Oculus Co-Founder Sees AI Replacing Film Crews With Synthetic Stars

The Rise of AI in Film Production: How Machine Learning is Changing Storytelling With UC Berkeley — Ai4

Ai4

Generative machine learning pipelines will compress multi-year production schedules into months, rendering physical sound stages and CAD artistry obsolete as synthetic performers compete for industry awards.

AI-assisted summary of The Rise of AI in Film Production: How Machine Learning is Changing Storytelling With UC Berkeley — original source. Check the original for full context.

Source published: 2026-08-15

Detailed notes

## Background Published on YouTube by Ai4 on August 15, 2026, this presentation features Oculus co-founder and UC Berkeley Board of Trustees member Jack McCauley. The session explores how emerging machine learning tools are transforming scriptwriting, editing, visual effects, and overall storytelling workflows across the entertainment industry.

## TL;DR Tracing his four-decade career from engineering the mouse scroll wheel, Guitar Hero hardware, and Oculus VR to pioneering Unix VFX workflows on Terminator 2, Jack McCauley contrasts historical manual asset drafting with modern generative AI pipelines. He demonstrates how text-to-3D synthesis will collapse multi-year development timelines—such as the $256 million production cycle of Grand Theft Auto V—down to months, triggering severe workforce disruptions for entry-level programmers and CAD artists. While showcasing AI-generated synthetic acting and predicting the rise of virtual, Oscar-caliber digital stars, McCauley emphasizes that human authorship remains indispensable for high-level narrative structure, emotional depth, and spontaneous live performance.

## Atmosphere & style McCauley delivers an unvarnished, pragmatic talk infused with the quiet authority and dry wit of a lifelong builder and low-level systems engineer. The presentation weaves easily between nostalgic anecdotes of late-night Chinese hotel rooms and Sun workstation render farms to stark, sobering assessments of white-collar creative displacement. The atmosphere feels conversational yet deeply technical, grounding dramatic forecasts of synthetic Hollywood cinema in decades of tangible hardware evolution.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=4s) Jack McCauley's co-founding of Oculus VR and its subsequent $3+ billion acquisition by Facebook was made possible by long-term professional networking, having been recommended to Palmer Luckey through contacts from his time engineering Guitar Hero. - [02:21](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=141s) McCauley originally invented the mouse scroll wheel to navigate a third dimension in gaming; failing to patent the uncredited invention taught him to strictly protect all subsequent intellectual property. - [04:21](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=261s) Groundbreaking early 1990s visual effects on films like Terminator 2 relied on clusters of Unix Sun workstations running weeks-long software renders prior to the advent of dedicated graphics and tensor processing units. - [06:49](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=409s) Historical mechanical computing based on Napier's principles provided exact fractional accuracy up to astronomical scales, avoiding the rounding limitations inherent in modern floating-point arithmetic. - [09:00](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=540s) Multi-year, manual world-building pipelines—exemplified by the $256 million budget spent drafting and texturing Grand Theft Auto V—are being replaced by real-time text-to-3D generative asset synthesis. - [11:37](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=697s) Generative world-building will compress multi-year media production timelines to months, causing significant disruption to technical art roles and unprecedented friction in entry-level computer science employment. - [13:20](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=800s) Graphics compute has evolved from assembly-coded custom engines like Doom (1993) to accessible commercial engines running on modern smartphones with graphics performance matching eighth-generation home consoles. - [15:28](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=928s) Traditional Hollywood production lots and physical crews are increasingly economically vulnerable to synthetic cinema, where solo creators can render Oscar-caliber performances with entirely virtual actors. - [18:02](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=1082s) Even as machines take over visual effects, camera blocking, and asset rendering, foundational plot construction and emotional narrative architecture still necessitate deliberate human authorship. - [19:40](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=1180s) Spontaneous, vulnerability-driven human creative mediums—such as live improvisational comedy and concert musical performance—remain fundamentally insulated from generative AI replacement.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=4s) The Oculus Acquisition and the Power of Staying Connected

Reflecting on a 40-year career across the entertainment and technology industries, Jack McCauley credits much of his success to serendipity and maintaining professional relationships. After co-founding Oculus VR to build virtual reality gaming headsets, he found himself surprised by the company's meteoric trajectory. McCauley viewed himself strictly as an engineer rather than an executive engaged in corporate dealmaking, and he had initially been skeptical of consumer appetite for VR headsets, wondering who would actually want to walk around wearing them. While stationed in a hotel room in China overseeing hardware manufacturing, he received a phone call at 3:00 in the morning from the CEO delivering life-changing news: Facebook had acquired Oculus for over $3 billion. As the boss quipped, McCauley now had "a stack of dollar bills from here to the moon."

That pivotal venture originated purely through maintaining professional connections. McCauley had previously served as the head engineer on the smash-hit video game franchise *Guitar Hero*. Throughout his career, he made a habit of calling former colleagues once a year just to check in. That habit paid off when contacts at Activision recommended him to an 18-year-old Palmer Luckey and the founding team as a seasoned digital hardware and software engineer capable of taking their virtual reality hardware prototype into production.

- Facebook (now Meta) acquired Oculus VR in 2014 for approximately $3.2 billion ($2 billion initial announcement, totaling over $3 billion with earnouts and employee packages). - McCauley oversaw the hardware engineering and overseas manufacturing operations in China. - The Oculus connection was brokered directly by contacts from McCauley's previous role as chief engineer on *Guitar Hero* at Activision/RedOctane.

---

[02:21](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=141s) The Scrolling Mouse Wheel and the Lessons of Intellectual Property

McCauley describes himself as an inventor from childhood, having won a scholarship at age nine for building a toy. His lifelong passion for programming and low-level development led him to Logitech in the 1990s, where he had the opportunity to collaborate with computer pioneer Doug Engelbart, renowned for presenting "The Mother of All Demos" and inventing the original computer mouse. At the time, mice operated primarily in DOS environments and lacked any scrolling mechanism. Seeking a way to navigate a third dimension in video games—adding an in-out depth axis to the standard X and Y axes—McCauley added the scrolling wheel to the mouse.

Although the scrolling wheel ultimately became an industry-wide standard for scrolling through desktop windows and navigating user interface menus rather than McCauley’s intended gaming depth axis, he failed to file a patent for the mechanism. At the time, patent law seemed overly complicated and uninteresting to him. After watching the entire industry copy his scroll-wheel design without consequence, McCauley made it a strict rule to patent nearly every subsequent invention he developed. When launching Oculus, he and a tight-knit team of about five people filmed their Kickstarter campaign video inside his own facility and managed all marketing, PR, and product development internally.

```mermaid flowchart TD accTitle: Evolution of McCauley's Hardware Innovations accDescription: Flowchart tracing McCauley's key hardware projects from the mouse wheel to Oculus VR.

subgraph Peripherals ["Logitech Era (1990s)"] A["Doug Engelbart's Mouse"] --> B["McCauley adds Scroll Wheel for Z-Axis Gaming"] B --> C["Copied across industry for GUI scrolling; Unpatented"] end

subgraph Gaming ["Activision / RedOctane Era"] C -.-> D["Guitar Hero Hardware Engineering"] end

subgraph VR ["Oculus VR Era (2012–2015)"] D --> E["Activision network connects McCauley to Founders"] E --> F["Hardware built & Kickstarter filmed in McCauley's shop"] F --> G["Acquired by Facebook / Meta"] end ```

---

[04:21](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=261s) Hollywood VFX on Sun Workstations and Academic Roots

Beyond interactive consumer electronics, McCauley worked as a technical contributor in Hollywood on classic feature films, including *Terminator 2: Judgment Day*, *Dick Tracy*, and *Edward Scissorhands*. His contributions focused on digital multi-channel audio systems, which contributed to an Academy Award win for his technical group. During the production of *Terminator 2* around 1990, dedicated graphics processing units (GPUs) and modern tensor processing units (TPUs) did not yet exist. Generating the groundbreaking visual effects required a network of 26 Unix-based Sun-1 and Sun Microsystems workstations running X Window environments, where engineers shelled out commands and compiled code directly through terminal windows.

Recognizing early on that artificial intelligence would eventually reshape computing, McCauley completed mathematics-heavy coursework through UC Berkeley Extension in the 1990s to master neural network theory, even though contemporaneous computer hardware lacked the processing horsepower required to run practical AI workloads. Beyond computing, his interests expanded into green energy and electric vehicles (EVs). A UC Berkeley alumnus, McCauley serves on the university's Board of Trustees, teaches green energy and EV powertrain technologies at Berkeley, and contributes research to both the RAND Corporation and the Mathematical Sciences Research Institute (MSRI / SLMath) in Berkeley.

- *Terminator 2: Judgment Day* (1991) visual effects rendering relied on clusters of Sun Microsystems workstations executing software renders before hardware GPUs existed. - McCauley pursued mathematical foundations of machine learning in the 1990s through UC Berkeley Extension. - Academic and institutional roles include UC Berkeley Trustee, green energy instructor, RAND Corporation research contributor, and affiliate of the Mathematical Sciences Research Institute.

---

[06:49](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=409s) Engineering Pursuits: Custom Cars, Drone Jammers, and Napier's Calculator

Operating a personal machine shop equipped with CNC mills and lathes, McCauley built custom vehicles, including a lightweight Fiat 500 retrofitted with a motorcycle engine. Turning his attention to defense hardware in 2019, he anticipated the threat of low-cost commercial quadcopters being weaponized with improvised explosives—a tactic that later became widespread in modern conflicts like the war in Ukraine. To counter this, he engineered a handheld anti-drone "blaster." By integrating a spectrum analyzer with directional antennas, the device scans radio frequencies and jams the RF control signals transmitted between ground operators and their drones, disabling their flight commands.

McCauley also explored mechanical computing history by hand-fabricating a precision calculating apparatus based on the mathematical principles of John Napier, the late-16th-century Scottish mathematician who developed logarithms and "Napier's bones." Frustrated by cheap plastic commercial replicas, McCauley built a functional model capable of computing large mathematical operations up to $10^{20}$—astronomical scale figures—as well as performing square roots, cube roots, multiplication, and division. He emphasizes that historic fractional mathematics offered absolute precision, unlike modern floating-point arithmetic using decimal points, which inherently introduces rounding limits and orders of precision.

---

[09:00](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=540s) From Six-Week Renders to AI Generation: Grand Theft Auto V and Modern Pipelines

Demonstrating the exponential progress in visual computing, McCauley highlights a sequence from *Terminator 2* that required 20 Sun workstations running continuously for six weeks just to render a single visual effects shot—an operation that modern video game engines can now render in real time on the fly. As a low-level systems programmer who has worked on PlayStation POSIX operating system layers, McCauley contrasts historical manual asset creation with the emerging paradigm of artificial intelligence.

To illustrate the sheer scale and expense of traditional digital world-building, he points to Rockstar Games' *Grand Theft Auto V* (published by Take-Two Interactive). Developing the title required overhauling the proprietary RAGE (Rockstar Advanced Game Engine) and sending teams to Los Angeles to photograph real buildings, which digital artists then manually drafted into 3D models using AutoCAD before applying custom textures and lighting. This laborious development process cost an unprecedented $256 million, though it paid off dramatically by grossing over $1 billion within days of launch. McCauley argues that AI is fundamentally disrupting this model: instead of teams spending weeks modeling and texturing individual buildings by hand, modern machine learning systems allow artists to generate production-ready 3D game assets and complex architectural environments instantly through natural language prompts.

[11:37](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=697s) Generative World-Building and the Tech Workforce Shock

Specifying digital environments is undergoing a radical shift through prompt-driven generation. Rather than manually modeling individual assets, a creator can simply describe the scene—demanding a brick building, broken windows, and walls covered in graffiti—and the system synthesizes it instantly. That manual artistry was previously the cornerstone of development and did not exist when franchises like *Grand Theft Auto* first emerged. This shift will compress production timelines from eight-year slogs down to six months or even a single month. Teams composed of digital artists with CAD and specialized programming backgrounds will see their conventional roles eliminated as generative tools take over asset generation.

Beyond game design, this efficiency shock is sweeping through market research and broader technical employment. Rather than retaining extensive research teams, granular industry metrics and historical data can be retrieved instantly using AI search. The downstream effects on the white-collar labor market are already manifesting in entry-level hiring. A friend’s daughter recently graduated with a degree in computer science and remains unable to secure a job—a scenario that was virtually unthinkable fifteen or twenty years ago when computer science majors routinely fielded multiple offers before graduation. While the ultimate labor equilibrium remains uncertain, the disruption to content creation is definitive.

