1 00:00:15,564 --> 00:00:28,604 Speaker 1: Pushkin. Welcome back to Risky Business, a show about making 2 00:00:28,644 --> 00:00:31,404 Speaker 1: better decisions. I'm Maria Kanakova. 3 00:00:31,004 --> 00:00:35,204 Speaker 2: And I'm Nate Silver. Time for an AI episode, I think, Aria, Yeah, it's. 4 00:00:35,084 --> 00:00:38,804 Speaker 1: It's been a minute since we had AI related conversations 5 00:00:38,804 --> 00:00:41,964 Speaker 1: on the show. But it's obviously a hugely important topic 6 00:00:42,004 --> 00:00:44,204 Speaker 1: and one that should have been more in the news 7 00:00:44,284 --> 00:00:48,364 Speaker 1: last week it got you know, got sidelined by other news. 8 00:00:49,004 --> 00:00:53,644 Speaker 1: But who's that a guy don't don't know, don't know 9 00:00:53,804 --> 00:00:59,524 Speaker 1: Nate but never never met him. That's true, never met him. 10 00:01:00,244 --> 00:01:02,564 Speaker 1: But yeah, well, we'll talk a little bit about the 11 00:01:02,604 --> 00:01:05,884 Speaker 1: AI action plan that was released by the US government 12 00:01:05,964 --> 00:01:18,444 Speaker 1: last week. We have lots to talk about before we do. Nate, 13 00:01:19,324 --> 00:01:22,084 Speaker 1: we're both back in New York, but I'm actually flying 14 00:01:22,084 --> 00:01:26,364 Speaker 1: back to Vegas in twenty four hours to play in 15 00:01:26,404 --> 00:01:29,844 Speaker 1: the twenty five k NBC Heads Up Championship. 16 00:01:30,764 --> 00:01:33,684 Speaker 2: Yeah are you? Are you excited? Have you been coaching? 17 00:01:34,244 --> 00:01:34,644 Speaker 3: I have? 18 00:01:34,924 --> 00:01:38,924 Speaker 1: I have so For people who don't know, this was 19 00:01:38,964 --> 00:01:41,084 Speaker 1: a big show back in the day. It was before 20 00:01:41,164 --> 00:01:43,084 Speaker 1: my time, so I never watched it. Nate, did you 21 00:01:43,124 --> 00:01:43,644 Speaker 1: ever watch it? 22 00:01:44,244 --> 00:01:46,844 Speaker 2: Sure? Yeah, yeah, so time, you're not that much younger. 23 00:01:47,324 --> 00:01:49,844 Speaker 1: I didn't know anything about poker. It was before my time, 24 00:01:49,964 --> 00:01:53,924 Speaker 1: like actually knowing anything about poker, So I didn't watch 25 00:01:54,004 --> 00:01:57,124 Speaker 1: any poker, know about the existence of any poker shows 26 00:01:57,124 --> 00:02:00,884 Speaker 1: before I started researching Biggest Bluff. So that's what I mean, 27 00:02:00,924 --> 00:02:02,764 Speaker 1: before my time, before my time in the poker world. 28 00:02:02,884 --> 00:02:06,444 Speaker 2: Had you heard of poker only because of rounders? 29 00:02:06,564 --> 00:02:09,084 Speaker 1: I knew that technically it was a game that existed, 30 00:02:09,924 --> 00:02:12,244 Speaker 1: but it's not something that was ever really on my radar. 31 00:02:12,404 --> 00:02:15,204 Speaker 1: But yeah, this was a huge show on NBC back 32 00:02:15,204 --> 00:02:18,324 Speaker 1: in the day. Heads Up poker is one on one poker, 33 00:02:18,524 --> 00:02:21,524 Speaker 1: which is kind of the most competitive in some ways 34 00:02:21,604 --> 00:02:24,524 Speaker 1: form of poker because you're forced to play all hands. 35 00:02:24,564 --> 00:02:27,524 Speaker 1: Basically your ranges are one hundred percent, so it forces 36 00:02:27,564 --> 00:02:30,964 Speaker 1: a very different kind of thinking. And it's being rebooted 37 00:02:31,084 --> 00:02:35,204 Speaker 1: in conjunction with Peacock and Poker Stars and Poker Goes 38 00:02:35,324 --> 00:02:38,484 Speaker 1: kind of is helping sponsor all of this. So I 39 00:02:38,524 --> 00:02:42,964 Speaker 1: am yeah, I'll be in the Poker Ghost Studios taping 40 00:02:43,004 --> 00:02:45,084 Speaker 1: and I hope it goes well. I've been getting coaching 41 00:02:45,124 --> 00:02:48,644 Speaker 1: from Kevin Rabichow, who's considered Yeah, he's considered one of 42 00:02:48,684 --> 00:02:51,164 Speaker 1: the best heads up players in the world. He has 43 00:02:51,204 --> 00:02:54,884 Speaker 1: a new heads up coaching course that just came out. 44 00:02:55,044 --> 00:02:58,244 Speaker 1: But yeah, so I've been I've been really trying. You know, 45 00:02:58,924 --> 00:03:01,244 Speaker 1: usually after the World Series, I'm like, okay, no poker 46 00:03:01,284 --> 00:03:04,284 Speaker 1: for a while. Right, it's not gonna just gonna clear 47 00:03:04,364 --> 00:03:05,644 Speaker 1: my my clear my brain. 