OpenAI to Ask Future AGI to Create Investor Returns, Altman Says

奥特曼称OpenAI将让未来的AGI为投资者创造回报 OpenAI目前没有商业收入模式，并计划在通用人工智能建成后指示其为投资者创造回报，以此证明设定100倍利润上限的合理性，防止任何单一实体攫取未来的所有经济价值。 萨姆·奥特曼回顾了他在Y Combinator的任期，剖析了创业者原型、10分钟面试流程，以及Demo Day后缺乏自律的风投所带来的后续风险。他解释了自己转投OpenAI的原因，认为构建安全的通用人工智能（AGI）其意义远超任何传统科技创业公司，也是人类解决气候变化等复杂危机的最佳机会。针对机器学习社区对GPT-2分阶段发布的强烈反对，奥特曼作出了直接回应，为OpenAI非常规的100倍利润上限模式进行辩护，并概述了包括自动化网络安全攻击在内的近期AI威胁。最后，他探讨了OpenAI为何不排除与军方合作、全球人才如何驱动YC的良性循环，以及为何改革美国移民制度对技术进步至关重要。 萨姆·奥特曼从Y Combinator转投OpenAI，专注于构建安全的通用人工智能（AGI），他认为这可能创造出超越科技领域以往任何成就的价值。 (24:42) OpenAI重组为利润受限实体，允许首轮投资者获得最高100倍的回报，同时将所有盈余价值导向公共利益，以防止任何单一群体攫取未来价值的光锥。 (44:31) OpenAI透露其目前营收为零且没有具体的商业模式，而是依赖与投资者的一项默契协议，即一旦AGI研发成功，将由AGI自行设计出投资回报方案。 (30:30) 奥特曼为不公开完整GPT-2模型权重的决定辩护，表示AI的进步严重依赖原始算力的扩展，尽早针对自动化虚假信息采取安全防范措施总比为时已晚要好。 (34:37) 除了长期的AGI安全，奥特曼指出AI迅猛发展所导致的灾难性网络安全漏洞是他当前最紧迫的近期运营担忧。 (41:28)

[27:47] Sam Altman: My overall belief is that the best way to get climate change solved is to build really strong AI first I think it's that hard of a problem. [44:31] Sam Altman: The problem with AGI specifically is that if we're successful, and we tried, maybe we could like capture the light-cone of all future value in the universe, and that is for sure not okay for one group of investors to have. [15:00] Sam Altman: I think people that run companies are either usually they're either a good leader or a good manager and I think I am a good leader and a bad manager. [15:00] Sam Altman: I think I think further out into the future than most people do and I think I appreciate exponential curves more than most people do. [50:02] Sam Altman: Fix immigration. If I could pick one policy that I think would help the global economy, and particularly the US economy, it would be that.

[0:00] so now I am very happy to be announcing [0:04] Sam Altman who is really like one of a [0:07] handful of people who doesn't really [0:08] need an introduction in Silicon Valley [0:09] but I'll say just the same that he is [0:12] the YC chairman and the president of [0:15] open AI Sam come on over here [0:17] thanks on the ground thank you Sam [0:22] so I was just delighted when Sam agreed [0:25] to do this I had the pleasure of [0:27] interviewing him a couple of years ago [0:28] at disrupting it was very fun so I [0:30] remember yes what's nice [0:31] Thanks having me back um so I was just [0:34] telling Sam that although I feel like [0:36] it's maybe boring for him I I always [0:39] love hearing him talk about founders in [0:40] fact I was um I run slowly and I listen [0:44] to podcasts sometimes and I was [0:45] listening to your recent interviewer [0:48] maybe was in February with this [0:50] Economist Tyler Cohen who is pretty well [0:52] known [0:52] he works at George Mason University [0:54] outside of Washington and he had a long [0:57] sit-down with the Sam and they were [0:58] talking a little bit about you know [1:01] pattern recognition [1:02] Sam's obviously seen a lot of founders [1:04] there's a lot of founders here so one [1:06] thing I thought you said that was really [1:07] interesting was and you said you weren't [1:10] even necessarily happy with this [1:12] discovery but that you felt a lot of the [1:15] best founders you'd met come from a [1:18] certain background they are upper-middle [1:21] class but not from like super rich [1:23] families maybe not from middle class [1:24] certainly not from poorer and so I just [1:27] wondered why is that do you think yeah I [1:30] mean I obviously think it's a real shame [1:32] that there aren't more successful [1:33] founders who grew up a super super poor [1:38] or super rich for that matter you sort [1:41] of need enough stability and like [1:48] blessings in life to even sort of have a [1:51] shot at compounding through your [1:54] education and early career long enough [1:55] to have a to be able to be a really [1:58] successful founder you know I think I [2:01] think that I have said before and [2:03] fundamentally believed to be true is [2:05] that the two great gifts I got in life [2:07] or a great education and the love of my [2:09] parents speaking of my mom is here yeah [2:13] I didn't wasn't planning on that I got [2:17] very lucky and got the best mom ever [2:19] which probably had more to do with my [2:21] success than anything else honestly I [2:22] mean that but I think if you don't grow [2:26] up in a stable environment with a great [2:28] education and loving parents and the [2:31] ability to you know like afford [2:34] everything that it takes between [2:36] starting your life and starting a [2:38] company it's very hard to have a real [2:39] shot at doing that it's interesting it a [2:42] little bit echos something that Doug [2:44] Leonie of Sequoia Capital told me during [2:47] a separate interview last year he was [2:49] saying you know my children will be [2:51] deprived in one way or disadvantaged [2:54] which is that they'll never be as hungry [2:55] as I was as a you know this young [2:57] immigrant from Italy when he was a kid [2:59] obviously he's done very well for [3:01] himself [3:02] you also said which I thought was [3:04] interesting and I am aware of this but [3:05] when applicants to YC used to apply it [3:09] was like an hour long process [3:11] yeah and then it got down to ten minutes [3:12] and you said you thought I could get [3:14] down to five minutes [3:15] so I think what I meant to say is I [3:19] don't think it's like socially [3:20] acceptable to make it five minutes yeah [3:23] I think that'd be pretty rude but I [3:24] think we could make decisions that are [3:27] 90 percent or 95 percent as good as we [3:29] make with the 10-minute interview in [3:31] five minutes [3:32] I I interviewed for YC in the first [3:34] batch ever it was 40 minutes or maybe 45 [3:37] like there's some did we don't quite [3:38] remember which of those but one of those [3:40] two and at the time Paul and Jessica [3:45] thought that was like as short as it [3:47] could be and we have found to our great [3:50] surprise that although we make many [3:53] other mistakes and our interview process [3:54] at the other stages like once we get the [3:56] founders in the room once we get to sit [3:58] down and sort of talk to them we don't [4:01] mess that up too often and it doesn't [4:03] take much data so many of the skills [4:05] about what it takes and so many of kind [4:08] of the sorts of thought processes and [4:10] the way people react to being pushed on [4:11] their idea are evident in the first few [4:15] minutes you talk to somebody that one [4:18] way we could scale YC but I think might [4:21] work is to just figure out how to do [4:23] 25,000 in-person interviews [4:25] you're right you said it before I would [4:27] since Donna Shing but I also just [4:29] wondered beyond the thought processes [4:31] and it sounds like maybe that's like a [4:32] little bit adversarial where they come [4:33] in and you're gonna try to poco zone or [4:35] no not really [4:36] you know even the founders that we don't [4:38] fund which is a lot they say it's really [4:40] fun and it's definitely fun for us an [4:43] interview if it's going really well ends [4:46] up feeling like the first set of office [4:48] hours we have with the company where we [4:49] talk about sort of their plan to sort of [4:51] make this great product or service or [4:54] whatever and even if we don't end up [4:58] funding you it's it's usually still fun [5:03] like there's this cottage industry [5:06] that's popped up like college admissions [5:08] for YC and so like we can usually tell [5:12] who's done it to it doesn't help but if [5:16] it's going really well it's not what [5:18] people expect it is a like let's talk [5:21] together about how to build this company [5:22] into something really big and see if we [5:24] can figure out how to do that well one [5:26] thing Paul said years ago I think maybe [5:29] even before you were president I saw him [5:31] talking about meeting with founders and [5:33] one thing that disqualified them [5:36] immediately for him was if he saw one [5:38] founder talking and the other founder [5:40] rolled their eyes I mean it's such an [5:42] obvious like you know display of disdain [5:44] I guess but that's a little bit more [5:46] obvious if you're looking for it are [5:47] there any sort of physical clues that [5:49] you're looking for um I mean contempt is [5:51] bad like that's that's true you know [5:56] people think we have all these like [5:57] secret things that we look for and yes [6:01] like open contempt as in any [6:03] relationship is like that's a sign for [6:05] us to like walk away slowly [6:07] other than that no there are no like [6:10] secret physical things we're looking for [6:13] and we say all the time like what we [6:18] look for [6:18] like fortunately it's really hard to [6:20] fake one of the things that I look for [6:23] the most is someone who's a clear [6:26] communicator and that's good because if [6:30] you're not it like I can tell someone [6:31] hey that's what I'm gonna be evaluating [6:33] you on and if you're not a clear [6:34] communicator that one's really hard to [6:35] fake and yet to be an effective CEO like [6:39] 95% of your job is effective [6:41] communication and evangelism and telling [6:43] the story of the company to people [6:44] you're trying to hire and retain and [6:46] sell to and get pressed from and [6:47] whatever else so there's a lot of things [6:50] like that where we just tell people what [6:51] it is like if you cannot clearly [6:53] communicate an ambitious vision will [6:55] probably not fund you and you can't fake [6:57] us out on that I heard Dave Goldberg the [7:03] founder of SurveyMonkey talked about [7:04] that once that it's salesmanship which i [7:07] think is another way of saying good [7:08] communication is so underrated in [7:11] entrepreneurship I did wonder since this [7:13] is sort of always of interest to a lot [7:16] of people here does age factor into your [7:17] calculations or well they have to be 18 [7:22] know like there was one batch where we [7:26] funded they don't as soon as I said that [7:28] I realized in one batch we funded a [7:29] fifteen year old whose parents signed [7:31] documents for him and also I think [7:34] someone who was like 78 or 80 yeah the [7:37] oldest the oldest and the youngest YC [7:39] founder ever in the same batch you know [7:44] like we do I think it's no secret that [7:47] we've made like most of our best [7:50] investments have been in people that are [7:52] sort of in their you know mid-20s to [7:55] mid-30s [7:56] but there's exceptions on all sides of [7:58] that and over time I expect there to be [8:00] more what we do care about is for [8:03] whatever circumstances in life you were [8:05] born into and have had and for whatever [8:07] age you are you have excelled at every [8:11] step of the way you know whatever hand [8:13] life tell you to play you have done as [8:15] well as anyone we can imagine and so we [8:19] grade harder unlike a 40 year old and a [8:21] 20 year old but that's probably the [8:26] difference what can I say one more thing [8:30] about Lee um I think when many people [8:33] say age they mean it as a shortcut for [8:35] someone who's willing to like work super [8:37] hard and I think it is true and I think [8:40] it's a real disservice when people [8:42] pretend like you can be a founder of a [8:44] company and not have to work extremely [8:46] hard and make it at least for some [8:48] period of time like the priority in your [8:50] life but [8:52] there are people of any age that are [8:53] willing to do that in people of any age [8:54] that are willing not to and I think [8:56] maybe that was sort of part of the [8:58] question or at least people off of me [9:01] somebody's got hard stops you know this [9:03] is gonna be sort of problematic on their [9:05] founder journey [9:06] it's definitely people do it but we try [9:11] to really be upfront about how hard it [9:12] is because the only time that I've heard [9:15] people feel bad about doing YC was when [9:20] they feel like you weren't honest with [9:21] me about just how much I was gonna have [9:23] to sacrifice in my life to do this and [9:25] so I think it does us no good and in [9:27] fact is actively hurtful us being like [9:30] the ecosystem of of Silicon Valley to [9:32] pretend like you can start a startup [9:34] that is going for hyper scale and not [9:37] have to work extremely hard one of the [9:40] worst things that can happen to you is [9:42] that your company succeeds you know when [9:44] you're really young yeah yeah it takes [9:46] over your life to a degree that people [9:48] don't believe us when we tell them one [9:52] other thing he had said historically and [9:56] it's changed so much at YC over the [9:57] years was that you had to be lean he was [9:59] telling founders you know you wanted to [10:01] sort of get to ramen profitable now YC [10:05] companies are graduating and they're [10:07] getting millions of dollars because of [10:09] YC tens of millions tens of millions [10:12] great so um look I if I could control [10:20] the market which obviously the free [10:22] market is gonna do its thing I would not [10:23] have YC companies raise the amount of [10:26] money they raise or the valuations they [10:27] do I have thought more than I care to [10:33] admit [10:33] about like improbable schemes to reduce [10:35] that because I do think it is on net bad [10:38] for the startups even though in some [10:40] individual cases it turns out to be [10:41] really good it isn't it is this this is [10:45] gonna sound like it's not true I wish it [10:48] were not the case that just because you [10:50] got into YC means you get to raise [10:52] millions of dollars or tens of millions [10:54] of dollars at the end of YC without I in [10:57] my opinion enough diligence or [11:02] discipline around price because I think [11:04] that has down [11:05] affects it hurts the companies in their [11:07] next round a whole bunch of other things [11:09] but that is you know we do our stuff the [11:13] ecosystem adapts to us oh I do they [11:19] never listen [11:21] but you know the ecosystem is the [11:26] ecosystem I think it will all correct [11:27] over time like in any given year can be [11:32] frustrating but over time I think [11:34] capital just sort of sloshes around [11:36] seeking for the best return we created a [11:38] return for a while that was incredibly [11:40] unusual so money flooded into our [11:44] ecosystem prices that come up it should [11:46] normalize someone at some point I hope [11:50] but I also think like not only does YC [11:52] get really good companies we actually [11:54] significantly increase the number of [11:56] good companies in the world and so in [11:58] some sense this is a response to that [12:00] and that part I hope keeps going before [12:06] YC came along you if you wanted to start [12:10] a company you had to have like I've gone [12:14] to like a fancy business school I've [12:15] written this nice business plan have [12:17] like experienced