```mermaid flowchart TD accTitle: Traditional vs AI-Driven Production Pipeline subgraph Traditional["Traditional Production (Years)"] direction LR T1["Manual CAD & 3D Modeling"] --> T2["Handcrafted Shading & Texturing"] T2 --> T3["Extensive Code & Asset Assembly"] end subgraph Generative["AI Generative Pipeline (Months)"] direction LR A1["Natural Language Prompting"] --> A2["Instant Scene & Texture Synthesis"] A2 --> A3["Automated Engine Integration"] end Traditional -.->|"Compressed Workflows & Labor Reduction"| Generative ```

- Production cycles for massive open-world franchises are projected to shrink from eight years down to a year or six months. - Traditional technical art roles—specifically artists with CAD and programming training—face widespread obsolescence. - Entry-level software engineering and computer science graduates are experiencing unprecedented hiring friction compared to previous market cycles.

[13:20](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=800s) Graphics Compute Evolution: From Doom's Assembly to Photorealistic Game Engines

Contrasting graphics milestones highlights how drastically development overhead has transformed over thirty years. In 1993, the year Nvidia was founded, id Software launched *Doom*, a landmark title hand-crafted in low-level assembly language by pioneers like John Carmack (who later joined Oculus). By 2023, independent projects like the bodycam first-person shooter *Unrecord* achieved near-photorealistic visual fidelity by utilizing off-the-shelf, commercially licensable game engines rather than proprietary low-level code.

```mermaid flowchart LR accTitle: Evolution of Compute and Tooling subgraph Y1993["1993: Doom Era"] E1["Low-Level Assembly Code"] --> P1["Custom Software Engines"] end subgraph Y2002["2002: Sixth Gen Console"] P2["PlayStation 2: 6.2 GFLOPS"] end subgraph Y2020s["2020s: Modern Compute"] E2["Commercial Engines (Unreal)"] --> M1["iPhone Pro (PS4-Class FLOPS)"] end Y1993 --> Y2002 --> Y2020s ```

During the growth of Oculus, securing appropriate content presented a cultural and corporate dilemma after being acquired by an advertising giant. Headset sales demanded intense, visceral experiences like first-person shooters, yet the parent company operated an ad-supported platform governed by strict anti-violence sensibilities. Meanwhile, compute capability skyrocketed: while the 2002 PlayStation 2 operated at 6.2 billion floating-point operations per second (6.2 GFLOPS), modern smartphones like the iPhone 14 and iPhone Pro carry graphics horsepower comparable to a PlayStation 4, constrained only by thermal throttling and battery draw.

- *Doom* (1993) required low-level assembly programming, whereas modern photorealistic shooters leverage rented, commercial game engines. - Oculus faced strategic friction aligning visceral gaming software with a parent company reliant on advertising revenue and averse to violent media. - The PlayStation 2 pushed 6.2 GFLOPS in 2002, a benchmark dwarfed by contemporary smartphones whose GPU performance matches eighth-generation home consoles.

[15:28](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=928s) Synthetic Cinema and the Disruption of Hollywood

The legacy infrastructure of Hollywood—physical sound stages, production lots, dedicated film editors, and extensive hair and makeup crews—is facing severe economic vulnerability, prompting studios to steadily abandon Los Angeles. AI-driven synthetic media can now bypass traditional physical shoots entirely. To illustrate the shift, a clip created by an independent thirty-year-old creator is showcased, featuring a fully AI-rendered, synthetic likeness of Robert Downey Jr. performing in an interrogation drama.

The synthetic scene plays out as an intense dialogue between an investigator and a suspect: "You didn't come here for nostalgia," the suspect begins, with the interrogator countering, "I came because you agreed to talk. Talking isn't a crime. Last night, between 10:00 and 11:00, where were you?" When the suspect hesitates and answers, "At home. Alone," the interrogator observes that "people pause when they're remembering," eventually pressing that a figure matching the suspect's height was spotted outside the victim's window at 10:07. Although the footage carries visible digital artifacts, it underscores an imminent shift where television, feature films, and streaming platforms like Netflix rely on synthetically generated casts: "You could have a virtual star, a person that doesn't even exist in the world, and he can win an Oscar."

- Physical studio lots and production crews in Los Angeles are dwindling as generative pipelines substitute on-set labor. - Solo creators can write, direct, and render dialogue scenes featuring recognizable celebrity likenesses without cameras or actors. - Fully synthetic, non-existent digital personas will soon possess the performance capabilities required to compete for premier industry accolades like Academy Awards.

[18:02](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=1082s) Q&A: Human Creative Direction, Musical Craft, and Comedy

Addressing an audience inquiry regarding the fate of human creativity if traditional Hollywood production collapses, the necessity of core human narrative vision remains central. While cameras, crews, and physical actors may be supplanted by rendered archetypes akin to digital Homer Simpsons, human storytellers must still construct the foundational plot, dialogue, and dramatic conflicts that engage audiences. The emotional architecture and narrative artifacts will continue to require deliberate human authorship, even if the public fixates on synthetic celebrities rather than real-world tabloid personalities.

```mermaid flowchart TD accTitle: Human Role versus Automated Machine Tasks subgraph HumanCore["Irreplaceable Human Roles"] H1["Plot Construction & Narrative Drive"] H2["Subversive Comedic Writing & Timing"] H3["Live Performance & Shared Audience Presence"] end subgraph Automated["Automated Machine Tasks"] A1["Set Architecture & Environmental Dressing"] A2["Lighting Calculations & Texture Mapping"] A3["Automated Camera Blocking & Render Output"] end HumanCore -->|"Directs & Instructs"| Automated ```

When asked about the fate of soundtrack composition and music scoring, an anecdote from the creation of *Guitar Hero 1* provides perspective. Developed on a shoestring budget by a team of roughly ten people in Santa Clara, the production initially could not secure master audio tracks because major artists dismissed the concept as foolish. Early skeptics questioned why players would not simply learn a real instrument, an objection countered by the sheer difficulty of mastering real guitar compared to an accessible interactive game. Subsequent commercial triumphs enabled *Guitar Hero 2* to license authentic master recordings from bands like Metallica. Musical composition remains a nuanced, distinct craft, and live performance environments cannot simply be reduced to automated sound playback.

Turning to a final question from an improv actor regarding the visceral pushback against AI within Chicago's comedy community, a clear boundary exists between digital visual effects and improvisational theater. Live comedy hinges entirely on audience rapport, vulnerability, and genuine human wit: "writing funny shows is done by a funny person." While generative AI will effortlessly automate digital set construction, procedural texturing, virtual lighting, and dynamic camera angles, reproducing the spontaneous, intuitive timing of human comedic performance remains well outside the machine's immediate reach.

- Human authors remain indispensable for high-level narrative architecture, thematic tension, and scriptwriting. - *Guitar Hero 1* was built by roughly ten people in Santa Clara who were initially denied master recording licenses because record labels doubted the concept. - Live musical and comedic performances rely on shared human physical presence and organic timing, insulating them from purely synthetic replacement.