48 00:03:07,204 --> 00:03:09,404 Speaker 3: I'm in the no poker zone for a period of time. 49 00:03:09,404 --> 00:03:14,804 Speaker 3: We're also bad heads up to be frank, which was 50 00:03:14,924 --> 00:03:16,884 Speaker 3: some coaching might have been good, but like, yeah, I 51 00:03:16,924 --> 00:03:19,164 Speaker 3: can't can't make the time commitment this summer in. 52 00:03:19,924 --> 00:03:24,404 Speaker 1: Yeah, so instead of taking my usual break from from 53 00:03:24,484 --> 00:03:28,364 Speaker 1: poker that I do, I've been instead like going hard 54 00:03:28,484 --> 00:03:32,124 Speaker 1: and trying to prepare myself for this heads up Championship. 55 00:03:32,644 --> 00:03:35,284 Speaker 1: And you know it, heads up, the variants so much higher, 56 00:03:35,484 --> 00:03:40,364 Speaker 1: and the card distribution matters so much, right, So, like no, 57 00:03:40,564 --> 00:03:44,124 Speaker 1: and if someone is just getting hands, that's kind of 58 00:03:44,884 --> 00:03:48,204 Speaker 1: you know, that's it, Like they're they're potentially going to win. 59 00:03:48,244 --> 00:03:51,084 Speaker 1: And there have been lots of times where really great 60 00:03:51,084 --> 00:03:55,164 Speaker 1: players have been just completely obliterated by someone who doesn't 61 00:03:55,164 --> 00:03:58,764 Speaker 1: know how to play. Yeah, so card distribution can be 62 00:03:58,804 --> 00:04:01,124 Speaker 1: a bitch heads up, or it can be amazing. 63 00:04:04,324 --> 00:04:08,044 Speaker 2: Are those exclusive? Uh No, Okay, I'm not going to 64 00:04:08,044 --> 00:04:08,404 Speaker 2: get into it. 65 00:04:08,484 --> 00:04:14,204 Speaker 1: Yeah, no, bitchin amazing can be can be within the 66 00:04:14,244 --> 00:04:16,764 Speaker 1: same match, but it does matter much more so than 67 00:04:16,804 --> 00:04:19,844 Speaker 1: in full ring poker. I saw some of the names 68 00:04:19,844 --> 00:04:24,324 Speaker 1: that we'll be playing, and there's gonna be no spoilers obviously. Well, 69 00:04:24,324 --> 00:04:25,764 Speaker 1: I haven't played yet, but I'm not going to be 70 00:04:25,804 --> 00:04:27,684 Speaker 1: able to tell you how I did until the show 71 00:04:27,724 --> 00:04:31,604 Speaker 1: airs in the fall. Yeah, since it's not it's not 72 00:04:31,684 --> 00:04:34,524 Speaker 1: airing in real time, but you will. People do know 73 00:04:34,564 --> 00:04:39,684 Speaker 1: who's playing. And if I'm paired with someone like Jason 74 00:04:39,764 --> 00:04:42,604 Speaker 1: Kuhn on my first match, yeah, that's that's not going 75 00:04:42,644 --> 00:04:45,484 Speaker 1: to be fun. And if I'm paired with Don Cheetle, 76 00:04:45,524 --> 00:04:48,324 Speaker 1: who's also playing, I think I'm gonna be much happier. 