WEBVTT - America’s New AI Strategy 0:00:15.564 --> 0:00:28.604 Pushkin. Welcome back to Risky Business, a show about making 0:00:28.644 --> 0:00:31.404 better decisions. I'm Maria Kanakova. 0:00:31.004 --> 0:00:35.204 And I'm Nate Silver. Time for an AI episode, I think, Aria, Yeah, it's. 0:00:35.084 --> 0:00:38.804 It's been a minute since we had AI related conversations 0:00:38.804 --> 0:00:41.964 on the show. But it's obviously a hugely important topic 0:00:42.004 --> 0:00:44.204 and one that should have been more in the news 0:00:44.284 --> 0:00:48.364 last week it got you know, got sidelined by other news. 0:00:49.004 --> 0:00:53.644 But who's that a guy don't don't know, don't know 0:00:53.804 --> 0:00:59.524 Nate but never never met him. That's true, never met him. 0:01:00.244 --> 0:01:02.564 But yeah, well, we'll talk a little bit about the 0:01:02.604 --> 0:01:05.884 AI action plan that was released by the US government 0:01:05.964 --> 0:01:18.444 last week. We have lots to talk about before we do. Nate, 0:01:19.324 --> 0:01:22.084 we're both back in New York, but I'm actually flying 0:01:22.084 --> 0:01:26.364 back to Vegas in twenty four hours to play in 0:01:26.404 --> 0:01:29.844 the twenty five k NBC Heads Up Championship. 0:01:30.764 --> 0:01:33.684 Yeah are you? Are you excited? Have you been coaching? 0:01:34.244 --> 0:01:34.644 I have? 0:01:34.924 --> 0:01:38.924 I have so For people who don't know, this was 0:01:38.964 --> 0:01:41.084 a big show back in the day. It was before 0:01:41.164 --> 0:01:43.084 my time, so I never watched it. Nate, did you 0:01:43.124 --> 0:01:43.644 ever watch it? 0:01:44.244 --> 0:01:46.844 Sure? Yeah, yeah, so time, you're not that much younger. 0:01:47.324 --> 0:01:49.844 I didn't know anything about poker. It was before my time, 0:01:49.964 --> 0:01:53.924 like actually knowing anything about poker, So I didn't watch 0:01:54.004 --> 0:01:57.124 any poker, know about the existence of any poker shows 0:01:57.124 --> 0:02:00.884 before I started researching Biggest Bluff. So that's what I mean, 0:02:00.924 --> 0:02:02.764 before my time, before my time in the poker world. 0:02:02.884 --> 0:02:06.444 Had you heard of poker only because of rounders? 0:02:06.564 --> 0:02:09.084 I knew that technically it was a game that existed, 0:02:09.924 --> 0:02:12.244 but it's not something that was ever really on my radar. 0:02:12.404 --> 0:02:15.204 But yeah, this was a huge show on NBC back 0:02:15.204 --> 0:02:18.324 in the day. Heads Up poker is one on one poker, 0:02:18.524 --> 0:02:21.524 which is kind of the most competitive in some ways 0:02:21.604 --> 0:02:24.524 form of poker because you're forced to play all hands. 0:02:24.564 --> 0:02:27.524 Basically your ranges are one hundred percent, so it forces 0:02:27.564 --> 0:02:30.964 a very different kind of thinking. And it's being rebooted 0:02:31.084 --> 0:02:35.204 in conjunction with Peacock and Poker Stars and Poker Goes 0:02:35.324 --> 0:02:38.484 kind of is helping sponsor all of this. So I 0:02:38.524 --> 0:02:42.964 am yeah, I'll be in the Poker Ghost Studios taping 0:02:43.004 --> 0:02:45.084 and I hope it goes well. I've been getting coaching 0:02:45.124 --> 0:02:48.644 from Kevin Rabichow, who's considered Yeah, he's considered one of 0:02:48.684 --> 0:02:51.164 the best heads up players in the world. He has 0:02:51.204 --> 0:02:54.884 a new heads up coaching course that just came out. 0:02:55.044 --> 0:02:58.244 But yeah, so I've been I've been really trying. You know, 0:02:58.924 --> 0:03:01.244 usually after the World Series, I'm like, okay, no poker 0:03:01.284 --> 0:03:04.284 for a while. Right, it's not gonna just gonna clear 0:03:04.364 --> 0:03:05.644 my my clear my brain. 