managers on the team [12:19] and most importantly you had to network [12:24] which is like you know basically [12:26] shorthand for like be from the right CEO [12:29] economic class to have a chance or a [12:32] white dude or whatever to have a chance [12:34] at getting investment like VC's used to [12:37] brag about how they like wouldn't take a [12:40] meeting without an intro from someone I [12:42] trust [12:43] and I think YC has only ever had maybe [12:45] like five total innovations but one of [12:49] them is that anyone in the world could [12:50] apply and they do and sometimes they're [12:53] really great and we put in the effort to [12:54] find them and I think that plus another [13:00] one of our very few innovations which is [13:02] we don't keep any of our advice secret [13:03] like we just publish all of it we [13:06] encourage people to read it I think [13:09] those two things together have increased [13:10] the number of good companies in the [13:12] world that people decide to start them [13:13] to make it into the ecosystem it's [13:15] interesting there is so much [13:15] transparency I'm not sure what started [13:17] maybe did [13:17] start with YC but I think that's had a [13:19] huge impact okay so shifting to the [13:22] personal when you were made president [13:25] the sort of story is there was not even [13:28] anyone else being considered yeah look I [13:31] think all of these things get canonized [13:36] over time I that probably wasn't true [13:38] and like I know that has become the myth [13:40] but I'm sure they had other ideas if I [13:42] didn't say yes well I was just gonna ask [13:44] where you think you are stronger than [13:47] Paul and Jessica and where you think [13:49] maybe you're weaker I mean weaker in [13:55] almost every way like what what PG and [13:58] Jessica did with YC is a miracle that [14:00] has happened in the investment industry [14:02] like a number of times I feel I can [14:05] count on one hand the change like I was [14:09] basically handed a thing that all I had [14:11] to do was not screw up and I think I've [14:13] done like a good job certainly a better [14:15] job than expected but anyone would have [14:19] done a pretty good job [14:21] I like YC was there's this thing that [14:26] you can tell in a company when it is [14:29] gonna be successful and it's so much [14:32] more successful on the inside already [14:33] than people realize and by the time I [14:35] took over YC it was definitely in that [14:38] category [14:38] and so like any monkey could have done a [14:41] pretty good job and I'm proud of the job [14:44] that I did and I think I did a few [14:46] things extremely well and some badly but [14:49] it's definitely case where they get the [14:51] credit for this one and it's not false [14:53] modesty there's plenty of other things [14:54] in my life I would take credit for well [14:58] I did want to ask you in my vein of [15:00] corny questions what you do think you're [15:02] sort of biggest strengths and weaknesses [15:04] are generally wow this is like a bad job [15:06] interview I've never been asked that [15:08] question in real life before man so I [15:15] think I think people that run companies [15:19] are either usually they're either a good [15:21] leader or a good manager and I think I [15:24] am a good leader and a bad manager so I [15:26] think oh you wrote something mean about [15:29] me that I'm just remembering that well [15:30] as many [15:31] but one that just came to mind I didn't [15:35] like that article by the way just that's [15:41] fine you do you as the kids say no it's [15:45] cool I'll I might I'm a big County fan [15:46] I'm here I don't do many of these I [15:48] think you said something like I'm not [15:50] known for being a warm terribly warm but [15:53] I wanted to ask about this because [15:56] you've said you don't care what people [15:57] think about you but this was something [15:59] that I had talked to about people and [16:01] they said not in a disparaging way but [16:03] that Sam is a little bit removed and not [16:05] a terribly warm boss and you like pretty [16:08] quickly called me out on that so what [16:10] did I hurt your feelings or was it just [16:12] that you thought it was such a gross [16:13] mischaracterization both I think it is a [16:19] meme again these like stories get told [16:22] like I I think I am I am NOT the kind of [16:30] manage that wants to spend like a lot of [16:31] time with people in like daily [16:33] one-on-ones talking about their feelings [16:35] I'm much more interested in saying like [16:38] okay like there's this interesting thing [16:39] let's go figure out to get it done but [16:43] yeah I think that was a [16:44] mischaracterization [16:45] but in that something is true in that [16:47] direction I don't I'm not like a super [16:51] warm and fuzzy person my mom can vouch [16:54] that's probably true she thinks I am to [16:58] her and like three other people in the [16:59] world but it's a short list so yeah I [17:05] think a weakness is like I don't I am [17:08] NOT a like a woman fuzzy balls a [17:12] strength I would like I think I I think [17:17] I think further out into the future than [17:19] most people do and I think I appreciate [17:21] exponential curves more than most people [17:23] do and your seemingly and what your your [17:29] career sort of this underscores these [17:31] things but you've talked publicly about [17:33] the importance of support including your [17:35] your mother's you've talked about Paul [17:38] and Jessica I just wondered who else you [17:41] look up to the most [17:50] I'm not saying this just because she's [17:55] here but I know it's true like I have a [18:01] special relationship with my mom I [18:02] remember I just I've recorded okay it's [18:07] fine I think I can tell this story I [18:11] think like I I got when I was 17 or [18:19] something like that I got this this [18:22] thing called the like a full ride [18:24] scholarship to a sort of college that is [18:27] good but not as good as Stanford and [18:29] there's like four kids in my family and [18:31] I knew my mom had like a lot of like a [18:35] lot of college tuition in her future and [18:38] on the other hand I didn't like my life [18:40] dream to go to Stanford and do study [18:42] computer science and also I'm gay in [18:46] this other school is like in the south [18:47] and that my visit there was not perfect [18:49] and this was 2003 and my mom and dad but [18:56] my mom was sort of like the driver of [18:59] this conversation took me out to dinner [19:00] without my siblings there's a big treat [19:02] and sat down and said like you know look [19:05] like we've been saving for you to go to [19:06] college we want you to do this it is [19:10] like we're okay we're gonna make it work [19:12] like you should like we know that it [19:13] like you're a computer nerd enlike this [19:15] is your life dream and you should do [19:17] that and we'll figure it out and we'll [19:18] support you and like you don't have to [19:20] feel bad about this thank you and that [19:23] turned out to be a pivotal life decision [19:25] for me and there have been others like [19:28] that where my mom was like pushed me and [19:31] you know I think I get plenty of like [19:33] good business advice but it's like there [19:35] are a handful of people in your life [19:36] that I think you can depend on to give [19:39] you like good life advice and I've been [19:42] you know like yeah it's like your [19:46] parents