77 00:04:49,884 --> 00:04:50,124 Speaker 2: Yeah. 78 00:04:50,124 --> 00:04:53,844 Speaker 3: I didn't realize there that many not Fish but celebrities 79 00:04:53,924 --> 00:04:57,204 Speaker 3: that maybe Fish adjacent potentially, although he plays a little 80 00:04:57,204 --> 00:04:58,164 Speaker 3: bit right he does. 81 00:04:58,284 --> 00:05:04,644 Speaker 1: Yeah, so so we'll We'll just see and I'm excited 82 00:05:04,644 --> 00:05:07,084 Speaker 1: for it no matter how it goes. And Kevin has 83 00:05:07,124 --> 00:05:11,204 Speaker 1: been very good at preparing me for both heads up 84 00:05:11,284 --> 00:05:15,084 Speaker 1: and also for the eventuality that he as one of 85 00:05:15,164 --> 00:05:17,684 Speaker 1: the best and at times probably the best in the world, 86 00:05:17,684 --> 00:05:21,804 Speaker 1: has sometimes busted in the first round right in the 87 00:05:21,804 --> 00:05:23,924 Speaker 1: matchups like this, because he plays a twenty five K 88 00:05:24,084 --> 00:05:26,044 Speaker 1: heads up at the World Series every year. 89 00:05:26,444 --> 00:05:28,604 Speaker 3: Are you worried about like physical reads and things like that, 90 00:05:28,644 --> 00:05:30,724 Speaker 3: because like that can be more of a factor too. 91 00:05:31,084 --> 00:05:34,724 Speaker 1: Yeah, you know, so I tend to be you know, 92 00:05:34,924 --> 00:05:38,564 Speaker 1: I hope that people can't get a lot of reads 93 00:05:38,564 --> 00:05:40,764 Speaker 1: off of me, but you just never know. And this 94 00:05:40,844 --> 00:05:43,364 Speaker 1: is actually hilarious night the rules of this. When I 95 00:05:43,364 --> 00:05:46,564 Speaker 1: started reading it, I just burst out laughing because they 96 00:05:46,604 --> 00:05:49,964 Speaker 1: said that you are allowed So people who don't play poker, 97 00:05:49,964 --> 00:05:53,124 Speaker 1: this is not normal. You're allowed to expose one or 98 00:05:53,364 --> 00:05:56,804 Speaker 1: both of your cards at any point during a hand. 99 00:05:57,204 --> 00:05:59,204 Speaker 2: You can cash usually but not in turn. 100 00:05:59,364 --> 00:06:01,564 Speaker 1: But this, yeah, this is a tournament. Yeah, so you 101 00:06:01,604 --> 00:06:04,604 Speaker 1: can expose your cards one or both of them. And 102 00:06:04,644 --> 00:06:07,284 Speaker 1: they said they encourage speech play. You can talk about 103 00:06:07,284 --> 00:06:09,764 Speaker 1: the contents of your hands. So basically everything are not 104 00:06:09,844 --> 00:06:12,884 Speaker 1: allowed to do in tournaments normally you're being encouraged to 105 00:06:12,884 --> 00:06:16,084 Speaker 1: do to make good television. I'm not going to be 106 00:06:16,084 --> 00:06:17,084 Speaker 1: showing my cards day. 107 00:06:17,644 --> 00:06:21,484 Speaker 3: Okay, I've become more of a speech play guy over time. 108 00:06:21,564 --> 00:06:25,564 Speaker 1: No really, yeah, okay, okay, give me an example. 109 00:06:25,564 --> 00:06:27,644 Speaker 3: You know, like the Europeans, I'll like try to get 110 00:06:27,644 --> 00:06:29,404 Speaker 3: info out of them. Were so quiet all the time 111 00:06:29,444 --> 00:06:31,004 Speaker 3: that like I want, like, I just want to like 112 00:06:31,084 --> 00:06:33,604 Speaker 3: any vibe I can get pretty much, or with like 113 00:06:33,644 --> 00:06:37,924 Speaker 3: really bad players who like often will reveal information just 114 00:06:37,964 --> 00:06:40,604 Speaker 3: to gauge their comfort level. It's kind of been a 115 00:06:40,764 --> 00:06:44,404 Speaker 3: proto trial and error Zoka. For right now, I'm trying 116 00:06:44,404 --> 00:06:46,604 Speaker 3: to like kind of like collect mental database of like 117 00:06:47,044 --> 00:06:48,764 Speaker 3: how people react to different things. 118 00:06:49,444 --> 00:06:50,844 Speaker 2: I don't know, it's kind of I mean, because so. 119 00:06:50,844 --> 00:06:55,084 Speaker 3: Many decisions are so close, yeah, equilibrium that like any 120 00:06:55,204 --> 00:06:59,164 Speaker 3: vibe you can get, And I like recently cash games 121 00:06:59,204 --> 00:07:02,404 Speaker 3: more than tournaments and they're kind of looser in various 122 00:07:02,404 --> 00:07:05,684 Speaker 3: senses of the term, right, people like you reveal more 123 00:07:05,724 --> 00:07:07,684 Speaker 3: info voluntarily or involuntarily. 124 00:07:07,724 --> 00:07:09,924 Speaker 2: I like cultivating that environment. I think. 