0:03:07.204 --> 0:03:09.404 I'm in the no poker zone for a period of time. 0:03:09.404 --> 0:03:14.804 We're also bad heads up to be frank, which was 0:03:14.924 --> 0:03:16.884 some coaching might have been good, but like, yeah, I 0:03:16.924 --> 0:03:19.164 can't can't make the time commitment this summer in. 0:03:19.924 --> 0:03:24.404 Yeah, so instead of taking my usual break from from 0:03:24.484 --> 0:03:28.364 poker that I do, I've been instead like going hard 0:03:28.484 --> 0:03:32.124 and trying to prepare myself for this heads up Championship. 0:03:32.644 --> 0:03:35.284 And you know it, heads up, the variants so much higher, 0:03:35.484 --> 0:03:40.364 and the card distribution matters so much, right, So, like no, 0:03:40.564 --> 0:03:44.124 and if someone is just getting hands, that's kind of 0:03:44.884 --> 0:03:48.204 you know, that's it, Like they're they're potentially going to win. 0:03:48.244 --> 0:03:51.084 And there have been lots of times where really great 0:03:51.084 --> 0:03:55.164 players have been just completely obliterated by someone who doesn't 0:03:55.164 --> 0:03:58.764 know how to play. Yeah, so card distribution can be 0:03:58.804 --> 0:04:01.124 a bitch heads up, or it can be amazing. 0:04:04.324 --> 0:04:08.044 Are those exclusive? Uh No, Okay, I'm not going to 0:04:08.044 --> 0:04:08.404 get into it. 0:04:08.484 --> 0:04:14.204 Yeah, no, bitchin amazing can be can be within the 0:04:14.244 --> 0:04:16.764 same match, but it does matter much more so than 0:04:16.804 --> 0:04:19.844 in full ring poker. I saw some of the names 0:04:19.844 --> 0:04:24.324 that we'll be playing, and there's gonna be no spoilers obviously. Well, 0:04:24.324 --> 0:04:25.764 I haven't played yet, but I'm not going to be 0:04:25.804 --> 0:04:27.684 able to tell you how I did until the show 0:04:27.724 --> 0:04:31.604 airs in the fall. Yeah, since it's not it's not 0:04:31.684 --> 0:04:34.524 airing in real time, but you will. People do know 0:04:34.564 --> 0:04:39.684 who's playing. And if I'm paired with someone like Jason 0:04:39.764 --> 0:04:42.604 Kuhn on my first match, yeah, that's that's not going 0:04:42.644 --> 0:04:45.484 to be fun. And if I'm paired with Don Cheetle, 0:04:45.524 --> 0:04:48.324 who's also playing, I think I'm gonna be much happier. 0:04:49.884 --> 0:04:50.124 Yeah. 0:04:50.124 --> 0:04:53.844 I didn't realize there that many not Fish but celebrities 0:04:53.924 --> 0:04:57.204 that maybe Fish adjacent potentially, although he plays a little 0:04:57.204 --> 0:04:58.164 bit right he does. 0:04:58.284 --> 0:05:04.644 Yeah, so so we'll We'll just see and I'm excited 0:05:04.644 --> 0:05:07.084 for it no matter how it goes. And Kevin has 0:05:07.124 --> 0:05:11.204 been very good at preparing me for both heads up 0:05:11.284 --> 0:05:15.084 and also for the eventuality that he as one of 0:05:15.164 --> 0:05:17.684 the best and at times probably the best in the world, 0:05:17.684 --> 0:05:21.804 has sometimes busted in the first round right in the 0:05:21.804 --> 0:05:23.924 matchups like this, because he plays a twenty five K 0:05:24.084 --> 0:05:26.044 heads up at the World Series every year. 0:05:26.444 --> 0:05:28.604 Are you worried about like physical reads and things like that, 0:05:28.644 --> 0:05:30.724 because like that can be more of a factor too. 0:05:31.084 --> 0:05:34.724 Yeah, you know, so I tend to be you know, 0:05:34.924 --> 0:05:38.564 I hope that people can't get a lot of reads 0:05:38.564 --> 0:05:40.764 off of me, but you just never know. And this 0:05:40.844 --> 0:05:43.364 is actually hilarious night the rules of this. When I 0:05:43.364 --> 0:05:46.564 started reading it, I just burst out laughing because they 0:05:46.604 --> 0:05:49.964 said that you are allowed So people who don't play poker, 0:05:49.964 --> 0:05:53.124 this is not normal. You're allowed to expose one or 0:05:53.364 --> 0:05:56.804 both of your cards at any point during a hand. 0:05:57.204 --> 0:05:59.204 You can cash usually but not in turn. 0:05:59.364 --> 0:06:01.564 But this, yeah, this is a tournament. Yeah, so you 0:06:01.604 --> 0:06:04.604 can expose your cards one or both of them. And 0:06:04.644 --> 0:06:07.284 they said they encourage speech play. You can talk about 0:06:07.284 --> 0:06:09.764 the contents of your hands. So basically everything are not 0:06:09.844 --> 0:06:12.884 allowed to do in tournaments normally you're being encouraged to 0:06:12.884 --> 0:06:16.084 do to make good television. I'm not going to be 0:06:16.084 --> 0:06:17.084 showing my cards day. 0:06:17.644 --> 0:06:21.484 Okay, I've become more of a speech play guy over time. 0:06:21.564 --> 0:06:25.564 No really, yeah, okay, okay, give me an example. 