your partner my parents my [19:48] partner and like a small handful of [19:49] other people that I just absolutely [19:51] trust peih-gee and jessica are in that [19:53] category for sure and there's other [19:56] people who give me incredible [19:57] fashion advice but you know that's [20:00] always like you get to some point in [20:03] your career where people are afraid to [20:05] offend you or say something you might [20:07] not want to hear and I am definitely [20:10] aware that I get stuff filtered and [20:12] planned out ahead of time at this point [20:13] so you it's like a limited number of [20:15] people so I wanted to ask to you about [20:20] your siblings there was this awful so I [20:25] think I mean are they all here your [20:27] brother one's an investor they've all [20:28] been here on and off one of them is will [20:32] come back here soon I think it's [20:33] traveling the world one of them lives [20:34] here and runs lattice lattice hq.com you [20:37] should all check it out series B just [20:40] raise the series B and my sister lives [20:41] in LA okay great and are you close to [20:43] them yeah I mean we all my brothers then [20:45] I free across two houses and many years [20:47] all live together [20:48] that's adults are they still living with [20:50] you uh no okay cuz a couple of years ago [20:52] there was this New Yorker article that [20:54] said that they were we're gonna live [20:55] with you temporarily and that it turned [20:56] into sort of a more yeah I had told them [20:59] in back in like 2013 or something that [21:02] they could come stay with me or 2012 [21:04] maybe while they found a place it was [21:06] like that kind of dealing I don't know [21:07] dad but it was really fun it was great [21:10] and like you know I we definitely have [21:14] the kind of relationship where they [21:15] still just drop in unannounced and it's [21:18] nice parents are the best um I did want [21:21] to talk to you so when we that article [21:23] in The New Yorker come out like pretty [21:25] soon before I had sat down and saved you [21:28] and I said oh I thought this article was [21:30] so charming and you're like you you know [21:32] that counts one of us and I didn't ask [21:34] what you hated it you know I reread it [21:36] for the first like three years ago than [21:38] it came out and I reread it recently [21:42] actually thought it was pretty good so I [21:47] was like oh my mom called me unisexual [21:51] and it I still don't know what that [21:53] means that was funny oh as a teenager no [21:59] I thought but I thought it was pretty I [22:00] thought it was pretty good [22:01] rereading it with more hindsight felt [22:05] pretty good I think you [22:11] any profile like that you'd never feel [22:13] like it gets across what you really are [22:17] and what you really care about and I [22:19] still don't it still feels like a [22:20] caricature but I read mine as this [22:27] person like wanted to do a book on me [22:28] it's something some person looked like [22:29] and I really don't want to do that which [22:31] I'm not doing and I but in the process [22:35] of that it was like well go read art [22:37] like the article was pretty positive so [22:39] I read it again and I thought it was [22:41] like okay and then I read the profiles [22:42] that like other tech investors got and I [22:45] was like man I got it [22:50] but it still it doesn't feel like me [22:51] yeah so I think that's what I didn't [22:53] like about it [22:54] I never been written about but I can [22:58] appreciate the sentiment but one thing [23:00] you said in the article that I thought [23:01] actually didn't I didn't pay attention [23:03] to when it came out but in sort of [23:06] looking at it began before this sit-down [23:08] I thought wow that was really appreciate [23:09] this you know before Donald Trump was [23:11] elected president you said democracy [23:13] only works in a growing economy yeah [23:14] without a return to economic growth the [23:16] democratic experiment will fail [23:18] you're probably as you wouldn't [23:20] acknowledge yourself is insulated as [23:22] anyone in Silicon Valley so like what [23:24] clues were you picking up on that so [23:26] many people weren't like you know I [23:28] definitely think I'm much less insulated [23:30] than most people [23:31] oh really um I I think like well I don't [23:37] read to that much which really helps I [23:39] think it's very easy in Silicon Valley [23:41] to get super isolated and it's insulated [23:43] and I think it's super valuable to sort [23:45] of travel and try to meet people and [23:48] stay in touch with people who are not in [23:50] this vortex and yeah I mean now it's [23:55] like old news but at the time it sort of [23:57] felt vaguely controversial to say most [23:59] people are not enamored with the tech [24:01] industry and what it's doing and I think [24:05] I think what's happening is the winners [24:09] are winning bigger but there's sort of [24:11] it's this narrower and narrower rally [24:12] and it's felt like that now for a while [24:15] it feels untenable to me part of the [24:19] reason that I've always been excited [24:20] about YC is that in some sense we're [24:22] able to rapidly expand [24:24] access and part of the reason I'm now so [24:26] excited open AI is this is like too [24:29] strong of a statement to make but [24:30] directionally I believe that only [24:32] something on the order of the impact of [24:35] general artificial intelligence can [24:37] restore global economic growth equitably [24:39] to everybody okay last corny question [24:42] before we get into open air well okay so [24:47] I thought I thought about you I was [24:49] again running slowly listening to a [24:51] podcast I love this podcast Conan [24:53] O'Brien interviews comedians and writers [24:56] he's so funny but it's very human in a [25:00] surprising way so he was talking to [25:01] Jimmy Kimmel and Jimmy Kimmel was saying [25:04] it was it was interesting he was saying [25:06] that his life is really a grind and that [25:09] you know from the outside looking in it [25:11] looks like he's got it made but he feels [25:13] sort of trapped and and sort of you know [25:17] it's his own making but he's worried [25:20] that if he quits he's not sure if he'll [25:22] be relevant he's not sure what how his [25:25] self perception will change and I just [25:26] wondered your sort of star soared so [25:29] high at Y Combinator I just wondered as [25:33] you're transitioning out or thinking [25:34] about leaving like was that a concern [25:36] for you did you feel like you're so [25:39] wrapped up in YC oh this would be I mean [25:41] I do care about two things I do care [25:46] super deeply about YC and I and I'm sure [25:48] I have like some more ego than I'd like [25:51] to admit in all this however the thought [25:54] of like being a little bit less [25:56] recognized feels quite appealing and [25:59] more than but more than that like I [26:01] really do believe as much as I would [26:04] like love to go back to a life of [26:06] relative anonymity I really do believe [26:08] the work that we're doing at open I will [26:10] like not only for our Clips the work [26:12] that I did it YC but the work that [26:16] anyone in the tech industry does and so [26:18] it didn't really I didn't have that [26:20] product if I were really just retiring [26:22] if I were like you know what I'm like [26:24] gonna go off and not work anymore [26:26] sure that'd be hard because I think