## Notable quotes > [00:04](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=4s) a stack of dollar bills from here to the moon

> [02:21](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=141s) The Mother of All Demos

> [06:49](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=409s) Napier's bones

> [15:28](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=928s) You didn't come here for nostalgia — Suspect

> [15:28](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=928s) I came because you agreed to talk. Talking isn't a crime. Last night, between 10:00 and 11:00, where were you? — Interrogator

> [15:28](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=928s) At home. Alone. — Suspect

> [15:28](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=928s) people pause when they're remembering — Interrogator

> [15:28](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=928s) You could have a virtual star, a person that doesn't even exist in the world, and he can win an Oscar. — Jack McCauley

> [18:02](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=1082s) writing funny shows is done by a funny person — Jack McCauley

## Mentioned Jack McCauley, Palmer Luckey, Doug Engelbart, John Napier, John Carmack, Robert Downey Jr., Metallica, Oculus VR, Facebook, Meta, Activision, RedOctane, Logitech, Sun Microsystems, UC Berkeley, RAND Corporation, Mathematical Sciences Research Institute (MSRI / SLMath), Rockstar Games, Take-Two Interactive, Nvidia, id Software, Netflix, Guitar Hero, Guitar Hero 1, Guitar Hero 2, Terminator 2: Judgment Day, Dick Tracy, Edward Scissorhands, Grand Theft Auto V, Doom, Unrecord, Rockstar Advanced Game Engine (RAGE), AutoCAD, PlayStation 2, PlayStation 4, iPhone 14, iPhone Pro, Fiat 500

## Video description Jack McCauley - Co-founder Oculus, Board of Trustee at UC Berkeley, UC Berkeley

Explore how AI tools are utilized in scriptwriting, editing and visual effects to enhance the filmmaking process.

[Source](https://www.youtube.com/watch?v=7ejoDa3u8pw)

中文简报

Oculus联合创始人：AI将以虚拟影星取代摄制团队生成式机器学习流程将把数年的制作周期压缩至数月；随着合成演员参与角逐行业大奖，实体摄影棚与CAD美术制作或将过时。回顾从研发鼠标滚轮、《Guitar Hero》硬件、Oculus VR到在《Terminator 2》中开创Unix视觉特效流程的四十年职业生涯，Jack McCauley将以往的人工作资绘制与现代生成式AI流程进行了对比。他展示了文本生成3D技术如何将数年的开发周期（例如《Grand Theft Auto V》耗资2.56亿美元的制作周期）压缩至数月，并对初级程序员和CAD美术师的就业造成严重冲击。在展示AI合成表演并预测具有奥斯卡水准的虚拟数字明星即将崛起的同时，McCauley也强调，人类在高级叙事结构、情感深度和即兴现场表演中的主导地位依然不可或缺。 Jack McCauley 凭借多年维护的人际关系，经由《吉他英雄》（Guitar Hero）时期的联系人引荐给 Palmer Luckey，联合创立了 Oculus VR，并最终促成 Facebook 以超过 30 亿美元将其收购。 (0:04) McCauley 最初为了在游戏中导航三维深度轴而发明了鼠标滚轮，但因未申请专利被行业广泛免费借鉴，这促使他在随后的发明中坚决执行严格的专利保护策略。 (2:21) 从《终结者2》耗时数周的 Sun 工作站渲染，到《侠盗猎车手5》（GTA V）耗资 2.56 亿美元的手工建模，传统数字世界构建管线正被实时文本生成 3D 资产的生成式 AI 全面颠覆。 (9:00) 生成式技术将把数年的制作周期压缩至数月，不仅对 CAD 建模等技术美术岗位造成巨大冲击，也给计算机科学专业的初级就业市场带来了前所未有的阻力。 (11:37) 随着独立创作者能够完全依托虚拟演员渲染出奥斯卡级别的表演，好莱坞实体摄影棚与现场制作人员正面临合成影像带来的严峻经济冲击，但核心剧本叙事与现场即兴表演依旧离不开人类。 (15:28)

Reading summary

Jack McCauley argues that advances in machine learning and compute hardware are radically disrupting digital entertainment, collapsing multi-year asset creation and physical studio pipelines into automated generative workflows. However, he maintains that core narrative architecture, comedic timing, and live performance remain fundamentally human domains. Exponential Compute Gains and Automated World-Building (9:00) McCauley contrasts the exponential growth of graphics hardware with historical production constraints. In 1991, generating a single visual effects sequence for Terminator 2: Judgment Day required 20 to 26 Sun Microsystems Unix workstations running software renders continuously for six weeks without dedicated graphics processing units. By comparison, Rockstar Games spent an unprecedented $256 million manually photographing Los Angeles and drafting 3D models in AutoCAD to build Grand Theft Auto V (2013). Today, off-the-shelf commercial game engines run photorealistic graphics in real time, and modern smartphones rival eighth-generation home console compute. According to McCauley, generative artificial intelligence fundamentally upends this manual asset pipeline. By allowing creators to synthesize complex architectural environments, geometry, and textures directly through natural language prompts, development timelines for massive open-world titles could compress from eight-year cycles down to six months or even a single month. Hollywood Disruption and White-Collar Labor Shocks (11:37) The transition toward synthetic media threatens traditional studio infrastructure and technical employment. McCauley observes that physical sound stages, production lots, and specialized crew roles in Los Angeles are steadily declining as generative pipelines substitute on-set labor. Independent creators can now write, direct, and render dialogue scenes with synthetic celebrity likenesses without cameras or actors—a shift McCauley forecasts will eventually lead to non-existent digital performers competing for premier industry accolades like Academy Awards. Concurrently, this automation is triggering labor friction across technical disciplines. Traditional digital art roles combining CAD and specialized programming face obsolescence as generative systems take over geometry and texturing. McCauley notes this compression is already impacting entry-level employment, citing recent computer science graduates unable to secure positions in a job market that historically offered multiple competing offers before graduation. Human Authorship in Narrative, Comedy, and Performance (18:02) Despite widespread automation in technical rendering, McCauley emphasizes that machines cannot replace foundational human creative vision. While artificial intelligence can seamlessly execute environmental dressing, procedural texturing, virtual lighting, and dynamic camera angles, human authors are still required to craft plot arcs, dramatic conflict, and emotional resonance. Audiences may accept synthetic performers, but the underlying emotional architecture still requires human authorship. McCauley draws a sharp boundary around live performance, particularly comedy and music. Recalling the early skepticism faced during the development of Guitar Hero before it secured licensed master tracks, he notes that musical composition and live execution retain an organic craft. Similarly, improvisational comedy relies on shared physical presence, vulnerability, and spontaneous wit—intuitive human comedic timing that McCauley argues remains well outside the reach of machine learning models. While generative tools drastically lower capital requirements and technical barriers in film and game production, the entertainment industry faces unresolved questions regarding the displacement of specialized physical crews and the economic stability of entry-level technical talent as synthetic media matures.

精炼阅读摘要

硬件工程师兼Oculus联合创始人Jack McCauley指出，底层计算性能的飞跃与生成式人工智能正在从根本上解构影视与游戏制作的传统劳动密集型管线；尽管自动化生成大幅压缩了资产构建周期并对技术就业造成结构性冲击，但核心叙事架构、幽默时机与现场人际共鸣仍然牢固依赖人类的独创性。算力跃升与制作管线范式转移 (9:00) 回顾过去三十年的数字图形演进，计算能力的飞跃彻底改变了数字资产的生产方式。在1991年制作《终结者2：审判日》（Terminator 2: Judgment Day）时，业界尚未出现专用的图形处理器（GPU）或张量处理器（TPU），团队依赖20至26台基于Unix架构的Sun-1及Sun Microsystems工作站集群进行软件渲染，仅完成单个特效镜头就需要连续运算长达六周；而1993年发布的初代《毁灭战士》（Doom）亦全凭低级汇编语言手写打造。相比于2002年PlayStation 2主机6.2 GFLOPS（每秒62亿次浮点运算）的性能，当代如iPhone 14等智能手机的图形算力已可比肩第八代主机PlayStation 4，使现代独立项目能够直接借助成熟的商业游戏引擎呈现高拟真画质。这种算力演进推动了生产流程从高成本手工建模向即时合成转型。以Take-Two Interactive旗下Rockstar Games开发的《侠盗猎车手5》（Grand Theft Auto V）为例，该作耗资2.56亿美元，需在重构专有RAGE引擎的基础上，派遣团队前往洛杉矶实地测绘拍照，并由数字艺术家通过AutoCAD手动建立三维模型并人工贴图打光，耗费数年制作。McCauley强调，现代机器学习模型颠覆了这一范式：创作者如今只需通过自然语言提示词即可即时生成具备复杂几何结构与贴图的生产级三维场景，使过去数周的人工资产建模与纹理流程得以秒级完成。制作周期压缩与技术劳动力市场震荡 (11:37) 生成式技术的成熟预计将使开放世界数字场景的开发周期从以往长达八年的漫长工程压缩至六个月甚至一个月。在传统开发模式中，根据具体要求（如砖石建筑、破损窗户与涂鸦墙面）进行逐项手工建模与着色是研发的核心基石；而随着自然语言提示直接合成完整场景，原本由具备CAD制图与编程背景的专业技术艺术家所承担的常规岗位正面临大范围边缘化甚至消亡。这一效率冲击不仅局限于游戏美术，还迅速波及至市场调研与软件开发等更广泛的白领技术就业领域。McCauley提到，AI检索已能即时调取详尽的行业数据，直接缩减了企业对传统研究团队的雇佣规模；而在基础软件开发领域，初级毕业生的就业难度空前加剧。以其熟人家庭为例，计算机科学专业毕业生当前甚至难以获得一份工作，这与十五至二十年前该专业学生毕业前通常手握多个录用名额的市场行情形成了鲜明反差，表明内容与技术生产领域的就业结构已受到实质性震荡。虚拟影业兴起与好莱坞物理体系解构 (15:28) 在传统影视工业方面，依赖实体摄影棚、外景场地以及庞大常驻工种（如剪辑师、化妆与发型团队）的旧有模式在经济上面临严峻考验，导致制片厂正在持续撤离洛杉矶。AI驱动的合成媒体正在逐步绕开传统的现场实拍模式：一名三十岁的独立创作者即可凭借AI软件，在无摄像机与实景演员的条件下独立编导并渲染出一场审讯室双人对白戏，并逼真再现演员Robert Downey Jr.的面部特征与台词神态。 McCauley指出，尽管当前的AI生成画面仍存在局部数字伪影，但流媒体与院线电影由纯合成角色出演已成为明确趋势。他据此推断，未来的影视制作可能会推出完全虚构、在物理世界中并不存在的数字角色，而这种纯算法生成的虚拟演员完全有可能凭借出色的数字表演角逐并赢得奥斯卡奖（Academy Award）等行业最高奖项。人类叙事架构与现场表演的不可替代性 (18:02) 即便机器能够在自动化布景、程序化材质纹理、虚拟光影与动态运镜上取代人工操作，McCauley强调核心的叙事构建依然离不开人类主导。无论观众未来关注的是真实明星还是数字虚拟偶像，驱动戏剧张力、情节冲突与情感共鸣的剧本和情感框架，仍必须由人类创作者自发设计与编排。在音乐与即兴表演领域，机器替代的边界尤为明确。McCauley以早期在圣克拉拉仅由十人小团队开发《吉他英雄》（Guitar Hero）的经历为例指出，该项目初期曾因唱片行业质疑“玩家为何不直接学习弹奏真吉他”而无法取得母带授权，但该作的商业大获成功以及续作赢得Metallica等乐队的认可，证明了交互模拟与深层音乐创作之间的本质差异。同样，面对即兴戏剧演员对AI技术的抵触情绪，McCauley认为现场喜剧高度依赖演员暴露自我的脆弱性、与观众的现场互动以及严苛的即兴幽默时机把控，这种完全由人类个体主导的自发性表达，是自动化算法在短期内无法复刻的坚实护城河。 McCauley从自身跨越数十年的硬件工程与视觉特效实践出发，论证了生成式AI加速重塑数字生产管线的必然性；然而，生成式工具在带来效率跃升的同时，对初级技术研发者的就业冲击、行业实体资产的萎缩，以及完全由算法生成的虚拟演员在艺术评价机制中的伦理合法性，仍属于未决且极具争议的现实课题。