0:06:25.564 --> 0:06:27.644 You know, like the Europeans, I'll like try to get 0:06:27.644 --> 0:06:29.404 info out of them. Were so quiet all the time 0:06:29.444 --> 0:06:31.004 that like I want, like, I just want to like 0:06:31.084 --> 0:06:33.604 any vibe I can get pretty much, or with like 0:06:33.644 --> 0:06:37.924 really bad players who like often will reveal information just 0:06:37.964 --> 0:06:40.604 to gauge their comfort level. It's kind of been a 0:06:40.764 --> 0:06:44.404 proto trial and error Zoka. For right now, I'm trying 0:06:44.404 --> 0:06:46.604 to like kind of like collect mental database of like 0:06:47.044 --> 0:06:48.764 how people react to different things. 0:06:49.444 --> 0:06:50.844 I don't know, it's kind of I mean, because so. 0:06:50.844 --> 0:06:55.084 Many decisions are so close, yeah, equilibrium that like any 0:06:55.204 --> 0:06:59.164 vibe you can get, And I like recently cash games 0:06:59.204 --> 0:07:02.404 more than tournaments and they're kind of looser in various 0:07:02.404 --> 0:07:05.684 senses of the term, right, people like you reveal more 0:07:05.724 --> 0:07:07.684 info voluntarily or involuntarily. 0:07:07.724 --> 0:07:09.924 I like cultivating that environment. I think. 0:07:10.444 --> 0:07:13.804 Yeah, no, I'm someone who definitely is chatty and tends 0:07:13.844 --> 0:07:16.764 to be like very friendly at poker tables. But you know, 0:07:16.844 --> 0:07:18.724 I'm not someone who's going to be like, oh, you know, 0:07:18.764 --> 0:07:21.524 do you have aces, like do you do things like 0:07:21.564 --> 0:07:24.084 that or is it because you know there are players 0:07:24.084 --> 0:07:26.684 who do or is it something a little bit different 0:07:26.764 --> 0:07:27.884 just to try to engage. 0:07:28.204 --> 0:07:30.324 Yeah, maybe it's like you're would have decided to fold 0:07:31.124 --> 0:07:33.404 already if they want to like reveal something that they 0:07:33.484 --> 0:07:37.844 might not reveal otherwise, or to send a bassline for 0:07:37.884 --> 0:07:39.764 the next time we play a hand, things like that. 0:07:40.164 --> 0:07:41.604 Yeah, that's that's actually smart. 0:07:41.804 --> 0:07:44.164 So how can you reveal afterward? Right? 0:07:44.284 --> 0:07:46.324 If you're friendly in the right way, they might reveal 0:07:46.444 --> 0:07:47.524 information later on. 0:07:47.764 --> 0:07:48.404 Right for sure. 0:07:48.684 --> 0:07:51.524 Yeah, that's actually something that I have capitalized on because 0:07:51.564 --> 0:07:55.124 I am a friendly player. Like I'll sometimes when it's obvious, 0:07:55.204 --> 0:07:57.164 like I've made a big fold or something like that, 0:07:57.324 --> 0:08:00.564 and I like make it obvious, they'll show me. And 0:08:00.644 --> 0:08:03.644 that's that's very nice because it's good information. But I'm 0:08:03.684 --> 0:08:06.524 always I do try to be careful, and I tell 0:08:06.564 --> 0:08:10.164 people this, you know, with with poker and with just 0:08:10.204 --> 0:08:14.444 like reads and our ability to spot deception in general, 0:08:14.484 --> 0:08:16.684 that we do tend to be not great at it 0:08:17.084 --> 0:08:20.684 as human beings. And the other kind of the other 0:08:20.764 --> 0:08:23.324 element of that is that we think we're better than 0:08:23.364 --> 0:08:26.484 we are. And we also often reveal information when we're 0:08:26.524 --> 0:08:30.844 engaging in speech play without realizing it because of how 0:08:30.924 --> 0:08:33.284 we're talking, what we're doing. And so it's something that 0:08:33.324 --> 0:08:36.644 I think just in general, with the psychology of the thing, 0:08:36.964 --> 0:08:38.324 you have to be careful. 