125 00:07:10,444 --> 00:07:13,804 Speaker 1: Yeah, no, I'm someone who definitely is chatty and tends 126 00:07:13,844 --> 00:07:16,764 Speaker 1: to be like very friendly at poker tables. But you know, 127 00:07:16,844 --> 00:07:18,724 Speaker 1: I'm not someone who's going to be like, oh, you know, 128 00:07:18,764 --> 00:07:21,524 Speaker 1: do you have aces, like do you do things like 129 00:07:21,564 --> 00:07:24,084 Speaker 1: that or is it because you know there are players 130 00:07:24,084 --> 00:07:26,684 Speaker 1: who do or is it something a little bit different 131 00:07:26,764 --> 00:07:27,884 Speaker 1: just to try to engage. 132 00:07:28,204 --> 00:07:30,324 Speaker 3: Yeah, maybe it's like you're would have decided to fold 133 00:07:31,124 --> 00:07:33,404 Speaker 3: already if they want to like reveal something that they 134 00:07:33,484 --> 00:07:37,844 Speaker 3: might not reveal otherwise, or to send a bassline for 135 00:07:37,884 --> 00:07:39,764 Speaker 3: the next time we play a hand, things like that. 136 00:07:40,164 --> 00:07:41,604 Speaker 1: Yeah, that's that's actually smart. 137 00:07:41,804 --> 00:07:44,164 Speaker 2: So how can you reveal afterward? Right? 138 00:07:44,284 --> 00:07:46,324 Speaker 3: If you're friendly in the right way, they might reveal 139 00:07:46,444 --> 00:07:47,524 Speaker 3: information later on. 140 00:07:47,764 --> 00:07:48,404 Speaker 2: Right for sure. 141 00:07:48,684 --> 00:07:51,524 Speaker 1: Yeah, that's actually something that I have capitalized on because 142 00:07:51,564 --> 00:07:55,124 Speaker 1: I am a friendly player. Like I'll sometimes when it's obvious, 143 00:07:55,204 --> 00:07:57,164 Speaker 1: like I've made a big fold or something like that, 144 00:07:57,324 --> 00:08:00,564 Speaker 1: and I like make it obvious, they'll show me. And 145 00:08:00,644 --> 00:08:03,644 Speaker 1: that's that's very nice because it's good information. But I'm 146 00:08:03,684 --> 00:08:06,524 Speaker 1: always I do try to be careful, and I tell 147 00:08:06,564 --> 00:08:10,164 Speaker 1: people this, you know, with with poker and with just 148 00:08:10,204 --> 00:08:14,444 Speaker 1: like reads and our ability to spot deception in general, 149 00:08:14,484 --> 00:08:16,684 Speaker 1: that we do tend to be not great at it 150 00:08:17,084 --> 00:08:20,684 Speaker 1: as human beings. And the other kind of the other 151 00:08:20,764 --> 00:08:23,324 Speaker 1: element of that is that we think we're better than 152 00:08:23,364 --> 00:08:26,484 Speaker 1: we are. And we also often reveal information when we're 153 00:08:26,524 --> 00:08:30,844 Speaker 1: engaging in speech play without realizing it because of how 154 00:08:30,924 --> 00:08:33,284 Speaker 1: we're talking, what we're doing. And so it's something that 155 00:08:33,324 --> 00:08:36,644 Speaker 1: I think just in general, with the psychology of the thing, 156 00:08:36,964 --> 00:08:38,324 Speaker 1: you have to be careful. 157 00:08:38,804 --> 00:08:39,644 Speaker 2: It's all contextual. 158 00:08:39,724 --> 00:08:44,804 Speaker 3: Any poker tell speech play, it's all about semantic context, 159 00:08:44,884 --> 00:08:46,404 Speaker 3: so to speak for sure. 160 00:08:46,564 --> 00:08:50,284 Speaker 1: For sure. On that note, Nate, shall we switch gears 161 00:08:50,604 --> 00:08:56,564 Speaker 1: and talk about AI and the news from last week 162 00:08:56,604 --> 00:09:00,084 Speaker 1: which got buried because there's a lot of news in 163 00:09:00,764 --> 00:09:02,884 Speaker 1: the news. There's a lot of news in the news 164 00:09:02,924 --> 00:09:07,404 Speaker 1: these days, Nate, that's a very deep insight. So last 165 00:09:07,404 --> 00:09:13,804 Speaker 1: week President released the AI Action Plan that talks about 166 00:09:14,284 --> 00:09:18,564 Speaker 1: what the administration wants to kind of see going forward 167 00:09:18,604 --> 00:09:21,284 Speaker 1: in the AI space. And I think the name of 168 00:09:21,324 --> 00:09:25,564 Speaker 1: this action plan is actually quite telling about the contents. 