Detailed notes — zh

## Background 本视频由 Ai4 于 2026 年 8 月 15 日在 YouTube 平台发布。视频记录了 Oculus 联合创始人兼加州大学伯克利分校校董杰克·麦考利（Jack McCauley）的深度演讲，探讨人工智能工具在剧本创作、影视剪辑与视觉特效中的落地应用，以及机器学习如何深刻改变影视与故事叙事的全流程。 ## TL;DR 麦考利结合自身 40 年横跨《吉他英雄》、好莱坞特效《终结者2》以及 Oculus 创立与并购的传奇工程履历，系统回顾了图形计算与渲染技术的代际飞跃。他指出生成式 AI 正在彻底颠覆传统游戏与影视工业的人力密集型资产制作模式，有望将 3A 级大作的开发周期从 8 年骤减至数月，但也将引发 3D 建模师与软件开发者的就业震荡。通过展示纯 AI 渲染的影视片段，他预言虚拟演员与全流程数字化将瓦解传统好莱坞制片体系，同时在问答环节强调，人类在核心叙事构思、音乐艺术性与即兴幽默等情感共鸣维度的价值依然无法被算法替代。 ## Atmosphere & style 整场演讲兼具资深硬件极客的从容自嘲与硬核技术洞见，麦考利以信手拈来的行业掌故和生动演示牢牢抓住了听众的注意力。随着话题从算力演进转向 AI 对影视制作与就业市场的剧烈冲击，现场氛围逐渐由轻松幽默转为充满危机感的深刻反思。在随后的问答互动中，他对艺术灵性与人类独特创造力的理性坚守，为这场关于技术颠覆的探讨注入了兼具温度与远见的人文底色。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=4s) 杰克·麦考利分享了 Oculus 的创立历程：早年在《吉他英雄》积累的人脉促成了与年轻创始人的合作，仅凭约5人的小团队完成了早期研发与众筹，最终以超30亿美元被 Facebook 收购。 - [02:21](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=141s) 麦考利在罗技工作期间为解决 DOS 游戏的视角缩放问题发明了鼠标滚轮，但因未申请专利被全行业广泛仿制，这一教训促使他日后极度重视知识产权保护。 - [02:21](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=141s) 好莱坞在90年代初严重缺乏现代 GPU 算力，制作《终结者2》等经典大片的液态金属特效时，需动用26台 Sun 工作站集群进行长达数周的离线渲染。 - [09:00](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=540s) 传统3A游戏如《GTA 5》耗费2.56亿美元且极度依赖美术师纯手工 CAD 建模，而生成式 AI 正在将这种高昂的劳动力密集型开发流程推向物理与成本极限的颠覆点。 - [11:37](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=697s) 生成式 AI 将把大型游戏 8 年的制作周期压缩至 6 个月甚至更短，但也将给 CAD 建模美术师、初级程序员及基础调研岗位带来残酷的就业替代冲击。 - [13:20](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=800s) 现代移动设备（如旗舰 iPhone）的图形算力已可比肩家用主机 PS4，而底层开发也从早期的硬核汇编语言演进为高度封装的商业引擎与多模态生成工具。 - [13:20](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=800s) Oculus 在被 Meta 收购后遭遇内容困局：作为依靠数字广告盈利且文化上反暴力的母公司，难以全力支持最能驱动硬件销量但包含暴力元素的硬核 FPS 游戏。 - [15:28](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=928s) 传统好莱坞制片体系正面临瓦解，AI 全流程影视制作未来不仅能节省数千万美元开销，甚至可能催生出斩获奥斯卡奖项的纯虚拟演员。 - [18:02](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=1082s) 即便传统实体制作工业衰退，人类的核心叙事价值仍不可替代，剧情构思、戏剧对白和情感内核将由极少数人类创意构架师主导。 - [19:05](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=1145s) 音乐作为一门极富仪式感与现场互动的手艺，其情感传达和艺术灵性构成了坚固壁垒，短期内难以被单纯的算法和像素渲染彻底取代。 - [20:37](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=1237s) AI 的核心优势在于计算机图形学、场景建模与光影渲染等物理世界的数字化，而在即兴喜剧等极度依赖现场气场共鸣与人类微妙幽默感的领域存在明显能力边界。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=4s) 职业生涯与 Oculus 的创办及收购内幕在娱乐和硬件工程领域深耕40年的杰克·麦考利（Jack McCauley）回顾自己的职业生涯时，坦言自己非常幸运，而这一切的核心在于“在对的时间结识了对的人”。麦考利早年曾在红极一时的音乐游戏《吉他英雄》（Guitar Hero）中担任首席工程师。正是因为在动视（Activision）工作期间建立并维系的人脉，让他结识了 Oculus 的初创团队——一位年仅18岁的年轻创始人和团队管理者。动视的熟人当时向对方引荐他：“你们需要硬件工程师、数字电路专家还是软件工程师？我认识一个人可以全包。” 麦考利起初对虚拟现实（VR）头显的市场前景持怀疑态度，甚至吐槽：“谁会买这种东西？戴着那玩意儿到处走，怎么可能！”但事实证明他是错的。Oculus 随后在麦考利的实验室里录制了具有里程碑意义的 Kickstarter 众筹视频，当时整个团队仅有大约五个人，包揽了所有的公关、市场宣传与工程实现。随着公司的爆发式增长，科技巨头伸出了橄榄枝。麦考利回忆道，作为一名纯粹的技术极客，他完全没有参与幕后的并购谈判：“我就是个工程师，不负责市场营销，也不管收购谈判。”当时他正在中国跟进 VR 头显的工厂产线，凌晨3点在酒店房间里突然接到老板的电话：“你猜怎么着？Facebook 刚刚以约32亿美元收购了我们的业务！恭喜你，你现在的钞票堆起来能一路叠到月球了！”虽然实际金额没那么夸张，但这笔交易彻底改变了他的生活。 - **关键人脉习惯**：即使退休多年，麦考利依然保持着每年给行业老朋友通一次电话维系关系的习惯。 - **Oculus 早期规模**：早期团队仅约5人，在麦考利的自有场地内完成了早期的核心研发与众筹宣传视频拍摄。 - **收购时刻**：麦考利在驻扎中国产线期间于凌晨3点接到收购喜讯，Facebook（现 Meta）以超30亿美元的估值完成收购。 ```mermaid flowchart TD accTitle: Oculus 创立与并购历程 accDescription: 展示杰克·麦考利从《吉他英雄》到联合创立 Oculus 并被 Facebook 收购的演进过程。 subgraph Phase1["人脉积累与起步"] A["《吉他英雄》首席工程师 （动视人脉）"] --> B["结识 Oculus 创始人 （18岁帕尔默·拉奇等人）"] end subgraph Phase2["车库式创业与众筹"] B --> C["麦考利工程实验室 （5人团队研发）"] C --> D["Kickstarter 众筹爆发 （颠覆性 VR 头显）"] end subgraph Phase3["量产与退出"] D --> E["驻扎中国把控硬件制造"] E --> F["2014-2015 被 Facebook/Meta 收购 （总额逾30亿美元）"] end ``` --- ### [02:21](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=141s) 从鼠标滚轮发明到好莱坞大片的工程传奇麦考利自幼便展露出惊人的发明与编程天赋，9岁时就因制作玩具获得奖学金。20世纪90年代在罗技（Logitech）工作期间，他曾与“所有演示之母”（The Mother of All Demos）的主角、计算机鼠标的发明者道格拉斯·恩格尔巴特（Douglas Engelbart）共事。当时的电脑鼠标只有 X 轴和 Y 轴，且运行在 DOS 系统下。为了在电子游戏中实现视角的拉近拉远（Z 轴控制），麦考利发明了带有滚轮的鼠标机制。然而，这项发明最终并没有如他最初设想的那样主要用于游戏缩放，而是随着 Windows 系统的普及被广泛用于菜单和文档的上下滚动。由于当时对专利缺乏认知，麦考利并未为其申请专利，导致该设计被全行业迅速仿制。这一教训让他在此后的职业生涯中将专利保护放到了极其关键的位置。除了消费级硬件，麦考利还曾深入好莱坞顶级特效制作的前线，作为核心技术贡献者参与了《终结者2：审判日》（Terminator 2: Judgment Day）、《至尊神探》（Dick Tracy）以及《剪刀手爱德华》（Edward Scissorhands）等多声道数字音频与图形计算工作，其所在团队曾斩获奥斯卡奖项。他钟爱 Sun Microsystems 的工作站（如 Sun-1 / Sun-4 系列），借助其强大的 Unix 终端与 X Window 视窗环境高效编译代码。据他回忆，在1990年前后没有现代 GPU 的时代，为了渲染《终结者2》中标志性的液态金属人等视觉特效，动用了多达26台 Sun 工作站进行集群渲染。此外，麦考利早在90年代就预见到人工智能的发展前景，并在加州大学伯克利分校（UC Berkeley）Extension 进修了密集的 AI 数学理论课程，只是受限于当年的算力瓶颈，这些算法在当时无法直接落地。 - **鼠标滚轮诞生背景**：原意为 DOS 游戏提供前后缩放的第三轴（Z轴），未申请专利后被行业普遍采纳为窗口滚动工具。 - **好莱坞经典视效算力**：1990年制作《终结者2》时缺乏独立 GPU，依靠26台 Sun 工作站集群进行长时间离线渲染。 - **学术与理论储备**：90年代便系统学习了基于高等数学的机器学习基础理论，亲历了从早期算法到现代 TPU / GPU 算力爆发的转变。 --- ### [05:45](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=345s) 跨界狂热：从绿色能源、改装车到无人机干扰器与古代计算器离开工业界一线后，麦考利的极客精神并未减退，反而跨界涉足了多个前沿与硬核制造领域。他曾就职于加州利弗莫尔的早期风能企业 US Windpower，如今在母校加州大学伯克利分校担任客座讲师，教授新能源、绿色电网与电动汽车（EV）平台架构技术；同时兼任兰德公司（RAND Corporation）的研究贡献者以及伯克利数学科学研究所（MSRI）的校董成员。在动手制造方面，麦考利曾经营过配备完整 CNC 机床和车床的高规格工坊，亲手将摩托车引擎改装进了一辆菲亚特500（Fiat 500）微型车内部。不仅如此，麦考利始终对尖端技术与地缘冲突保持着敏锐的工程师直觉： - **无人机防御干扰枪（Drone Blaster）**：早在2019年，麦考利就预见到搭载炸药的廉价民用无人机将对地面部队构成灾难性威胁（正如后来在俄乌冲突中所见）。他利用现成射频元器件与频谱分析仪，手工拼装打造了一台信号干扰设备，成功阻断操作员通过 iPad 或遥控器发向无人机的控制链路（由于无线电管制法规，该原型未申请专利亦未商用）。 - **复原纳皮尔算筹计算器（Napier's Calculator）**：麦考利对16世纪末苏格兰数学家约翰·纳皮尔（John Napier）的机械计算装置十分着迷，嫌市面上的塑料复刻品粗糙，便亲自加工打造了一台高精度的实体机械计算器。该装置完全基于分数数学运行，能在没有小数点的时代精确完成高达 $10^{20}$ 宇宙级天文数字的乘除法与开平方、开立方运算。 --- ### [09:00](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=540s) 从《终结者2》的六周渲染到《GTA 5》的手工资产与 AI 生成变革麦考利现场展示了一段《终结者2》的经典特效片段（因台词包含脏话而特意静音播放）。他感慨地指出，在90年代初，仅仅为了渲染电影中的这一个单一特效镜头，就动用了20多台 Sun 工作站整整计算了6个星期。而得益于半导体和算力的飞跃，当年需要耗时一个半月的渲染任务，如今在现代视频游戏引擎中已经可以实现毫秒级的实时渲染（On the fly）。谈及游戏工业资产生产的演进，麦考利以游戏史上的标杆巨作《侠盗猎车手5》（Grand Theft Auto V，2013-2014年发行）为例进行了深度剖析。开发商 Rockstar Games 为了达到极致的写实度，不仅彻底重写了 RAGE 游戏引擎，还派出了庞大的美术团队前往洛杉矶实地采风拍摄。整座虚拟城市中成千上万栋建筑、道路、材质贴图与色彩，全部是由美术师坐在电脑前使用 AutoCAD 等工具纯手工耗时数年一点一滴建模复刻出来的。这种极其劳动力密集型的开发模式，使得《GTA 5》的研发预算高达惊人的2.56亿美元——相比之下，当年动视收购整个《吉他英雄》团队也才花费了1亿多美元。尽管《GTA 5》在发售后10天内狂揽20亿美元成为商业奇迹，但这种依赖海量人力堆砌 3D 资产的传统开发流程已经逼近物理与成本极限。麦考利在此点明演讲核心：随着现代生成式 AI 的崛起，开发者如今只需输入一段自然语言描述，即可直接生成高精度的 3D 建筑与游戏资产，游戏与电影工业的内容生产模式正迎来颠覆性的代际革命。 | 对比维度 | 90年代初影视特效（如《终结者2》） | 2010年代大作开发（如《GTA 5》） | 现代 AI 驱动时代 | |:--- |:--- |:--- |:--- | | **核心算力与工具** | 20~26台 Sun 工作站集群 | 重构 RAGE 引擎、AutoCAD 手工建模 | GPU / TPU 集群与多模态生成模型 | | **单场景/资产耗时** | 单个镜头离线渲染耗时长达 **6周** | 整个城市建筑全部依靠人力手动绘制建模，耗时数年 | 自然语言描述即时生成，渲染达到毫秒级实时响应 | | **开发成本与门槛** | 硬件极其昂贵、算力极度匮乏 | 研发预算高达 **2.56亿美元** | 极大幅度降低 3D 资产与场景构建的边际成本 | ### [11:37](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=697s) AI 颠覆游戏开发周期与就业市场的震荡 Jack McCauley 生动地描绘了场景生成方式的范式转移：在过去开发《侠盗猎车手》（Grand Theft Auto，GTA）初期，如果要设计一栋带碎玻璃窗、外墙布满涂鸦的红砖楼，必须由拥有 CAD 背景的艺术家和程序员纯手工逐笔构建；而如今，创作者只需用自然语言下达描述，AI 便能瞬间自动生成。