0:08:38.804 --> 0:08:39.644 It's all contextual. 0:08:39.724 --> 0:08:44.804 Any poker tell speech play, it's all about semantic context, 0:08:44.884 --> 0:08:46.404 so to speak for sure. 0:08:46.564 --> 0:08:50.284 For sure. On that note, Nate, shall we switch gears 0:08:50.604 --> 0:08:56.564 and talk about AI and the news from last week 0:08:56.604 --> 0:09:00.084 which got buried because there's a lot of news in 0:09:00.764 --> 0:09:02.884 the news. There's a lot of news in the news 0:09:02.924 --> 0:09:07.404 these days, Nate, that's a very deep insight. So last 0:09:07.404 --> 0:09:13.804 week President released the AI Action Plan that talks about 0:09:14.284 --> 0:09:18.564 what the administration wants to kind of see going forward 0:09:18.604 --> 0:09:21.284 in the AI space. And I think the name of 0:09:21.324 --> 0:09:25.564 this action plan is actually quite telling about the contents. 0:09:25.564 --> 0:09:29.884 It says, winning the Race America's AI Action Plan. So 0:09:29.964 --> 0:09:33.644 it's titled winning the Race. And that's how it's framed, right, 0:09:33.884 --> 0:09:37.484 It's framed as a race between the US and everyone else, 0:09:37.604 --> 0:09:42.964 especially China, but really everyone else, mostly China, but mostly no, 0:09:43.084 --> 0:09:47.604 there isn't. And the overall point of all of the 0:09:47.684 --> 0:09:51.884 different sections is how can we make the US the 0:09:52.004 --> 0:09:56.004 leader in AI the most competitive and how can we 0:09:56.044 --> 0:09:59.764 do that by also kind of screwing over China in 0:09:59.804 --> 0:10:03.204 some respects by not allowing it access to the types 0:10:03.244 --> 0:10:06.124 of things that will allow our allies to access and 0:10:06.164 --> 0:10:08.364 some of the things in it are contradictory, but that's 0:10:08.444 --> 0:10:12.564 kind of the overall message. Did you read the report night, 0:10:12.644 --> 0:10:14.004 Do you have any initial thoughts? 0:10:14.164 --> 0:10:14.284 So? 0:10:14.444 --> 0:10:17.724 V Maschwitz is one of the most prominent and best 0:10:18.084 --> 0:10:21.324 AI bloggers. He writes the substack, don't worry about the vase, 0:10:23.124 --> 0:10:25.364 and he calls it as he sees that he's more 0:10:25.404 --> 0:10:28.884 concerned about ped doom than a lot of people. P 0:10:29.004 --> 0:10:31.324 doom is the possibility if things going really badly, extinction, 0:10:32.164 --> 0:10:34.364 loss of control, et cetera. And he said he thought 0:10:34.404 --> 0:10:38.604 the plan was actually his term pretty good. I don't 0:10:38.924 --> 0:10:43.044 know his politics in detail, but he really dives down 0:10:43.084 --> 0:10:45.604 into like every last word you can read that. So 0:10:45.644 --> 0:10:49.644 I'm kind of relying on that for like general vibes, right. 0:10:49.684 --> 0:10:53.364 I mean, it is interesting that like, even like even 0:10:53.364 --> 0:10:57.044 people who are concerned like V about AI safety, deeply 0:10:57.044 --> 0:11:02.564 concerned in his case, have this ambiguous relationship with China, 0:11:02.964 --> 0:11:05.124 right where if we were the only country developing AI 0:11:05.244 --> 0:11:08.124 and we had like centralized control, then it would be 0:11:08.124 --> 0:11:11.604 easier to argue that Okay, let's speak appropriately cautious here, 0:11:12.124 --> 0:11:17.244 take it slow right with China. You know a lot 0:11:17.284 --> 0:11:18.564 of people I don't want it to be of you 0:11:18.684 --> 0:11:20.924 to him I do think that, well, we have no 0:11:21.044 --> 0:11:25.684 choice but to race, and maybe we can race in 0:11:25.804 --> 0:11:29.724 ways that involves some degree of coordination or cooperation. It's 0:11:29.764 --> 0:11:31.684 hard to run races like that. But you know, I 0:11:31.684 --> 0:11:35.924 mean the notion of like securing American leadership instead of 0:11:36.684 --> 0:11:41.244 seeding it to China, I don't think is terribly controversial. 