like [26:28] work is so much of my identity but it [26:32] feels like this is just this like [26:36] significant this is like a this is like [26:38] a big project we're embarking on okay so [26:41] I think everybody here has heard of open [26:43] AI and I think people are still [26:44] wondering what is open AI exec sure we [26:48] are trying to build safe artificial [26:51] general intelligence so it is my belief [26:53] that in the next few decades someone [26:56] some group of humans will build a [26:59] software system that is smarter and more [27:03] capable than humans in every way and [27:05] that very quickly it will go from me a [27:06] little bit more capable than humans to [27:08] something that is like a million or a [27:10] billion times more capable than humans [27:11] and this is hard to think about but at [27:15] least in my own case once I accepted [27:18] that that was gonna happen almost [27:21] impossible think about anything else so [27:23] we're trying to figure out how we do [27:24] that technically make it safe and [27:27] equitable share the benefits of it the [27:29] decision-making with over the world and [27:30] so that when we get to this point where [27:34] you know like the von Neumann probes [27:38] leave earth and sort of spread [27:40] consciousness throughout the entire [27:42] universe until the heat death that we [27:44] build something that we're proud of how [27:47] are you so sure it's going to happen so [27:49] lots of researchers are sort of afraid [27:51] to even associate their name with a [27:55] timeline you're talking about decades [27:56] there was a story I think maybe in the [27:58] verge or Vox yeah I think there is a lot [28:02] of debate about the time frame and like [28:04] people getting these huge arguments [28:05] about whether it's 10 or 20 or 30 years [28:07] and even if it's 50 or 70 like for the [28:13] most important thing in human history to [28:15] be that close I think the timeline [28:18] argument is a distraction perhaps but I [28:23] mean I think there are people who aren't [28:24] even sure if it's ever gonna happen they [28:26] think it might might happen in 50 years [28:28] might happen in 20 years I just wonder [28:30] like what about applying that big brain [28:31] to like climate change which is right [28:33] here right now and like threatening to [28:35] kill us oh well III think one of the [28:38] projects that we will use the technology [28:39] we develop at opening I to do is solve [28:41] climate change and I think if humans [28:43] knew how to solve that unaided we would [28:45] hopefully have solved it by now [28:46] I argue we're failing at both this [28:49] and the policy.we I do have like a [28:54] couple of pet ideas about climate change [28:56] that I'm still independently spending [28:58] some time on but my overall belief and [29:01] we're encouraging companies to apply to [29:02] YC and I would love you to do it [29:04] my overall belief is that the best way [29:05] to get climate change solved is to build [29:08] really strong AI first I think it's that [29:11] hard of a problem I hope you're right [29:14] so tell us sort of on a more granular [29:16] level what's happening at Albany how [29:19] many employees do you have how many PhDs [29:21] I don't know the number of PhDs we don't [29:24] count that but the number of employees [29:25] is you know around 90 I would guess [29:28] something like that and I guess how much [29:30] sort of like pure research are you doing [29:33] we're still at a point where everything [29:35] we do is pure research we're not [29:37] building any systems to solve climate [29:39] change or be industrial robots all the [29:42] work we do is still trying to figure out [29:44] what are the components that we need to [29:46] build for general artificial [29:49] intelligence and so we did a lot of work [29:53] in the last in 2017 2018 with [29:56] reinforcement learning robotics video [30:00] games things like that we're doing a lot [30:01] of work right now with unsupervised [30:03] learning so we have this now text model [30:05] that can read the Internet [30:07] learn to predict the next word but not [30:10] with any suit not with any labels just [30:11] true on supervised learning and it feels [30:14] to me like it's starting to understand [30:16] and then there's some other components [30:18] that we have to build as well but we're [30:20] definitely in the stage of a research [30:23] company there are there are big pieces [30:26] of the problem that we don't know how to [30:28] do yet I just wondered you know I'm sure [30:31] people are attracted to the organization [30:33] because of your involvement but I just [30:35] wondered you know if it's if it's hard [30:37] to compete I'm sorry well hard to [30:39] compete and also hard to hire you know [30:42] if you're sort of if you can go to open [30:44] AI or you can go to I don't know if who [30:45] you consider your competitors if it's [30:46] Google or Microsoft or yeah it's really [30:48] hard I mean I think the most in-demand [30:51] people to hire right now are sort of you [30:54] know really great AI researchers and we [30:57] are badly outgunned financially by any [31:00] of those companies so I think [31:03] a mission that is appealing to people [31:04] you do feel like you're sort of working [31:06] on behalf of the world not on behalf of [31:09] a single tech company and you're sort of [31:11] trying to the thrill of discovery is [31:14] really exciting so it's fun to work on [31:16] that instead of sort of optimizing ad [31:17] clicks but even then like it's extremely [31:21] hard to successfully recruit someone [31:24] that Google desperately wants to hire [31:26] also so you know Google has business [31:31] units and it's got objectives and it [31:33] just seems like open AI is so sort of [31:36] amorphous but it is a business so I'm [31:38] just wondering like eventually is the [31:40] idea to kind of like license [31:42] technologies will you have customers [31:45] that you're going to be customizing [31:46] algorithms for them or how is it gonna [31:48] work you know the honest answer is we [31:50] have no idea we we have never made any [31:54] revenue we have no current plans to make [31:56] revenue we have no idea how we may one [31:58] day generate revenue we have made a soft [32:01] promise to investors that once we've [32:03] build this sort of generally intelligent [32:05] system basically we will ask it to [32:09] figure out a way to generate an [32:10] investment return for you [32:14] it sounds like an episode of Silicon [32:16] Valley it really does I get it I can [32:19] laugh it's alright but it is what I [32:21] actually believe is gonna happen so [32:25] there aren't many things like this where [32:28] they're so difficult to reason about and [32:30] so powerful that if you can build it [32:32] like you know there's that joke that I [32:34] actually think it's very funny which is [32:35] like step one is do this thing step two [32:37] is do this thing question mark question [32:38] mark question mark profit but like you [32:43] know I think if a guy is on the table [32:45] that's actually okay and if we have to [32:47] take like you know one one thousandth of [32:49] 1% of the value that we create and [32:51] return to investors and then figure out [32:53] how to share the rest sort of equally [32:54] among the world that I'd field that [32:56] feels okay so I mean there's just no [32:59] metrics for them to rely on here it's [33:01] just