这种变革对娱乐行业的影响将是“震撼人心”的——以往动辄耗时 8 年的 3A 级大作，未来开发周期将缩短至 1 年甚至 6 个月。然而，这也伴随着残酷的就业震荡，大量依赖 CAD 建模的专业美术人员“将会彻底消失”。 ```mermaid flowchart LR accTitle: 游戏内容生产模式演进 subgraph Traditional["传统开发模式"] A["CAD / 美术团队手工建模"] --> B["耗时 8 年研发周期"] end subgraph AIEra["生成式 AI 时代"] C["自然语言 Prompt 描述场景"] --> D["AI 瞬时全自动生成"] D --> E["研发周期骤减至 6 个月"] end ``` 这种冲击不仅局限于游戏和影视，还将蔓延至市场研究乃至基础编程岗位。McCauley 提到，自己今天来演讲前直接用 Google 和 AI 瞬间查清了 GTA 的所有背景数据，AI 的信息检索与整合能力已相当强大。更令人担忧的是软件行业的就业严冬：他朋友的女儿刚从计算机科学专业毕业，如今却根本找不到工作——这在 15 到 20 年前是完全不可想象的，那时计算机专业的毕业生在离校前就能斩获多个高薪 Offer。虽然就业格局的最终走向尚不明朗，但对于内容创作产业而言，这场风暴已无可逆转。 - **研发周期压缩**：像 GTA 这类超大型游戏 IP 的制作周期预期将从 8 年缩减至 6 个月甚至更短。 - **高危岗位预警**：从事 CAD 建模的美术设计师、基础市场调研人员及初级程序员首当其冲。 - **计算机应届生就业现状**：行业需求发生结构性剧变，名校 CS 专业毕业生面临“毕业即失业”的罕见困境。 --- ### [13:20](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=800s) 从《Doom》到《Unrecord》：算力跃迁与 Oculus 的内容困局回溯图形计算的发展史，McCauley 将 1993 年 NVIDIA 成立之初的经典游戏《Doom》（《毁灭战士》）与 2023 年呈现极致拟真质感的战术射击游戏《Unrecord》进行了对比。他特别提到，当年在 Oculus 共事的元老之一（John Carmack）正是《Doom》的缔造者。1993 年的《Doom》完全由极底层的汇编语言（Assembly Language）硬核写成，而三十年后的现代游戏则是基于现成的商业商业游戏引擎开发，开发者只需租用引擎或支付版税即可快速搭建高规格世界。在此期间，McCauley 还剖析了 Oculus 被 Meta（Facebook）收购后遭遇的内容困局。作为一家靠数字广告盈利而非制作游戏内容起家的母公司，其企业文化带有天然的“反暴力”倾向，这导致团队在推行最能拉动 VR 头显销量的内容（如硬核 FPS 射击游戏）时举步维艰：“当你的母公司靠卖广告赚钱、并且文化上排斥暴力内容时，你该如何打造那种能真正卖爆头显的爆款内容？这极为困难。” 谈及硬件算力，他列出了一组惊人的浮点运算性能（FLOPS）跨越： - **1993 年 PC 硬件**：每秒浮点运算量极为初级，全凭底层汇编代码榨取硬件潜能。 - **2002 年 PlayStation 2**：浮点运算能力达到约 62 亿次/秒（6.2 GFLOPS）。 - **移动端硬件跨越**：如今一部 iPhone 14 / iPhone 17 Pro Max 具备的图形处理能力已经比肩甚至超越了当年家用主机 PlayStation 4 的水平；不过受限于散热与电池寿命，手机无法长期维持满血峰值运行。 --- ### [15:28](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=928s) 好莱坞的危机：纯 AI 影视片段演示与“虚拟奥斯卡影帝” McCauley 毫不客气地预测：传统好莱坞体系正在走向瓦解。如今洛杉矶的影视工业深陷危机，大量传统制片厂正在撤离，曾经庞大的片场搭建团队、后期剪辑师、化妆师群体生存艰难。为了佐证这一观点，他现场播放了一段由一位 30 岁左右的创作者完全使用 AI 渲染制作的黑色电影（Film Noir）风格短片：短片呈现了一场扣人心弦的审讯室对手戏，画面中赫然出现了 AI 生成的“小罗伯特·唐尼”（Robert Downey Jr.）的拟真形象。对白充满张力： > “你来这儿不是为了怀旧。” > “我来是因为你答应谈谈。谈话可不犯法。昨晚十点到十一点之间，你在哪儿？” > “在家里。一个人。”（伴随着回忆时的自然停顿） > “你是在指控我什么吗？如果是在指控，我要求见律师。” > “我没有指控你，我只是在听你说话的声音。十点零七分，有人看到一个身形与你吻合的人站在她的窗外……” McCauley 承认当前画面仍带有肉眼可见的“AI 渲染感”，但他坚信未来的电视剧、院线电影和 Netflix 流媒体剧集都将全面转向此类 AI 全流程制作。技术不仅能降低数千万美元的拍摄与布景开销，甚至还将催生出“完全不存在于现实世界的纯虚拟演员，并且这位虚拟影星未来完全有能力斩获奥斯卡最佳男主角”。如同当年 GPU 与游戏引擎重塑游戏产业一样，生成式 AI 将彻底重构影视创作底层逻辑。 --- ### [18:02](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=1082s) 问答一：好莱坞消失后，人类创造力将去向何方？现场观众提问：“如果好莱坞传统工业真的如你所言被彻底抹去，那么原本寄宿并孕育于该行业的创造力将何去何从？从业者会转型去从事什么？” McCauley 回应道，即便底层技术发生颠覆，人类的核心叙事价值依然无法取代。剧情构思（Plot）、戏剧对白（Dialogue）以及驱动观众情感共鸣的核心故事内核，必须由人类亲自创造。未来的制作模式将高度聚焦于极少数具备核心创意的关键人物。他幽默地将未来的虚拟明星类比为《辛普森一家》中的荷马·辛普森（Homer Simpson）——大众消费的是被渲染呈现出来的角色人格。传统小报关注的明星八卦或许会消亡，但角色的“人格侧写与魅力”依然具有商业号召力。人类创意人员将从繁杂的实体置景、服化道和琐碎制作中解放出来，转型为掌舵故事灵魂的核心构架师。 --- ### [19:05](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=1145s) 问答二：AI 会取代电影配乐吗？《吉他英雄》的版权往事一位观众提到自己的儿子正在攻读音乐编曲与作曲专业，焦虑地询问电影与游戏配乐是否也将全盘被 AI 替代。 McCauley 并没有给出冰冷的定论，而是结合自己主导硬核音乐外设《吉他英雄》（Guitar Hero）的初代研发往事作答：初代《吉他英雄》启动时预算极其微薄，位于加州圣克拉拉（Santa Clara）的硬件团队仅有约 10 人。起初他们完全拿不到 Metallica 等知名大牌乐队的原版母带（Master Tracks），唱片公司和艺人们一致觉得用塑料玩具吉他跟着打节拍是极其荒谬愚蠢的想法；连 McCauley 当时也半开玩笑地质疑：“为什么不教玩家弹真正的吉他呢？”（团队的回答则是“玩家并不想学，而且吉他太难了”）。直到一代大获成功，后来的《吉他英雄 2》才得以顺利拿到顶尖大师的原版母带。借此案例，McCauley 强调： - **音乐是纯粹的手艺与仪式（Craft）**：人类对现场演奏的追捧不可替代，“总不能开一场演唱会，台上就孤零零放着几个扬声器吧？” - **艺术性壁垒**：音乐中的情感传达和艺术灵性远比简单的像素渲染复杂，他不认为 AI 会在短时间内彻底取代人类作曲家的艺术地位。 --- ### [20:37](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=1237s) 问答三：即兴喜剧的抵触情绪与 AI 的能力边界最后一位来自芝加哥即兴喜剧（Improv Comedy）领域的观众提出了一个文化割裂的现象：在他的日常主业中大家都很拥抱 AI，但线下演员与艺术圈子内部却对 AI 生成的艺术抱有强烈的排斥与仇视态度，他询问这种反感情绪是否会成为阻碍 AI 技术普及的屏障。 McCauley 分析指出，剧场艺术与计算机图形学有着本质的不同： - **现场交互无法替代**：喜剧很大程度上依赖现场观众的即时互动与气场共鸣，活生生的人类演员在舞台上的张力是无法被算法替代的。 - **幽默感的不可计算性**：“写出真正幽默爆笑的剧本，必须由一个本身就风趣幽默的人来完成”，AI 很难真正理解人类幽默的微妙内核。 - **AI 真正的领地在于物理世界数字化**：McCauley 重申，他所预言的核心变革集中在**计算机图形学（Graphics）、虚拟场景搭建、光影渲染、材质贴图与机位运镜生成**。AI 可以秒级建立精美的虚拟摄影棚，但在即兴喜剧等极度依赖人类情感共振的深水区，AI 仍难以撼动人类演员的地位。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=4s) 你们需要硬件工程师、数字电路专家还是软件工程师？我认识一个人可以全包。 — 动视引荐人 > [00:04](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=4s) 谁会买这种东西？戴着那玩意儿到处走，怎么可能！ — 杰克·麦考利 > [00:04](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=4s) 我就是个工程师，不负责市场营销，也不管收购谈判。 — 杰克·麦考利 > [00:04](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=4s) 你猜怎么着？Facebook 刚刚以约32亿美元收购了我们的业务！恭喜你，你现在的钞票堆起来能一路叠到月球了！ — Oculus 团队管理者 > [13:20](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=800s) 当你的母公司靠卖广告赚钱、并且文化上排斥暴力内容时，你该如何打造那种能真正卖爆头显的爆款内容？这极为困难。 — 杰克·麦考利 > [15:28](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=928s) 你来这儿不是为了怀旧。 — AI 短片角色 > [15:28](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=928s) 我来是因为你答应谈谈。谈话可不犯法。昨晚十点到十一点之间，你在哪儿？ — AI 短片角色 > [15:28](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=928s) 在家里。一个人。 — AI 短片角色 > [15:28](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=928s) 你是在指控我什么吗？如果是在指控，我要求见律师。 — AI 短片角色 > [15:28](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=928s) 我没有指控你，我只是在听你说话的声音。十点零七分，有人看到一个身形与你吻合的人站在她的窗外…… — AI 短片角色 > [19:05](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=1145s) 为什么不教玩家弹真正的吉他呢？ — 杰克·麦考利 > [19:05](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=1145s) 玩家并不想学，而且吉他太难了 — 《吉他英雄》团队 > [19:05](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=1145s) 总不能开一场演唱会，台上就孤零零放着几个扬声器吧？ — 杰克·麦考利 > [20:37](https://www.youtube.com/watch?v=7ejoDa3u8pw&t=1237s) 写出真正幽默爆笑的剧本，必须由一个本身就风趣幽默的人来完成 — 杰克·麦考利 ## Mentioned 杰克·麦考利 (Jack McCauley), Oculus, 加州大学伯克利分校 (UC Berkeley), 《吉他英雄》(Guitar Hero), 动视 (Activision), Facebook / Meta, 罗技 (Logitech), 道格拉斯·恩格尔巴特 (Douglas Engelbart), 《终结者2：审判日》(Terminator 2: Judgment Day), 《至尊神探》(Dick Tracy), 《剪刀手爱德华》(Edward Scissorhands), Sun Microsystems, US Windpower, 兰德公司 (RAND Corporation), 伯克利数学科学研究所 (MSRI), 约翰·纳皮尔 (John Napier), 《侠盗猎车手5》(Grand Theft Auto V), Rockstar Games, RAGE 引擎, 《毁灭战士》(Doom), 《Unrecord》, 约翰·卡马克 (John Carmack), NVIDIA, PlayStation 2, PlayStation 4, 小罗伯特·唐尼 (Robert Downey Jr.), Metallica, 荷马·辛普森 (Homer Simpson), Netflix ## Video description Jack McCauley - Co-founder Oculus, Board of Trustee at UC Berkeley, UC Berkeley Explore how AI tools are utilized in scriptwriting, editing and visual effects to enhance the filmmaking process. [Source](https://www.youtube.com/watch?v=7ejoDa3u8pw)