0:11:41.284 --> 0:11:42.564 I mean there's a lot in the in the in 0:11:42.604 --> 0:11:47.444 the in the plan though, right, there are some Trumpian 0:11:47.564 --> 0:11:53.684 elements around factors like the political orientation of AI models, 0:11:53.764 --> 0:11:57.404 right where they don't want them to be too woke, 0:11:57.564 --> 0:12:01.204 They don't want them to be too political quote unquote right. 0:12:01.444 --> 0:12:03.164 Yeah, which is one of the things which is hard, 0:12:03.404 --> 0:12:05.044 Which is hard, and which is one of the things 0:12:05.044 --> 0:12:08.164 that ZV does point out because I read this as well, 0:12:09.244 --> 0:12:11.084 it's one of the one of the things that he 0:12:11.124 --> 0:12:15.804 points out is actually probably not possible given the type 0:12:15.844 --> 0:12:20.964 of language that that Trump employs, and that as a psychologist, 0:12:21.004 --> 0:12:25.764 I would say is not possible given how AI is developed, 0:12:25.804 --> 0:12:29.764 because there's just there's bias everywhere, and there's no such 0:12:29.844 --> 0:12:33.604 thing as like this is the objective thing, and if 0:12:33.604 --> 0:12:37.724 you're constantly policing quote unquote anti woke, even if that's 0:12:37.804 --> 0:12:40.404 kind of the unbiased position. But you're like, oh no, 0:12:40.564 --> 0:12:44.044 that's you know that, then you're going to it's problematic. 0:12:44.204 --> 0:12:47.404 It's problematic in terms of how your how you go 0:12:47.444 --> 0:12:47.844 about it. 0:12:48.164 --> 0:12:53.164 Yeah, with with Groc or more technically, the Twitter bought Groc, 0:12:53.204 --> 0:12:57.444 which is an instance of Elon Elon's overall modeled rock. Right, 0:12:58.844 --> 0:13:01.924 Elon just did a couple of like subtle things. It 0:13:01.964 --> 0:13:05.124 was like, oh, don't be too woke here, don't trust 0:13:05.164 --> 0:13:08.644 the media too much. Right, if you read the actual 0:13:08.724 --> 0:13:11.084 changes to the archae texture, they weren't that profound, And 0:13:11.164 --> 0:13:14.884 yet Kroc was calling itself. I believe the term is 0:13:14.924 --> 0:13:18.844 Mecha Hitler after some period of time engaging in elaborate 0:13:19.804 --> 0:13:23.724 rape fantasies. I wish this were a joke, but it's not. 0:13:24.004 --> 0:13:28.004 And like so yeah, I mean if you kind of 0:13:28.124 --> 0:13:32.444 nudge a model to override its inputs, it's hard to 0:13:32.444 --> 0:13:35.444 do so in a way that like has any type 0:13:35.484 --> 0:13:37.404 of dexterity. I mean Google had this problem too, right 0:13:37.444 --> 0:13:39.844 when they got to Gemini two years ago, a year 0:13:39.844 --> 0:13:42.844 and a half ago, right, they were like, oh my gosh, yeah, 0:13:43.044 --> 0:13:45.164 people are there's too many like white people, white men 0:13:45.164 --> 0:13:48.244 in these photos, So like literally it was forced to 0:13:48.324 --> 0:13:51.284 draw like multi racial Nazis. 0:13:51.524 --> 0:13:52.804 It always comes down to Nazis. 0:13:52.804 --> 0:13:56.764 And by the way, why well, in part because like 0:13:56.764 --> 0:14:00.044 if you're kind of mining text of arguments on the 0:14:00.044 --> 0:14:06.644 Internet and on Twitter, people invoke Nazi comparisons all the time. 0:14:06.724 --> 0:14:08.244 Right, Who's hot Sydney Sweeney? 