poof here's you know money and yeah [33:04] it's an odd fundraising conversation [33:06] right yeah well tell you that so is it [33:08] is it mostly individual investors I know [33:10] Reid Hoffman was obviously yeah it is [33:14] most it has historically been mostly [33:16] individual investors as we massively [33:19] ramp up our capital needs I expect that [33:22] to change institutions can I mean it [33:24] just seems like a really hard sell if [33:25] you're you know especially you don't [33:28] know when this is it could be hundreds [33:29] of years from now I mean I I don't think [33:31] it will be but you know it's an open [33:34] like I think one thing that no one knows [33:35] yet is just how much money you need [33:37] you know billions for sure tens of [33:39] billions maybe hundreds of billions I [33:40] sure hope not but the fundraising [33:44] strategy will have to adapt this when [33:45] you learn more about what we need i I [33:48] guess if you do raise billions of [33:50] dollars I mean what's what is gonna be [33:53] your sort of biggest differentiator then [33:55] like open ai versus like the mega corpse [33:59] I hope I hope that our biggest [34:02] differentiator is number one we do the [34:06] best research in the world and number [34:08] two we care the most about how it gets [34:12] deployed we're not trying to make Google [34:15] more ad revenue or whatever they do know [34:18] were we're not a government trying to [34:21] think about it like build super weapons [34:23] we sit there all day [34:26] and stress about what this means for [34:28] sort of the continuation of humanity and [34:31] how we want to design a structure where [34:33] the world gets input and benefit from [34:35] what we're doing okay now this is the [34:37] part of the interview where I shift into [34:39] criticisms I'm sure this is nothing [34:43] you've never heard before but you know I [34:44] talked to researchers I probably talked [34:46] in fairness like not the entire research [34:48] community probably talked to ten people [34:51] over time but people are really upset [34:53] about open AI I was really surprised by [34:56] this when we didn't release the GPT to [34:59] model people got really upset they got [35:01] really upset about it because I mean my [35:03] understanding and from again various [35:06] people is that you're not everything is [35:09] sort of qualitative not quantitative so [35:12] you know the language model was you [35:15] applying ten times the compute power 10 [35:18] times the GPUs and coming up with [35:19] something kind of interesting yeah but [35:21] nothing that was not developed [35:23] beforehand and sort of you know people [35:25] feel like you guys made it sound like [35:27] you invented something new well I I'm [35:32] sure that's not a Christian I've heard [35:34] I'm sure plenty people say it I think [35:35] the criticism that I felt a lot was you [35:41] all did do this really impressive new [35:43] thing and you're not sharing it and [35:45] that's against the norms of the [35:47] community the ML Research community has [35:49] traditionally been an academic community [35:51] where everything is open and shared and [35:53] it's available to anybody who wants it [35:55] and and and for the second time ever [36:00] with open and I we decided that not we [36:03] weren't going to release GPT to the full [36:06] model weights but they were going to do [36:07] it in stages so he said we've got this [36:09] thing we think it is so good that it can [36:12] be used for harm so powerful that it can [36:14] be used for harm and we are going to [36:17] tell the world about it first because we [36:19] have not yet built up sort of like [36:21] societal antibodies for fake text in the [36:24] way we have for photoshopped images and [36:26] then we will gradually release it over [36:28] time and so we've put out now [36:29] increasingly size models but still [36:32] nothing near the bleeding edge of our [36:34] internal capabilities and the academic [36:38] community did not [36:39] like that most of the rest of the world [36:42] said like that seems reasonable and [36:45] we're glad you're thinking about the [36:47] constants of consequences of Technology [36:49] instead of just releasing everything but [36:52] it is a hard decision and it is [36:53] definitely against the norm of the field [36:56] and I think as time goes on and we get [37:00] closer and closer to this true [37:02] generalized system or I think gonna have [37:06] to release less and less and I do think [37:08] people are gonna be upset by that they [37:10] were upset by that and I think the most [37:12] cynical take on that was oh look at the [37:14] timing they've decided to stop open [37:16] sourcing and now they're becoming this [37:17] capped profit company that's very much [37:19] like a for-profit company and you know [37:21] like I'm drawing a line from one thing [37:23] to another you say not yeah um I I think [37:31] like if we wanted to make money there [37:34] are things that we could do with what [37:35] we've currently developed that language [37:37] model is actually not the top of the [37:38] list but I think like our robotics work [37:40] for example and I get this suspicion and [37:45] it's particularly bad because like the [37:46] honest truth is like we do have to [37:47] figure out how to make a return for our [37:50] investors at some point and we don't [37:51] know how we're going to do it and that [37:54] should arise question marks for anyone [37:58] but I think the bigger issue at least in [38:01] the ml community will continue to be [38:03] that we're gonna have to not release [38:04] research well I'll tell you the other [38:06] criticism that I heard repeatedly was [38:08] well there's a few things that it's [38:12] qualitative not quantitative that also [38:17] the talk sort of which I think kind of [38:19] started with Elon Musk and seems to kind [38:21] of continue in the vein of like self [38:23] safe artificial general intelligence is [38:26] that you know it's a little bit of [38:28] fear-mongering in again I'm telling you [38:31] what I've heard yeah they say you know [38:34] it's not it doesn't even exist it's not [38:36] dangerous and open AI is sort of and [38:39] it's bad it's sort of a bad player of [38:41] first saying that it might be I do have [38:44] some sympathy to that argument but I [38:47] hope on the other hand we collectively [38:50] have learned in the tech industry that [38:52] if you're building [38:53] being on an exponential curve and you [38:56] don't think about the societal [38:57] consequences early um you know the same [39:00] people who say open the eye is [39:01] fear-mongering or whatever are the same [39:04] ones that say well shouldn't Facebook [39:05] have thought about this before they did [39:06] it and I think this is like us trying to [39:08] think about us both think about it [39:10] before we do it you know that I do think [39:16] that much of the secret to building AI [39:18] is just gonna be scaling these systems [39:19] bigger and bigger and bigger and so that [39:21] gbt to model you know maybe we were an [39:24] order of magnitude or two too early on [39:28] when we said okay we're not gonna [39:29] release this but I'd much rather be a [39:31] little bit too early then a little bit [39:32] too late and to push a button that you [39:34] can't on push and put something out into [39:35] the world and so in some sense I have a [39:39] lot of sympathy for this argument [39:41] because I don't like all the hand [39:44] wringing over like oh all of technology [39:46] is really awful and evil and and that on [39:49] the other sense like I think we as an [39:51] industry have failed the good faith test [39:53] of presuming that we're going to be [39:54] really thoughtful how we use all of this [39:55] and it's on us to show that we will be [39:59] and I hope that opening I can be [40:03] somewhat of a norms that are there I [40:05] think a thing that's wrapped up maybe in [40:08] that is we do get a lot of press [40:10] attention and I think that does bother [40:15] people and so maybe it's like well there [40:18] they did the thing that they did was an [40:20] improvement but only because of scale [40:22] and they got all this attention and that [40:23] feels unfair yeah [40:25] basically what I heard was these guys do [40:27] something incremental or they credit you [40:30] with doing something you know [40:31] interesting but because you added more [40:33] pieces and then you run to the press but [40:35] you know honestly I mean I did sort of [40:37] think about that and I thought well even [40:38] if it is maybe a little bit about [40:40] marketing maybe that's not such a bad [40:42] thing I mean are you trying to just you [40:44] get more we sort of attract more [40:47] attention to this we actually don't like [40:50] so one thing I think we do do is make [40:52] extremely good blog posts that are like [40:55] interactive and pretty compelling to [40:56] read and and in some sense that's [40:58] running to the press because people [41:00] write about it but we're not like [41:02] sitting there dialing every journalist [41:03] saying like please come write about this [41:06] thing I [41:07] felt caught off-guard by the amount of [41:09] attention that gbt two got and I thought [41:11] it was a great result so I think part of [41:14] it is people are like really interested [41:16] in covering this and we do in a way that [41:19] the field traditionally has not we do [41:21] produce I think pretty approachable and [41:24] cool demos I think we didn't ask like of [41:28] the risks that you're concerned of like [41:30] what's the most immediate is it do you [41:33] know enough about what you're staring [41:36] with a eye specifically or generally in [41:38] the world right now no are there sort of [41:42] priorities and you're sort of like this [41:44] is you know level red it's a pretty it's [41:47] honestly it's a it's a decently long [41:49] list I personally find the idea of [41:52] autonomous weapons and sort of thinking [41:55] through where that could go quite scary [41:58] but maybe the thing that I would say is [42:00] it top of the list although in some [42:04] sense a special case of that is I think [42:06] the cyber security issues we're going to [42:09] face with advanced AI are eye watering [42:14] ly bad and not taken particularly [42:17] seriously and given that these systems [42:20] are sort of getting one order of [42:22] magnitude more powerful per year it is [42:24] it's very difficult to reason about how [42:27] seriously we have to take that problem [42:28] so I think that'll be one of the first [42:30] areas where we see significant societal [42:34] threats if you just think about like how [42:36] much of our life is dependent on [42:39] connected computers at this point I'm [42:42] very worried about that [42:45] it's the sort of shortest term issue so [42:49] I know you don't have customers yet [42:51] you're not sure if and when you'll sell [42:53] to anybody you know talking about guns [42:57] and connectedness and I just wonder if [42:59] you could say how you would feel about [43:01] ever selling or working with the [43:04] military I would I'm definitely not [43:10] totally opposed to it and I think the [43:12] thing that like it's popular in Silicon [43:14] Valley right now I'm saying like we will [43:16] you know we're gonna make a sort of [43:18] loyalty test statement that we would [43:19] never [43:20] with the military under any conditions [43:21] like no I can definitely imagine [43:23] conditions where we would do anything we [43:26] could to help the military and I can [43:28] imagine a lot where we wouldn't but like [43:33] the answer like most of life is it [43:35] depends and can I also ask I meant to [43:39] ask this sooner but Elan musk had sort [43:42] of left the organization I guess maybe [43:44] last year he stepped off the board a [43:45] little bit yeah and he said it was sort [43:48] of just you know we're friends but it [43:50] was a disagreement can you say what the [43:52] disagreement was over it was mostly [43:55] because Tesla is working on a bunch of [43:57] AI stuff and some people from opening I [43:59] went to work a Tesla and it was like a [44:00] little bit messy the we certainly have [44:05] you know like occasional philosophical [44:08] we should proceed this direction or that [44:10] direction disagreements and also sort of [44:13] like how to think about the safety [44:15] issues but like we're close he's cool [44:18] like he's still extremely helpful to [44:20] open a I um let's see I know I've got to [44:24] let you go here I can keep going go for [44:26] it I guess I should probably let the [44:29] crowd ask you some questions too um wait [44:31] wait wait oh I didn't want to talk to [44:33] you about fundraising yeah okay this is [44:35] another thing people hate hate this [44:37] capped profit idea yeah well they think [44:40] it's very clever positioning and I have [44:42] to say I don't disagree although I do [44:43] think it's brilliant you know it's just [44:46] the the idea if you want explain to [44:47] people is that that when you do get [44:50] investment you'll your return your [44:54] investors and get up to 100 ex back [44:56] first before you put more money into the [44:59] sort of ad with his advocacy and [45:01] education notes before we just [45:02] distribute the money to the world or [45:04] whatever right but a hundred acts is a [45:06] huge bar I mean [45:07] whatsapp sold her twenty two billion [45:09] dollars for Facebook Sequoia I had [45:11] written I'm a sixty million dollar check [45:12] and that was like actually if I would [45:14] have entire but it was like a 50x return [45:15] so you'd have to be sort of one of the [45:18] most profitable companies or valuable [45:21] companies I guess in the world in order [45:22] for your investors to sort of look I'm I [45:25] think it's cool for people to make money [45:27] investing in companies I I think [45:29] capitals needs to adapt but fundamental [45:31] a system that I believe in the problem [45:35] with AGI specifically is that if we're [45:39] successful and we tried maybe we could [45:44] like capture the light-cone of all [45:45] future value in the universe and that is [45:47] for sure not okay for one group of [45:49] investors to have and so our thought was [45:54] and that was only for the first round [45:56] like the first round investors in the [45:58] what's that made much more than 100 X [45:59] you know subsequent investors in open [46:02] air I will make a cap of much less than [46:03] 100 X it will come down to hopefully 1 X [46:05] near the end but for our seed round [46:08] investors you know YC sometimes makes [46:10] 10,000 maybe a hundred thousand X I [46:12] don't know on some investments for our [46:15] seed round investors taking a risk on [46:18] this thing which is improbable at best [46:20] to get a hundred extra turn and then [46:22] willingly