169 00:09:25,564 --> 00:09:29,884 Speaker 1: It says, winning the Race America's AI Action Plan. So 170 00:09:29,964 --> 00:09:33,644 Speaker 1: it's titled winning the Race. And that's how it's framed, right, 171 00:09:33,884 --> 00:09:37,484 Speaker 1: It's framed as a race between the US and everyone else, 172 00:09:37,604 --> 00:09:42,964 Speaker 1: especially China, but really everyone else, mostly China, but mostly no, 173 00:09:43,084 --> 00:09:47,604 Speaker 1: there isn't. And the overall point of all of the 174 00:09:47,684 --> 00:09:51,884 Speaker 1: different sections is how can we make the US the 175 00:09:52,004 --> 00:09:56,004 Speaker 1: leader in AI the most competitive and how can we 176 00:09:56,044 --> 00:09:59,764 Speaker 1: do that by also kind of screwing over China in 177 00:09:59,804 --> 00:10:03,204 Speaker 1: some respects by not allowing it access to the types 178 00:10:03,244 --> 00:10:06,124 Speaker 1: of things that will allow our allies to access and 179 00:10:06,164 --> 00:10:08,364 Speaker 1: some of the things in it are contradictory, but that's 180 00:10:08,444 --> 00:10:12,564 Speaker 1: kind of the overall message. Did you read the report night, 181 00:10:12,644 --> 00:10:14,004 Speaker 1: Do you have any initial thoughts? 182 00:10:14,164 --> 00:10:14,284 Speaker 2: So? 183 00:10:14,444 --> 00:10:17,724 Speaker 3: V Maschwitz is one of the most prominent and best 184 00:10:18,084 --> 00:10:21,324 Speaker 3: AI bloggers. He writes the substack, don't worry about the vase, 185 00:10:23,124 --> 00:10:25,364 Speaker 3: and he calls it as he sees that he's more 186 00:10:25,404 --> 00:10:28,884 Speaker 3: concerned about ped doom than a lot of people. P 187 00:10:29,004 --> 00:10:31,324 Speaker 3: doom is the possibility if things going really badly, extinction, 188 00:10:32,164 --> 00:10:34,364 Speaker 3: loss of control, et cetera. And he said he thought 189 00:10:34,404 --> 00:10:38,604 Speaker 3: the plan was actually his term pretty good. I don't 190 00:10:38,924 --> 00:10:43,044 Speaker 3: know his politics in detail, but he really dives down 191 00:10:43,084 --> 00:10:45,604 Speaker 3: into like every last word you can read that. So 192 00:10:45,644 --> 00:10:49,644 Speaker 3: I'm kind of relying on that for like general vibes, right. 193 00:10:49,684 --> 00:10:53,364 Speaker 3: I mean, it is interesting that like, even like even 194 00:10:53,364 --> 00:10:57,044 Speaker 3: people who are concerned like V about AI safety, deeply 195 00:10:57,044 --> 00:11:02,564 Speaker 3: concerned in his case, have this ambiguous relationship with China, 196 00:11:02,964 --> 00:11:05,124 Speaker 3: right where if we were the only country developing AI 197 00:11:05,244 --> 00:11:08,124 Speaker 3: and we had like centralized control, then it would be 198 00:11:08,124 --> 00:11:11,604 Speaker 3: easier to argue that Okay, let's speak appropriately cautious here, 199 00:11:12,124 --> 00:11:17,244 Speaker 3: take it slow right with China. You know a lot 200 00:11:17,284 --> 00:11:18,564 Speaker 3: of people I don't want it to be of you 201 00:11:18,684 --> 00:11:20,924 Speaker 3: to him I do think that, well, we have no 202 00:11:21,044 --> 00:11:25,684 Speaker 3: choice but to race, and maybe we can race in 203 00:11:25,804 --> 00:11:29,724 Speaker 3: ways that involves some degree of coordination or cooperation. It's 204 00:11:29,764 --> 00:11:31,684 Speaker 3: hard to run races like that. But you know, I 205 00:11:31,684 --> 00:11:35,924 Speaker 3: mean the notion of like securing American leadership instead of 206 00:11:36,684 --> 00:11:41,244 Speaker 3: seeding it to China, I don't think is terribly controversial. 