0:14:08.324 --> 0:14:11.804 Right, there was some new American ego or commercial where 0:14:11.884 --> 0:14:14.964 like she's inn am pair of gens and she talks 0:14:14.964 --> 0:14:18.164 about how she is good genes. A little weird, right, 0:14:18.204 --> 0:14:21.924 but like it was kind of Nazi. It was a 0:14:21.924 --> 0:14:25.364 little weird. However, idiots on the internet were like, this 0:14:25.644 --> 0:14:29.124 has Nazi symbolism in it, you know, not just alluding 0:14:29.124 --> 0:14:31.724 to that saying the word Nazi. So if you're like groc, 0:14:32.284 --> 0:14:33.924 you're like, okay, well you want like kind of the 0:14:33.964 --> 0:14:39.004 garden variety political epitaph and calling people Nazi or Hitler, 0:14:39.484 --> 0:14:41.724 which should be more forbidden, you would think, But that 0:14:41.804 --> 0:14:45.084 taboo is a violated all the time. And so an 0:14:45.084 --> 0:14:48.644 AI trained on political speech, if you go and say, yeah, 0:14:48.764 --> 0:14:51.804 don't be so woke here, right, Well, can kind of 0:14:51.844 --> 0:14:53.804 go into like a downward spiral election people do where 0:14:53.844 --> 0:14:56.964 before long it's calling everybody or itself hitler. 0:14:57.764 --> 0:15:01.004 Yeah, no, I think that that's absolutely true, and you 0:15:01.044 --> 0:15:06.284 know it also your your point actually makes a broader 0:15:06.404 --> 0:15:10.524 point about this AI action play, and where a lot 0:15:10.564 --> 0:15:12.604 of the things and I did skim through it. I 0:15:12.644 --> 0:15:15.324 didn't read the whole thing word for word. It is 0:15:15.524 --> 0:15:19.044 small print on twenty eight pages, and I had other 0:15:19.164 --> 0:15:20.964 things to do this week, so I just prepare for 0:15:20.964 --> 0:15:22.884 the Heads Up Championship. So I didn't read the whole thing, 0:15:23.084 --> 0:15:25.324 but a lot of it actually, like there are things 0:15:25.364 --> 0:15:29.044 that sound reasonable in theory, but there's no practical way 0:15:29.124 --> 0:15:31.964 of like how is this actually going to happen? How 0:15:31.964 --> 0:15:34.084 do we make this happen? And so you know, I 0:15:34.084 --> 0:15:36.404 think that this is a very concrete this like anti 0:15:36.484 --> 0:15:39.644 woke is a very concrete example where like, okay, well, 0:15:39.724 --> 0:15:42.324 even if in theory we're saying we want everything to 0:15:42.364 --> 0:15:45.444 just be unbiased, let's like, let's just remove anti woke, right, 0:15:45.764 --> 0:15:49.164 we want everything to be like as unbiased as possible. 0:15:49.324 --> 0:15:51.764 Let's pretend that that's actually what it said, even though 0:15:51.764 --> 0:15:54.844 it didn't. It did use the terminology anti woke, and 0:15:54.844 --> 0:15:57.604 then there's even an anti woke executive order after it. 0:15:57.764 --> 0:16:00.284 Let's pretend it didn't, Like, let's just pretend it was 0:16:00.524 --> 0:16:04.004 we do not want bias. Great, I agree with that. 0:16:04.604 --> 0:16:07.004 How do you implement it? Like, how do you actually 0:16:07.044 --> 0:16:09.324 like do those weights? How do you It's it's base 0:16:09.484 --> 0:16:13.724 sickly impossible. If not, like, it's a ridiculously thorny problem. 0:16:14.044 --> 0:16:17.444 And that is just a very specific thing that you 0:16:17.484 --> 0:16:19.884 can kind of grab onto it to realize how a 0:16:19.924 --> 0:16:23.244 lot of these statements might sound good and like, Okay, 0:16:23.244 --> 0:16:25.324 well this is a reasonable aim, but if you don't 0:16:25.364 --> 0:16:28.404 actually have the implementation down and have the nitty gritty 0:16:28.404 --> 0:16:30.724 of how to do it, it's pointless. It's just words, 0:16:30.764 --> 0:16:35.844 it's just rhetoric. And we'll be back right after this. 0:16:48.964 --> 0:16:52.764 There's another really interesting part of this action plan that 0:16:52.844 --> 0:16:55.644 I think we should talk about because it was something 0:16:55.684 --> 0:16:58.564 that the Biden and administration also struggled with, which is 0:16:59.124 --> 0:17:02.284 what do we do in terms of proprietary versus open 0:17:02.324 --> 0:17:03.284 source models? 0:17:03.364 --> 0:17:03.524 Right? 