Source quotes

[0:04] 动视引荐人: 你们需要硬件工程师、数字电路专家还是软件工程师？我认识一个人可以全包。 [0:04] 杰克·麦考利: 谁会买这种东西？戴着那玩意儿到处走，怎么可能！ [0:04] 杰克·麦考利: 我就是个工程师，不负责市场营销，也不管收购谈判。 [0:04] Oculus 团队管理者: 你猜怎么着？Facebook 刚刚以约32亿美元收购了我们的业务！恭喜你，你现在的钞票堆起来能一路叠到月球了！ [13:20] 杰克·麦考利: 当你的母公司靠卖广告赚钱、并且文化上排斥暴力内容时，你该如何打造那种能真正卖爆头显的爆款内容？这极为困难。 [15:28] AI 短片角色: 你来这儿不是为了怀旧。 [15:28] AI 短片角色: 我来是因为你答应谈谈。谈话可不犯法。昨晚十点到十一点之间，你在哪儿？ [15:28] AI 短片角色: 在家里。一个人。 [15:28] AI 短片角色: 你是在指控我什么吗？如果是在指控，我要求见律师。 [15:28] AI 短片角色: 我没有指控你，我只是在听你说话的声音。十点零七分，有人看到一个身形与你吻合的人站在她的窗外…… [19:05] 杰克·麦考利: 为什么不教玩家弹真正的吉他呢？ [19:05] 《吉他英雄》团队: 玩家并不想学，而且吉他太难了 [19:05] 杰克·麦考利: 总不能开一场演唱会，台上就孤零零放着几个扬声器吧？ [20:37] 杰克·麦考利: 写出真正幽默爆笑的剧本，必须由一个本身就风趣幽默的人来完成

