manage 3D data [247:54] inside VLC and FFmpeg, right? It's obvious. [247:58] - So that's where it is moving, like the community is open. [248:01] - Not everyone in the community sees that, but like, as Kieran [248:05] and I, we are entrepreneurs, we know where it's going. We see that, right? [248:10] - So I suppose that there is a tension probably inside FFmpeg. It's like, [248:14] "Hey, listen, folks, we're really good at doing video and audio, [248:20] so like why expand? Like let's do the thing we're really good at doing." [248:25] - In order to answer that question, we need to answer the [248:28] definition of what is multimedia. [248:31] And multimedia is a digital representation of several [248:38] streams for the human senses. And we will do that, right? So imagine there is now a [248:46] way to not have a mic, but have an odor sensor- ... and a diffuser of odors. [248:52] It will get into FFmpeg. [248:54] - So your demuxer is coming up. [248:56] - Yes. Yes. [248:57] Of course, your demuxer has a new track type that is basically odors, right? [249:01] And you already have- [249:02] - Smell, touch. [249:03] - It's like audio. You'll have a left and right nose track. You have a left and right audio pair. It's easy. [249:07] - Yes, of course. [249:09] - Stereo smell. [249:09] - Stereo smell, yeah. [249:10] - So in VLC, for example, we already have a plugin for haptic. It's mostly [249:14] for what we call 4D cinemas, right? You know, those ones on hydraulic, [249:18] I don't know how you say that. All the hydraulic- [249:21] - Hydraulic arms. Hydraulic, um- [249:22] - Arms. And where everything is moving, like you have in theme parks, right? [249:25] And there is a data feed synchronized where, [249:29] which is basically transporting this information. [249:32] - Is there yet a standard for that? [249:33] - There are many standards, right? Um- [249:35] - This is... You make me so happy. [249:38] - And so of course, like we have a plugin which is not in the normal version of VLC- [249:43] - That's good. [249:43] - ... that is basically transporting those type of movements, which is physical movements, [249:47] which is haptic movements, right? It is a human sense, so it will get in. [249:54] - That's such an exciting future. Was it... I mean, it's a small [249:57] community of developers. How do you pull that off? [250:01] Like if you're a contributor to FFmpeg or VLC, it feels [250:05] stressful. Like it, just looking on Twitter, [250:09] it's like it's a huge amount of work to make [250:13] it work on all these different operating systems, an incredible effort. [250:17] - No, see it in the other direction. We are not [250:21] the contributors. We are the maintainers, [250:24] right? So we maintain for everyone. Meaning that, for example, every year there is [250:31] around 150 people who contribute to VLC and maybe [250:34] 300 on FFmpeg, right? Our goal as a small team is [250:38] to get all the contribution in. So if there is more [250:42] usage, there will be more contributions, and those people will [250:46] do the right module, the new format, and so on. We care [250:50] about the architecture of VLC, the architecture of FFmpeg, right? [250:54] Now we're doing things in VLC, which is spatial audio, right? We did the [250:58] demo not long ago. There was [251:02] changes needed on the architecture, and we did the first spatial audio [251:06] module. When it's going to add the second one, it's going to be easy, or the third one is going to be easy, [251:10] right? Our goal, and it's going to be the same for others or [251:14] haptic, right? We need to work the architecture so that modules can [251:18] be added to add future capabilities. [251:21] So yes, we are going... We are multimedia framework, so that's not [251:25] just audio and video. It's everything that is timed and-Represent [251:32] something that you can sense. And if it's brainwaves, it's going to be brainwaves. [251:36] - I think that's inevitable. Sorry. [251:38] - I love this on so many fronts because, so FFmpeg and VLC are [251:43] pushing companies and pushing the world to [251:47] standardize. So for example, to standardize- [251:49] ... brainwaves, right? So standardize... It would push, [251:53] like I hope Neuralink comes up with a standard for, for multimedia [252:00] via brain computer interfaces or for robots with haptic. [252:05] - By experience, what happens is always the same, right? [252:08] You start, it's a new topic. There is like five different standards [252:12] because everyone starts to do this. The hype goes down because every time the hype [252:16] goes down, then people start to say, "Well, you know what? You, we need to do a [252:20] standard." People, because two or three companies, usually not the leader, but [252:24] the two or three followers do a standard, [252:27] and then we implement the standard and, and then it's the end of the curve. [252:31] It starts to be more paper. [252:32] - And then the leader's kind of pressured into it because it is better to do a [252:36] standard. Yeah. [252:36] - Example, 3D audio, right? [252:39] Six or seven years ago, it was everything about 3D. You go, you had the Cardboard on [252:43] Android. You had two audio formats. They're all dead, right? [252:47] And now it's coming back with actual use cases, and we learn [252:51] from the mistakes of the past standard. So it will be the same everywhere. [252:56] - And not try to avoid closed. I saw somewhere you, [253:00] you didn't have too many nice