0:17:03.964 --> 0:17:07.844 Do we encourage people to kind of share all the 0:17:07.844 --> 0:17:11.004 code so that anyone can use it and anyone can 0:17:11.044 --> 0:17:13.524 kind of build on top of it, or is this 0:17:13.604 --> 0:17:17.444 something that we want to kind of keep inside, kind 0:17:17.444 --> 0:17:21.444 of keep internal. China went open source right almost immediately 0:17:21.524 --> 0:17:23.444 with kind of deep seak all of that those are 0:17:23.524 --> 0:17:27.564 open source models, and in the US, the Biden administration 0:17:27.604 --> 0:17:30.084 didn't really like they just like punted on it. They 0:17:30.164 --> 0:17:33.564 didn't they didn't have a position. And we do have both, right, 0:17:33.724 --> 0:17:36.444 we have proprietary models, we have some open source stuff, 0:17:36.844 --> 0:17:39.804 and there are arguments on both sides. But you had 0:17:39.844 --> 0:17:41.964 started off by talking about p doom, which is not 0:17:42.004 --> 0:17:44.684 something this report mentions at all, and one of the 0:17:44.804 --> 0:17:50.444 arguments against having everything be open sources. Well, okay, great, 0:17:50.564 --> 0:17:54.924 but then you know, rogue actors, people who don't have 0:17:54.964 --> 0:17:58.084 the best interests of the United States in mind will 0:17:58.124 --> 0:18:00.804 have this code as well. It's kind of the same 0:18:00.924 --> 0:18:03.924 arguments that have been made in the past with like 0:18:03.964 --> 0:18:07.804 when people provide the full genomic sequence of like deadly viruses, 0:18:07.884 --> 0:18:11.124 right or things like that, are you know, instructions for 0:18:11.164 --> 0:18:14.244 how to build kind of a bomb was step by step? 0:18:14.484 --> 0:18:18.084 Like hey, guys, like, why are we actually making all 0:18:18.124 --> 0:18:20.524 of this information publicly available? So that's kind of I 0:18:20.524 --> 0:18:23.524 think the strongest case against open source. There's a lot 0:18:23.564 --> 0:18:25.324 of pro let's let's talk about con first. 0:18:25.364 --> 0:18:27.524 Yeah, well, look, I think part of it was there's 0:18:27.564 --> 0:18:29.404 a period of time where I think the AI safety 0:18:29.404 --> 0:18:31.284 folks thought that hey, maybe we could just kind of 0:18:32.364 --> 0:18:37.284 contain it to open AI, Anthropic and Google or like 0:18:37.404 --> 0:18:39.964 or have been like the Big three, right, and maybe 0:18:39.964 --> 0:18:42.004 that we can avoid like a race dynamic if there's 0:18:42.004 --> 0:18:43.884 a finite number of players and they all believe in 0:18:43.924 --> 0:18:47.484 AI safety, right, and I think now a there's like 0:18:47.644 --> 0:18:51.804 less belief that like open AI in particular, is that 0:18:51.884 --> 0:18:56.444 concerned about AI safety. It's a legitmission, right, you know. 0:18:56.644 --> 0:18:59.484 B we've seen with deep Seek the Chinese model that 0:18:59.644 --> 0:19:02.844 like with some clever engineers, you can probably have a 0:19:02.884 --> 0:19:06.244 model it's like only six months to a year behind 0:19:06.284 --> 0:19:10.284 the frontier, probably not on the frontier. Sottitudes have changed 0:19:10.324 --> 0:19:13.244 again kind of as a resignation to the inotability, I 0:19:13.244 --> 0:19:16.084 think in part and also there is some protectionism, right, 0:19:16.164 --> 0:19:20.604 Like Meta has championed open models. Why well, partly because 0:19:20.644 --> 0:19:24.924 Meta's models don't suck, but they've been considered behind to 0:19:24.964 --> 0:19:28.604 the point where they're literally offering people AI engineers, in 0:19:28.644 --> 0:19:31.324 one case, allegedly, according to a Wired as much as 0:19:31.324 --> 0:19:33.964 a billion dollars to join Meta so they can catch. 0:19:33.804 --> 0:19:36.044 Up and become like a fourth part of the Big four. 0:19:36.444 --> 0:19:37.604 I guess it would be, right. 0:19:38.204 --> 0:19:41.164 So yeah, I mean, when it comes to election analysis 0:19:41.964 --> 0:19:44.444 and other models. Oftentimes, I found the people in that 0:19:44.524 --> 0:19:47.204 space who are advocating for total open source everything are 0:19:47.244 --> 0:19:50.884 people that like A don't need the income or be their. 