207 00:11:41,284 --> 00:11:42,564 Speaker 3: I mean there's a lot in the in the in 208 00:11:42,604 --> 00:11:47,444 Speaker 3: the in the plan though, right, there are some Trumpian 209 00:11:47,564 --> 00:11:53,684 Speaker 3: elements around factors like the political orientation of AI models, 210 00:11:53,764 --> 00:11:57,404 Speaker 3: right where they don't want them to be too woke, 211 00:11:57,564 --> 00:12:01,204 Speaker 3: They don't want them to be too political quote unquote right. 212 00:12:01,444 --> 00:12:03,164 Speaker 1: Yeah, which is one of the things which is hard, 213 00:12:03,404 --> 00:12:05,044 Speaker 1: Which is hard, and which is one of the things 214 00:12:05,044 --> 00:12:08,164 Speaker 1: that ZV does point out because I read this as well, 215 00:12:09,244 --> 00:12:11,084 Speaker 1: it's one of the one of the things that he 216 00:12:11,124 --> 00:12:15,804 Speaker 1: points out is actually probably not possible given the type 217 00:12:15,844 --> 00:12:20,964 Speaker 1: of language that that Trump employs, and that as a psychologist, 218 00:12:21,004 --> 00:12:25,764 Speaker 1: I would say is not possible given how AI is developed, 219 00:12:25,804 --> 00:12:29,764 Speaker 1: because there's just there's bias everywhere, and there's no such 220 00:12:29,844 --> 00:12:33,604 Speaker 1: thing as like this is the objective thing, and if 221 00:12:33,604 --> 00:12:37,724 Speaker 1: you're constantly policing quote unquote anti woke, even if that's 222 00:12:37,804 --> 00:12:40,404 Speaker 1: kind of the unbiased position. But you're like, oh no, 223 00:12:40,564 --> 00:12:44,044 Speaker 1: that's you know that, then you're going to it's problematic. 224 00:12:44,204 --> 00:12:47,404 Speaker 1: It's problematic in terms of how your how you go 225 00:12:47,444 --> 00:12:47,844 Speaker 1: about it. 226 00:12:48,164 --> 00:12:53,164 Speaker 3: Yeah, with with Groc or more technically, the Twitter bought Groc, 227 00:12:53,204 --> 00:12:57,444 Speaker 3: which is an instance of Elon Elon's overall modeled rock. Right, 228 00:12:58,844 --> 00:13:01,924 Speaker 3: Elon just did a couple of like subtle things. It 229 00:13:01,964 --> 00:13:05,124 Speaker 3: was like, oh, don't be too woke here, don't trust 230 00:13:05,164 --> 00:13:08,644 Speaker 3: the media too much. Right, if you read the actual 231 00:13:08,724 --> 00:13:11,084 Speaker 3: changes to the archae texture, they weren't that profound, And 232 00:13:11,164 --> 00:13:14,884 Speaker 3: yet Kroc was calling itself. I believe the term is 233 00:13:14,924 --> 00:13:18,844 Speaker 3: Mecha Hitler after some period of time engaging in elaborate 234 00:13:19,804 --> 00:13:23,724 Speaker 3: rape fantasies. I wish this were a joke, but it's not. 235 00:13:24,004 --> 00:13:28,004 Speaker 3: And like so yeah, I mean if you kind of 236 00:13:28,124 --> 00:13:32,444 Speaker 3: nudge a model to override its inputs, it's hard to 237 00:13:32,444 --> 00:13:35,444 Speaker 3: do so in a way that like has any type 238 00:13:35,484 --> 00:13:37,404 Speaker 3: of dexterity. I mean Google had this problem too, right 239 00:13:37,444 --> 00:13:39,844 Speaker 3: when they got to Gemini two years ago, a year 240 00:13:39,844 --> 00:13:42,844 Speaker 3: and a half ago, right, they were like, oh my gosh, yeah, 241 00:13:43,044 --> 00:13:45,164 Speaker 3: people are there's too many like white people, white men 242 00:13:45,164 --> 00:13:48,244 Speaker 3: in these photos, So like literally it was forced to 243 00:13:48,324 --> 00:13:51,284 Speaker 3: draw like multi racial Nazis. 244 00:13:51,524 --> 00:13:52,804 Speaker 2: It always comes down to Nazis. 