Source quotes

[0:04] a stack of dollar bills from here to the moon [2:21] The Mother of All Demos [6:49] Napier's bones [15:28] Suspect: You didn't come here for nostalgia [15:28] Interrogator: I came because you agreed to talk. Talking isn't a crime. Last night, between 10:00 and 11:00, where were you? [15:28] Suspect: At home. Alone. [15:28] Interrogator: people pause when they're remembering [15:28] Jack McCauley: You could have a virtual star, a person that doesn't even exist in the world, and he can win an Oscar. [18:02] Jack McCauley: writing funny shows is done by a funny person

Original transcript

[0:04] I've been very fortunate in my career and a lot of it came down to hooking up with the right people [0:10] and I worked in entertainment for 40 years. I retired in a while back after we sold our [0:17] business to Meta. I started a company with another couple people for VR headset and um [0:27] against my best guesses Meta acquired us. Facebook acquired us in 2015 after a 3.2 billion um and uh [0:38] The interesting note is before I launch into this is that I didn't know about any of those backroom [0:44] negotiations. I'm just an engineer guy, right? I'm not involved with marketing and you know, [0:48] negotiating with potential acquirees acquires rather. And I was in my hotel room in China cuz [0:55] we built our headsets. A VR headset is a virtual reality headset. You've probably heard of it. um [1:01] You play video games in it, which is really in my genre of work. um I got a a call from the [1:08] the boss. He says he goes, "Guess what?" And I go, "What?" It's 3:00 in the morning. He goes, [1:14] "uh Facebook bought our business and uh congratulations." And he said, [1:18] "You have a stack of dollar bills from here to the moon." Well, it wasn't quite that much, but [1:23] I've been very fortunate and I got introduced to the founder of excuse me, the CEO of Oculus via [1:31] a connection I had from working on Guitar Hero. I was a head engineer on Guitar Hero, a video game. [1:37] Another one I I I thought Oculus wasn't going to make it, right? I mean, I said, "You know, [1:42] who's going to buy that? They're going to walk around with those things." I mean, come on. [1:45] Kind of bright there actually at the end, but um I got introduced to [1:50] these guys at Oculus. A young guy is 18 and the boss um via connection I had at Activision [1:57] that I maintained. I call people, not so much anymore. I don't work anymore. [2:02] But I always call people uh once a year just to stay in touch with folks. And they go, "Well, [2:08] you know, you need a hardware engineer, a digital guy, a software engineer? I know a guy that can [2:13] do that for you." And so, I primarily worked in entertainment. So, let's let's get into this here. [2:21] Um I'm an inventor. Uh I have been since I was a little kid. I got a scholarship uh for building a [2:29] toy you know when I was nine. And I still do that stuff. I enjoy it. I like programming. Um my son [2:36] went to work for me briefly as a programmer. He's really good at it. He hates it. I don't. [2:42] Uh I don't know why he doesn't like it. He says it's boring. He's probably bored with it. He's [2:46] smarter than I am, so. But I've been doing this stuff for my whole life. I still do it to this [2:50] day. And at the end I'll tell you about something I'm working on but if we have time. So, that's [2:56] a scrolling wheel mouse. Again, connections. I work with Doug Engelbart who's a famous engineer [3:03] when I was at Logitech in the '90s. And uh Doug is the the mother of all demos guy and he invented a [3:10] computer mouse. I get this thing computer mouse is back then didn't have a scrolling wheel. And I go, [3:15] "How can I make it better?" Well, there's no windows back then. So, we were using DOS and I [3:21] I endeavored to put a scrolling mouse wheel on it. Now, it didn't end up being used the way [3:26] I intended. It's used to for windows for various things, moving menus up and down. I wanted it for [3:32] video games to go in and out. You have an X and Y axis. I needed another axis. Wasn't used that way. [3:38] And I didn't patent it. I had no idea about patents. Uh I didn't really know about those [3:43] things and sounded complicated. Didn't care. So, they copy it. Great. Well, they did copy it. [3:48] After that, I patented every just about everything I did. and mostly the company saw that it was [3:55] judicious to do so I would do so. Uh chief engineer co-founder of Oculus VR very fortunate [4:04] because of the connections Activision turned me on to those guys. I worked at Activision Guitar Hero. [4:09] Go talk to this dude. He came in and we talked. We filmed a Kickstarter video at my facility. We [4:15] did all the marketing PR and so forth internally with just about five people. [4:21] Uh I worked on feature films also. I worked on Terminator 2, Dick Tracy, [4:25] Edward Scissorhands as a technical contributor and I worked on audio. Uh multi-channel audio stuff [4:31] was what I was doing digital audio. And it got an Oscar. I didn't get the Oscar our group did. Um [4:37] my job there was working on a Sun 1 workstation. I love Sun workstations. I don't know if people here [4:43] are programmers but it's my favorite machine. Has X windows you could shell out and and compile your [4:49] source code using uh a Unix window and I just thought it was genius level stuff. [4:55] But when we did the work I'll show you a video in here in a minute on Sun 1 workstation took [5:00] 26 workstations on Terminator 2 to generate the visual part of it. It was all rendered mostly [5:07] and mix of things but mostly rendered. This is back you know, 1990 time frame when we didn't have [5:14] GPUs. They didn't come out. A GPU is a graphics processing unit that's used for video game stuff. [5:20] A A TPU which is a tensor processing unit is basically a GPU and TPUs used for AI. I endeavored [5:26] in the 90s also to learn all about an AI because I thought it was an up-and-coming thing. [5:31] So I took courses through the university UC Berkeley extension on the math part is [5:36] mathematically intensive different types of ways of doing it. So I'm kind of got a pretty good [5:40] background in that although I never used that in my career. There just wasn't horsepower to do it [5:45] way back in the day. Uh I worked in entertainment. I mentioned that already. Also other things too. [5:51] I'm a kind of a green energy guy and I worked at US Wind Power briefly. They're They make [5:57] windmill stuff and they're located in Livermore, California. From California. Uh I'm a professor [6:03] UC Berkeley. I teach energy uh green energy stuff there. EV stuff. I know the EV platforms very well [6:10] and the technology behind them. I did this all on my own. I learned this stuff. It has nothing to [6:14] do with my job. And somebody said, "You go work at Tesla." I don't know. Thanks. Work at Tesla? [6:20] Can you imagine? What would I do there? I mean, they got all the great guys there. So Uh and I [6:25] work I've got them I'm work RAND Corporation a research contributor there and Math and Science [6:31] Research Institute and during Berkeley. It is a math re- post-grad math research institute and [6:38] uh I also um I I'm alumni of UC Berkeley. I'm involved very much involved with the trustees [6:44] and so forth. Uh I build cars. I'm a car nut. Um and I'll get off me for a minute and we'll go into [6:50] the stuff I want to talk about. This is a Fiat 500 I put a motorcycle engine in. Um and we had [6:56] a shop with all of his CNC machinery and lathes and all that stuff. It I don't have it anymore. I [7:01] got rid of it. But built these cars there. Um not so much the silver one there, but the other two. [7:08] Um so uh the drone stuff in 2019 I came out with this machine. I could see that coming. These cheap [7:15] drones from China and other places with explosives attached on flying into people. You've seen that [7:21] probably those videos in Ukraine. It's horrible. And so I endeavor to build some but they got to [7:26] be able to jam them because there's a a guy on the ground with a remote control guiding it or [7:30] with you know, an iPad. And I said there's got to be a way to jam signal going from the controller [7:35] to the drone and then we were successful. This has a spectrum analyzer in it. So [7:40] and it's all cobbled together from various parts I bought. I It's not like I had to do [7:43] much engineering in it except the antennas and stuff, but um that is uh in 2019 and um that's [7:51] the drone blaster. I did some other stuff, too. Um Again, I didn't invent it. Guy I didn't patent [7:56] this. I don't think it's a patentable a jammer. It's not really not legal, either, I found out. [8:02] Uh that's a calculator I built. Um I was demoing this the other day for someone [8:07] and I know how to run it. I just got out kind of got out of the way and let them try to figure [8:11] out. It's really simple. Does huge numbers, 10 to the 20th, which is a gigantic number, [8:16] universe-size numbers. And this is from 1500s, late 1500s. Uh Jonathan Napier, mathematician. Um [8:24] but I built one. I thought it was so cool, I got to have one of those things. They're available, [8:27] not but not They're kind of like smaller and kind of made out of plastic. I wanted a real one. [8:33] Does multiplication and division and cube roots, square roots, stuff you don't probably [8:38] know much about, but I'm not going to dive into that too much, but it does really good [8:42] fractional mathematics, which is doesn't have a decimal point. Didn't have decimal points back [8:45] then. Have now. It's all fractions. Fractions are great because they're precise and decimal points [8:51] have an order of precision to them, so we use decimal points and not fractions so much. Okay, um [9:00] this is a video and I'm going to try to play this and I might have to ask for his help there because [9:04] I don't know how to run this screen remote, but I worked on this film and There you go. At least I I [9:14] stopped the sound on it because it's swearing and I don't want like good people you know, upset. [9:24] That was done on the Sun workstation, 20 of them. And that took [9:30] It took 6 weeks to render that one scene. 