give up all their future [46:24] returns and for the company to give up [46:27] and employees to give up the future [46:29] returns to share with the world as a [46:30] whole [46:31] felt really fair and I think a lot of [46:35] the criticism of us I've heard but this [46:38] particular thing people are like why is [46:40] it only 100x not a thousand like that [46:42] doesn't sound enough yeah I think it's [46:45] it's great to know that so it's gonna [46:46] come down over time oh for sure okay [46:48] assuming we're doing well right and then [46:50] and your student our risk adjusted [46:51] return hopefully just gets better and [46:53] better every year [46:54] well I think it's it's probably neither [46:56] way did anybody want to ask oh gosh okay [46:59] um let's see are you sure you can stick [47:02] around for a little bit a little bit [47:03] yeah okay [47:07] thank you hello my name is Benna as I [47:09] come from Bangladesh [47:11] nice to meet and I have actually come [47:12] all the way to meet you Wow [47:15] so my question is as you see startups [47:18] coming into AI Combinator what [47:20] percentage you see in yarny are growth [47:22] in terms of participation from outside [47:25] of us and what percentage you see or a [47:30] growth on the side of let's say purpose [47:33] first and then do you think you know the [47:37] next Silicon Valley is not from within [47:40] us or outside okay now I'll answer all [47:43] three in order at this point most of our [47:45] growth is outside the US we there's this [47:48] wonderful phenomenon that we have where [47:51] a company will come here from somewhere [47:53] else in the world let's say Bogota rappy [47:57] comes here they go on to be really [47:59] successful they go back and like we get [48:02] incredible free marketing out of that [48:03] and they tell everyone how great this [48:05] thing called YC is and that there's this [48:06] thing called a startup if they should [48:08] start and then after that happens now we [48:11] have a ton of Colombian startups [48:12] applying to YC and this happens a like [48:15] basically we have like this one success [48:17] case in a new market and then we are [48:20] flooded in it the best way with [48:22] companies from there and at this point I [48:25] won't say that's where all of our growth [48:27] comes from but I bet most most of it is [48:29] non-us so that's that's wonderful and I [48:34] think that is part of why I believe YC [48:36] has an easy 10x in front of it of growth [48:39] in the next 10 or 20 years in terms of [48:42] purpose driven companies yes in fact [48:45] that is now one of the first questions I [48:46] ask a company huh there's like you can [48:49] look out there at that sea of startups [48:51] literally and why you're gonna choose [48:56] one or the other [48:57] is unclear and the only thing that is [49:01] cheap and easy when capital is this [49:03] abundant is capital concentrated talent [49:06] and a company has never been harder than [49:07] news today everybody wants to start a [49:09] company or be employing number one and [49:10] the question of like how are you gonna [49:12] get enough really good people together [49:14] just sort of put their talents and time [49:17] and effort into your company that's like [49:20] one of the first questions I ask and the [49:22] companies that don't have an answer to [49:23] that they have a real challenge what was [49:24] your third question [49:27] Oh next Silicon Valley um well there was [49:32] never another Florence quite either it [49:35] all just sort of diffused and I think [49:37] this is a this has been a moment like [49:40] the Renaissance that we will not see [49:44] again for this one industry in any one [49:46] particular place it'll just sort of be [49:48] everywhere all right as I'm walking I'm [49:52] gonna ask you Softbank [49:53] got bad neutral you know I don't spend [49:58] much time thinking about it thank you [50:02] I'm Salman and I'm based in Pakistan and [50:05] UAE it was really a pleasure listen to [50:07] your thoughts and an honor to have your [50:09] mom with us thank you very much my [50:11] question is as a Silicon Valley [50:13] ambassador what is that one advice that [50:16] you have for President Trump on global [50:18] economy fix immigration if I could get [50:26] if I could pick one policy that I think [50:29] would help the global economy in [50:33] particularly the US economy it would be [50:35] that the number of our founders the [50:37] percentage of our founders and founders [50:38] of tech companies in general that are [50:41] immigrants is is quite high i I think it [50:45] has been one of the u.s. is great [50:49] advantages is the ability to concentrate [50:51] talent here and I think it's benefited [50:52] the whole world too and I'm very scared [50:56] about that [51:00] what oh really [51:03] I've been super busy all day so I [51:04] haven't gotten to read the news I know [51:06] there was a question in the back - yeah [51:09] okay I think we make this the last one [51:10] so we can let these guys go offer their [51:11] dinner okay another question Sam I don't [51:14] think you're a trump supporter oh I [51:16] definitely not okay so do you like any [51:18] the candidates for um honestly I would [51:21] take any of them over Trump I think they [51:27] would all be I happily take that trade [51:31] thanks so much Sam [51:33] I was curious your thoughts on the [51:35] facial recognition ban that just [51:37] happened [51:38] in the city so I heard that that [51:41] happen and that is literally the only [51:43] sense I know about it I it's hard to [51:48] have a strong opinion without [51:49] understanding it more I definitely have [51:53] some concerns about facial recognition [51:56] being widely deployed and then I think [51:58] there's like some ways in which it's a [51:59] really good thing to do but I'd say my [52:03] my posture towards it is one of more [52:05] concern not less and so it would depend [52:08] on the details but I think we have to [52:12] have a real conversation as a society [52:14] about how we want to do that I think [52:17] when every at every big new [52:19] technological revolution some of the [52:23] things that we sort of hold a society [52:25] but that have some Astra's like the [52:29] right to free speech you know that's [52:31] always had some Astra seize and now it's [52:32] gonna get some more I hope for the [52:34] Internet because that's different than [52:35] what you can say in a town square and [52:37] the right to privacy there's like one [52:40] version of this which is you know [52:42] sometimes you are allowed to break into [52:43] someone's home with the right stuff and [52:45] a different level of expectation on the [52:48] street and when everything interesting [52:50] is digital or digitally captured like [52:53] what does that mean and we're gonna have [52:55] to evolve all of that and so the [52:58] specifics of it matter a lot to me [52:59] there's versions of it that I think I [53:01] could support and there's versions of it [53:03] that I think would be unacceptable and I [53:06] don't know enough beyond that to comment [53:07] on this specific one I've always loved [53:10] about you that you're so brilliant but [53:12] you're willing to say when you don't [53:14] know enough about something I think it's [53:16] better than saying something wrong [53:17] absolutely Sam thank you so much such a [53:20] joy to talk to you I really really [53:22] appreciate it