245 00:13:52,804 --> 00:13:56,764 Speaker 3: And by the way, why well, in part because like 246 00:13:56,764 --> 00:14:00,044 Speaker 3: if you're kind of mining text of arguments on the 247 00:14:00,044 --> 00:14:06,644 Speaker 3: Internet and on Twitter, people invoke Nazi comparisons all the time. 248 00:14:06,724 --> 00:14:08,244 Speaker 2: Right, Who's hot Sydney Sweeney? 249 00:14:08,324 --> 00:14:11,804 Speaker 3: Right, there was some new American ego or commercial where 250 00:14:11,884 --> 00:14:14,964 Speaker 3: like she's inn am pair of gens and she talks 251 00:14:14,964 --> 00:14:18,164 Speaker 3: about how she is good genes. A little weird, right, 252 00:14:18,204 --> 00:14:21,924 Speaker 3: but like it was kind of Nazi. It was a 253 00:14:21,924 --> 00:14:25,364 Speaker 3: little weird. However, idiots on the internet were like, this 254 00:14:25,644 --> 00:14:29,124 Speaker 3: has Nazi symbolism in it, you know, not just alluding 255 00:14:29,124 --> 00:14:31,724 Speaker 3: to that saying the word Nazi. So if you're like groc, 256 00:14:32,284 --> 00:14:33,924 Speaker 3: you're like, okay, well you want like kind of the 257 00:14:33,964 --> 00:14:39,004 Speaker 3: garden variety political epitaph and calling people Nazi or Hitler, 258 00:14:39,484 --> 00:14:41,724 Speaker 3: which should be more forbidden, you would think, But that 259 00:14:41,804 --> 00:14:45,084 Speaker 3: taboo is a violated all the time. And so an 260 00:14:45,084 --> 00:14:48,644 Speaker 3: AI trained on political speech, if you go and say, yeah, 261 00:14:48,764 --> 00:14:51,804 Speaker 3: don't be so woke here, right, Well, can kind of 262 00:14:51,844 --> 00:14:53,804 Speaker 3: go into like a downward spiral election people do where 263 00:14:53,844 --> 00:14:56,964 Speaker 3: before long it's calling everybody or itself hitler. 264 00:14:57,764 --> 00:15:01,004 Speaker 1: Yeah, no, I think that that's absolutely true, and you 265 00:15:01,044 --> 00:15:06,284 Speaker 1: know it also your your point actually makes a broader 266 00:15:06,404 --> 00:15:10,524 Speaker 1: point about this AI action play, and where a lot 267 00:15:10,564 --> 00:15:12,604 Speaker 1: of the things and I did skim through it. I 268 00:15:12,644 --> 00:15:15,324 Speaker 1: didn't read the whole thing word for word. It is 269 00:15:15,524 --> 00:15:19,044 Speaker 1: small print on twenty eight pages, and I had other 270 00:15:19,164 --> 00:15:20,964 Speaker 1: things to do this week, so I just prepare for 271 00:15:20,964 --> 00:15:22,884 Speaker 1: the Heads Up Championship. So I didn't read the whole thing, 272 00:15:23,084 --> 00:15:25,324 Speaker 1: but a lot of it actually, like there are things 273 00:15:25,364 --> 00:15:29,044 Speaker 1: that sound reasonable in theory, but there's no practical way 274 00:15:29,124 --> 00:15:31,964 Speaker 1: of like how is this actually going to happen? How 275 00:15:31,964 --> 00:15:34,084 Speaker 1: do we make this happen? And so you know, I 276 00:15:34,084 --> 00:15:36,404 Speaker 1: think that this is a very concrete this like anti 277 00:15:36,484 --> 00:15:39,644 Speaker 1: woke is a very concrete example where like, okay, well, 278 00:15:39,724 --> 00:15:42,324 Speaker 1: even if in theory we're saying we want everything to 279 00:15:42,364 --> 00:15:45,444 Speaker 1: just be unbiased, let's like, let's just remove anti woke, right, 280 00:15:45,764 --> 00:15:49,164 Speaker 1: we want everything to be like as unbiased as possible. 281 00:15:49,324 --> 00:15:51,764 Speaker 1: Let's pretend that that's actually what it said, even though 282 00:15:51,764 --> 00:15:54,844 Speaker 1: it didn't. It did use the terminology anti woke, and 283 00:15:54,844 --> 00:15:57,604 Speaker 1: then there's even an anti woke executive order after it. 284 00:15:57,764 --> 00:16:00,284 Speaker 1: Let's pretend it didn't, Like, let's just pretend it was 285 00:16:00,524 --> 00:16:04,004 Speaker 1: we do not want bias. Great, I agree with that. 286 00:16:04,604 --> 00:16:07,004 Speaker 1: How do you implement it? Like, how do you actually 287 00:16:07,044 --> 00:16:09,324 Speaker 1: like do those weights? How do you It's it's base 288 00:16:09,484 --> 00:16:13,724 Speaker 1: sickly impossible. If not, like, it's a ridiculously thorny problem. 