things to say about Dolby. [253:03] - No, I don't. Um- [253:04] - What is can you educate me on why, [253:09] where they went, what, what did they do bad that made you mad? [253:13] - It used to be an amazing company doing tons [253:17] of great things with amazing engineers. They defined what sound was. [253:22] And now it's mostly- [253:25] - Lawyers [253:25] - ... lawyers and licensing things. [253:27] - Oh, so they're, yeah, it's, they're closing stuff off. They're trying to make money on licensing. [253:30] - No, it's just like they don't innovate as much as they did- [253:33] - I see [253:33] - ... and so on. It's a bit like I'm sorry to say, right, like HP, right? [253:38] - Very true. [253:40] - Oh, since we talked about Twitter a bunch in different contexts, do you have a, [253:44] do you have a favorite, do you have a, and least favorite, most embarrassing tweet [253:50] on either VideoLAN or FFmpeg Twitters? [253:53] - The two, my two favorites are, "Talk is cheap, send patches." I think that, that- [253:57] ... embodies a lot of the stuff doesn't get, as we've talked about, stuff [254:01] doesn't get built unless someone does it. It doesn't just appear from the ether. [254:05] The other one that I like is "FFmpeg, nothing is beyond our reach." [254:10] I think that comes from a US military satellite patch where I think they, they invented some [254:14] kind of monitoring system that could see the whole world, and this was released. [254:19] - Wasn't there something where FFmpeg was running on a rover on Mars also? [254:22] - Yeah, so FFmpeg is used by the Mars rover the Mars 2020 [254:26] rover to compress pictures. They really wanted—they [254:30] wrote a paper about it, and they really wanted to use as much commercial off-the-shelf technology as [254:34] possible. [254:34] - Oh, that's cool. [254:35] - FFmpeg runs on Mars, so we are a multi-planetary open source library. [254:39] - Nice. [254:41] - Very often we've seen— [254:43] - Nice [254:44] - ... Tweets for people using VLC in weird [254:48] places. A lot of the people doing Formula 1 [254:52] are in all the paddocks, they use VLC to play the live feed. We've [254:55] seen the European Space Agency. We've seen SpaceX, like, [254:59] monitoring the launches with VLC, and, like, it, like, [255:03] fills you with joy, right? So- [255:05] - I've seen a particle accelerator. [255:07] - Oh, yeah, yeah. We had one of the most amazing things that I [255:12] went for was to go to the CERN at the LHC because they were using VLC to monitor all [255:19] the sensors on the ring because the ring is 27 kilometers. [255:23] And so they had some analog cameras- [255:27] ... and they were using some of the capture cards to go to analog to [255:30] VLC, so VLC could stream on their multicast network for the whole CERN [255:34] to access that. And, like, I visited that in 2010 with Laurent and— [255:43] like, we fixed their issue in an hour or something like that, right? Because it was [255:47] some parameters maybe not well documented at that time. [255:51] And he said, "Okay, for the whole day, what do you want to do?" And we visited everything. Like- ... [255:55] things with antimatter and—and colliders and so on. [256:00] And that was, like, one of the most amazing days of my physics background. [256:06] - Yeah, it's used, like, everywhere. Any tweets, uh, Kieran, you regret? [256:12] - Tweets I regret? [256:14] - Or is it like that, how does the French song go? Regret nothing. [256:17] - "Je ne regrette rien." Yeah. [256:19] - Yes. Uh, that's very important for me, right? [256:21] Don't regret anything. No, it's because regrets [256:26] are a tax on your mind, right? So learn from your mistakes, but don't regret. [256:31] Because you've done it, so unless you have a time machine to go back in time, [256:37] don't regret, right? It's going to just tax your brain. Learn from your mistake, [256:41] sure. Don't regret. [256:43] - It's like it reminds me, it's beautiful. It's a tax on your brain. It reminds me of the Johnny Depp [256:47] quote I saw where he was saying, "Hate, you know, I [256:51] don't hate. That's, hate is a very expensive emotion." [256:56] - Are you comparing me to Johnny Depp? Because that would be your first one. [257:00] - Well, gentlemen like I said, I'm eternally [257:03] grateful for the software that, you know, the two of you and [257:07] the bigger community have been part of building with FFmpeg and VLC and everything [257:11] else. I'm eternally grateful for the spicy tweets. Never stop. [257:17] And I'm grateful that you would talk with me today and give me this [257:23] sexy hat. I feel like a wizard. I feel special. And I feel [257:28] special to get a chance to talk and celebrate the piece of software that brought me so much joy over [257:33] the years. So thank you for everything, and thank you for talking today. [257:36] - Thank you for having us. [257:37] - Thank you so much. [257:38] - Thanks for listening to this conversation with Jean-Baptiste Kempf and Kieran Kunhya. [257:43] To support this podcast, please check out our sponsors in the description where you can [257:47] also find links to contact me, ask questions, give feedback, and so on. [257:51] And now let me leave you with some words from the legendary Linus Torvalds. [257:57] "Most good programmers do programming not because they expect to get [258:01] paid or get adulation by the public, but because it is fun to program." [258:08] Thank you for listening, and I hope to see you next time.