0:19:50.724 --> 0:19:52.484 Models kind of suck, right. They're like, we're doing this 0:19:52.524 --> 0:19:53.164 really good the community. 0:19:53.164 --> 0:19:54.644 It's like, well, you're doing that because you're not gonna 0:19:54.724 --> 0:19:57.244 it's not really good enough to sell that product for 0:19:57.324 --> 0:20:00.564 a premium price, whereas I think mine is for example. Right, So, 0:20:00.604 --> 0:20:04.084 like there are all types of ulterior motives. There's regulatory capture, 0:20:04.164 --> 0:20:07.284 right of course Sam Milton might say, well, yeah, we 0:20:07.364 --> 0:20:10.004 have to be regulated. We can't have any and pop 0:20:10.364 --> 0:20:14.764 Shop burning an AI model, so therefore just just us three, right, 0:20:14.884 --> 0:20:18.244 just as Google and a thropic, and therefore we capture 0:20:18.324 --> 0:20:19.924 all the market share in perpetuity. 0:20:20.444 --> 0:20:24.444 Yeah. So so the the incentives here are definitely not 0:20:24.724 --> 0:20:27.404 as aligned as as one might think. And I think 0:20:27.444 --> 0:20:31.164 you you're also kind of subtly raising another point here, 0:20:31.564 --> 0:20:35.764 which is that in general, and stop me if you disagree, 0:20:35.804 --> 0:20:38.924 But it seems like over the last year, I sure 0:20:38.964 --> 0:20:41.884 like people are you know, worried about AI and like 0:20:41.924 --> 0:20:43.444 p doom and all this, but it seems like that 0:20:43.564 --> 0:20:47.684 discourse has actually quieted down and instead people are like, yeah, 0:20:47.724 --> 0:20:50.844 let's let's be more competitive with China. Okay, fuck it, Like, 0:20:50.964 --> 0:20:53.524 let's let's let's try to push to like do what 0:20:53.604 --> 0:20:56.764 we can to like to make it really good. And 0:20:56.844 --> 0:21:00.204 I think that people have actually just maybe it's because 0:21:00.204 --> 0:21:03.244 they've become, you know, just more used to having AI 0:21:03.364 --> 0:21:07.124 whatever it is. It feels like the safety concerns have 0:21:07.284 --> 0:21:14.484 taken kind of a secondary role to capitalistic and competitive instincts. 0:21:14.764 --> 0:21:18.404 I think there is maybe some view that, like this 0:21:18.524 --> 0:21:21.804 year has not seen as much progress in AI timelines 0:21:21.924 --> 0:21:24.924 as people would have thought back in January, that might 0:21:24.964 --> 0:21:27.004 be debated. I mean, I think people are also like 0:21:28.564 --> 0:21:30.724 a little reluctant to talk about this in the AI 0:21:30.724 --> 0:21:32.444 safety community because they don't want to make it seem 0:21:32.564 --> 0:21:36.364 like their guards are down right, But that's my sense, 0:21:36.444 --> 0:21:40.044 right almost more about what's being kind of on said now. Also, 0:21:40.084 --> 0:21:41.844 people can get bored of saying the same thing over 0:21:41.884 --> 0:21:44.284 and over. There's a new book out very soon about 0:21:44.284 --> 0:21:46.484 how AI is going to kill everybody if we don't 0:21:46.524 --> 0:21:49.044 learn how to control it quickly, right from people Olosio 0:21:49.084 --> 0:21:52.844 Yukowski and need Sours who are doomers, but very smart 0:21:52.884 --> 0:21:55.484 people widely respect to me, we'll have one of them 0:21:55.484 --> 0:21:58.284 on at some point. I don't know, right, but yeah, 0:21:58.364 --> 0:22:00.524 and I think it's kind of like also maybe a 0:22:00.604 --> 0:22:04.724 view that like, Okay, this technology is interesting and economically 0:22:05.364 --> 0:22:07.804 powerful and kind of fun, right, so we can't talk 0:22:07.844 --> 0:22:08.764 about the same thing all the time. 0:22:08.764 --> 0:22:10.284 I don't know, it's my that's my impression. 0:22:10.844 --> 0:22:14.284 Yeah, whatever, whatever the cases, whatever's going on behind the scenes. 0:22:14.524 --> 0:22:17.524