6 weeks. Yeah. That tech now can be done on the fly. [9:37] The rendering can be done on the fly in a video game, which is where I work. [9:41] I wasn't a graphics programmer in video games. I mostly did the hardware end of things, like the [9:47] computers and stuff like that in programming. Programming at the low-level programming. [9:52] I worked on the PlayStation POSIX operating system stuff like that. I didn't get involved. [9:57] I'll explain something. I'm going to talk about Grand Theft Auto. Okay, Grand Theft Auto V. [10:02] The work on Grand Theft Auto V, which was launched in 2014. I think I got that right. [10:07] Uh it took forever to do. They had to rewrite the rage game engine to get it to run the way [10:12] they wanted it to. All of the buildings and structures that you see in Grand Theft Auto V, [10:17] anyone familiar with GTA V? There's got to be some people. Were done by hand. They took a building [10:25] and took pictures of it. Group people went to LA and they sat down with AutoCAD and recreated the [10:30] building, including texturing, and colors, and so forth. Laborious work. That took forever. And uh [10:38] they spent $256 million on that game development. Unheard of sums back then cuz Guitar Hero we spent [10:46] m- the company bought us Activision bought us for $120. Um but we probably [10:51] had a low-budget company there. There was two guys, two brothers who started Guitar Hero, [10:56] me and about six or seven other people. We had a studio doing the game content part, but we [11:00] were doing all the hardware. To Now, uh Grand Theft Auto is a completely different story. Um [11:07] it's Rockstar Games. They're in the UK. Take-Two Interactive is the publisher. Uh it took forever [11:13] and it's all it's it's a great game. It's a it's a best-selling game. So, $256 million to make it [11:20] and in 10 days they made $2 billion after launch. It's a smash hit. This is part of my talk. Um [11:29] I can create a building, a an asset, a video game asset with AI by describing in language [11:37] what I want the building to look like. I want it to be a brick building. I want it to have broken [11:41] windows. I want graffiti all over it and it just does it. That was all done by hand and that tech [11:46] didn't exist when GTA was started. What's it going to How's it going to impact entertainment? This is [11:52] This is going to be mind-blowing in my view. Video games will not take eight years. They'll take [12:01] a year or six months to do. You have a franchise, let's say Grand Theft Auto, six months a month. [12:08] I've got a team of people, artists with a with programming background as in [12:12] in CAD and they're creating this stuff. Those guys are going to be gone. They will be. [12:17] This is the impact that AI think on go you know, globally in the terms of employment will [12:24] have a big impact on on employment, particularly artist, uh market research people. I just go to [12:30] Google and find out what I want about marketing and then I found out that the facts about GTA [12:36] before I got here today. I just I forgot all that stuff. I just looked it up. This is the [12:42] power of AI, the research power. It's It's going to be really substantial you know, significant. [12:48] These jobs of programmers so I have a friend who's got a a friend of ours has a kid [12:55] and she just graduated and she's a computer science major and can't find a job. Can't [13:00] find a job. Computer science major It was unheard of 15 or 20 years ago. You'd get a job you'd get [13:05] multiple offers before you even graduated. And and what How that's going to play out is not clear [13:12] uh in terms of employment um but for content creation, I think it's going to be substantial. [13:20] Uh I'll go to the next slide. Oops, did I go backwards? I certainly did. [13:27] Okay. This is 1993 when Nvidia launched uh in '93 and that's a video game called [13:34] Doom. One of the guys that worked at Oculus was creator of Doom. This is 2023. A few years back, [13:40] this is Unrecord. I made this video a couple years ago. Uh but if you could play that for me, please. [13:50] Unrecord is a good game. It's a first-person shooter FPS. [13:53] Are you familiar with that? It's a shooting game. [13:59] One of the problems we ran into with Oculus is what to do about content. We got acquired [14:04] by a company that doesn't make content. They sell ads. How do you do this kind of content [14:10] which would make the headset sell? How do you do that kind of content when you got a company [14:14] that makes ad and they have a culture that's like anti-violence so forth. Really difficult. [14:21] That Comparing contrast these two. 1993 Doom was written in assembler, [14:27] assembly language is a low very low-level programming language. [14:31] And in 2023 was written with a game engine. You can get commercially. You [14:35] can rent a game engine if you will to pay royalty to use someone else's game engine. [14:44] I don't know why it keeps doing that. So Nvidia's 1993 [14:49] uh this is the floating point operations per second. Just to give you an idea of [14:53] where the number of multiplications and divisions it can do. It's very, very high. [14:58] And 2002, the PlayStation 2, it's 6.2 billion floating point operations per second. Don't worry [15:04] about that. If you don't understand what that is, it's just Comparing contrast the numbers. [15:10] Uh iPhone 14, a little bit over slide. You can see where iPhone The iPhone has as much [15:15] processing power for graphics as a PlayStation 4. The Pro Max 17 has that kind of horsepower. [15:23] You can't run it like that. It'll burn the battery up and gets very hot. [15:28] I'm going to talk about this. Please don't play this. Talk about this. This is a film that a [15:33] person I know did. Here's my prediction about Hollywood, right? You know, they have a set in a [15:39] studio now and film editors and makeup people and blah blah blah, a whole bunch of people employed [15:46] making movies. That That industry is in trouble, by the way, in LA. And I talk to my friends [15:51] occasionally who are still working that business. A lot of studios have left. Reasons for that are [15:57] I don't want to get into that too much, but they've left. And but now I can create a movie [16:02] with artificial intelligence. And this was Here's an example. I'll go ahead and play that, please. [16:12] It looks AI-ish. I I grant you that. That's Robert Downey Jr. This is rendered just with AI changed. [16:20] You didn't come here for nostalgia. [16:26] I came because you agreed to talk. Talking isn't a crime. Last night, [16:34] between 10:00 and 11:00, where were you? [16:41] She did this. She's about 30 years old herself. [16:46] At home. Alone. People pause when they're remembering. [16:59] Are you accusing me of something? Because if you are, [17:06] I want a lawyer. [17:09] I haven't accused you. I'm just listening to how you speak. [17:14] At 10:07, someone was seen standing outside her window. [17:20] I think that's enough. I'm almost out of time. Thanks. Someone's matches your height. [17:25] You said you were just talking. Yes, I am talk Thank you. Um that's my prediction with the [17:34] entertainment industry TV films Netflix is that's all going to be done like that. You [17:38] could have a virtual star, a person that doesn't even exist in the world and he can win an Oscar. [17:45] You know, that's that's kind of where I'm headed with this talk here. Uh I think the impact of [17:50] this kind of technology is going to be profound and uh just as the GPU is profound in video games [17:57] and the the tools that came about video game engines engines the thing that runs the game. So, [18:02] have a few minutes for some questions if anyone has any. Thank you for sharing your experience. [18:08] If Hollywood your opinion is going to be wiped out, what do you think happens to the creativity [18:15] that lifts within the industry and and coming out of the industry? What would they transition to? [18:22] You have to have someone who has to create the plot, the dialogue, [18:26] all of those things. Artifacts have to be created by a person to keep to drive your interest. Now, [18:32] that's my prediction is that you'll have a uh a key people that will be involved. Um [18:37] the images of stars and so forth will still be relevant. The personality profile, you know, the [18:44] National Enquirer stuff about somebody that probably won't exist, but I think it I think it [18:49] what's going to happen is that we'll have people like sort of like Homer Simpson that you know, [18:54] you know Homer Simpson, right? They're that are rendered and played for you. So, that's [19:00] kind of my prediction, but I think it's going to have a significant im- impact on entertainment [19:05] and particularly where I work. I have a son who's currently studying to become I'm composer. [19:11] What do you think is going to happen to music and specifically soundtracks for movies and [19:17] video games? Do you think it's all going to go the same way where it's all going to be AI generated? [19:25] Oh, well, in the time I have left, I'll tell you a little talk a little bit about Guitar Hero. [19:30] And then I hope that answers your question. Guitar Hero 1 which launched super low budget, [19:36] about 10 people working in a studio in Santa Clara, California. Uh for the hardware, [19:42] the game was written somewhere else. We could not get the master tracks from the artist. You [19:46] It's a guitar game you play along with Metallica or whatever. We couldn't get them and that because [19:52] uh they thought it was a dumb idea. Kind of where I was with it, but [19:56] I said, "Why don't you teach them how to play a real guitar?" Well, they're not interested. [20:00] Uh and they can't learn a guitar from a video game. It's just too difficult. So, [20:05] with Guitar Hero 2 and the subsequent successes, we had master tracks and so [20:09] forth from the original artist. Do I think that music composition uh that's a a craft. Um I don't [20:17] know that that will happen. I mean, how do you do How do you concerts, right? You're going to have [20:21] What is it? Who's going to be there? Speakers? Kind of a little bit like that now, but I I don't [20:26] see it replacing or the artistic component like in music and so forth, but I Who knows? All right. [20:32] I didn't answer your question, though, did I? We have time for one more question right here. [20:37] Hi. Uh thank you for the talk. It was great. Um I uh I do a lot of improv comedy in Chicago uh and [20:45] work with a lot of actors and there's this really It's It's interesting cuz like in my day job, [20:51] people are very like high adopters of AI, but in this community specifically, there's a deep like [20:55] hatred and rejection of AI and um I'm wondering, you know, particularly around like AI created [21:02] art and and media and stuff like that. So, I'm wondering if you think that that's going to be a [21:07] barrier to uh like seeing more stuff like this or if you think that's just going to like fade over [21:12] time. It's like an immediate knee-jerk reaction. You know, most comedy shows are done live. [21:18] You know, this with an audience there. I don't see it think you can replace it with a real person, [21:22] but I don't know. Be honest with you what will happen and then it that [21:26] writing funny shows is done by a funny person. So how are you going to replace that with AI maybe? [21:33] I'm talking mostly in my talk about the graphics part of it and set design and set creation and [21:38] did being described into a computer. And this and the computer creates the set, [21:44] the lighting you want, the textures and so forth, the camera angles, that kind of stuff. I don't [21:49] I'm not sure that you can replace comedians with a with AI. Perhaps you can at some point. Not sure.