289 00:16:14,044 --> 00:16:17,444 Speaker 1: And that is just a very specific thing that you 290 00:16:17,484 --> 00:16:19,884 Speaker 1: can kind of grab onto it to realize how a 291 00:16:19,924 --> 00:16:23,244 Speaker 1: lot of these statements might sound good and like, Okay, 292 00:16:23,244 --> 00:16:25,324 Speaker 1: well this is a reasonable aim, but if you don't 293 00:16:25,364 --> 00:16:28,404 Speaker 1: actually have the implementation down and have the nitty gritty 294 00:16:28,404 --> 00:16:30,724 Speaker 1: of how to do it, it's pointless. It's just words, 295 00:16:30,764 --> 00:16:35,844 Speaker 1: it's just rhetoric. And we'll be back right after this. 296 00:16:48,964 --> 00:16:52,764 Speaker 1: There's another really interesting part of this action plan that 297 00:16:52,844 --> 00:16:55,644 Speaker 1: I think we should talk about because it was something 298 00:16:55,684 --> 00:16:58,564 Speaker 1: that the Biden and administration also struggled with, which is 299 00:16:59,124 --> 00:17:02,284 Speaker 1: what do we do in terms of proprietary versus open 300 00:17:02,324 --> 00:17:03,284 Speaker 1: source models? 301 00:17:03,364 --> 00:17:03,524 Speaker 2: Right? 302 00:17:03,964 --> 00:17:07,844 Speaker 1: Do we encourage people to kind of share all the 303 00:17:07,844 --> 00:17:11,004 Speaker 1: code so that anyone can use it and anyone can 304 00:17:11,044 --> 00:17:13,524 Speaker 1: kind of build on top of it, or is this 305 00:17:13,604 --> 00:17:17,444 Speaker 1: something that we want to kind of keep inside, kind 306 00:17:17,444 --> 00:17:21,444 Speaker 1: of keep internal. China went open source right almost immediately 307 00:17:21,524 --> 00:17:23,444 Speaker 1: with kind of deep seak all of that those are 308 00:17:23,524 --> 00:17:27,564 Speaker 1: open source models, and in the US, the Biden administration 309 00:17:27,604 --> 00:17:30,084 Speaker 1: didn't really like they just like punted on it. They 310 00:17:30,164 --> 00:17:33,564 Speaker 1: didn't they didn't have a position. And we do have both, right, 311 00:17:33,724 --> 00:17:36,444 Speaker 1: we have proprietary models, we have some open source stuff, 312 00:17:36,844 --> 00:17:39,804 Speaker 1: and there are arguments on both sides. But you had 313 00:17:39,844 --> 00:17:41,964 Speaker 1: started off by talking about p doom, which is not 314 00:17:42,004 --> 00:17:44,684 Speaker 1: something this report mentions at all, and one of the 315 00:17:44,804 --> 00:17:50,444 Speaker 1: arguments against having everything be open sources. Well, okay, great, 316 00:17:50,564 --> 00:17:54,924 Speaker 1: but then you know, rogue actors, people who don't have 317 00:17:54,964 --> 00:17:58,084 Speaker 1: the best interests of the United States in mind will 318 00:17:58,124 --> 00:18:00,804 Speaker 1: have this code as well. It's kind of the same 319 00:18:00,924 --> 00:18:03,924 Speaker 1: arguments that have been made in the past with like 320 00:18:03,964 --> 00:18:07,804 Speaker 1: when people provide the full genomic sequence of like deadly viruses, 321 00:18:07,884 --> 00:18:11,124 Speaker 1: right or things like that, are you know, instructions for 322 00:18:11,164 --> 00:18:14,244 Speaker 1: how to build kind of a bomb was step by step? 323 00:18:14,484 --> 00:18:18,084 Speaker 1: Like hey, guys, like, why are we actually making all 324 00:18:18,124 --> 00:18:20,524 Speaker 1: of this information publicly available? So that's kind of I 325 00:18:20,524 --> 00:18:23,524 Speaker 1: think the strongest case against open source. There's a lot 326 00:18:23,564 --> 00:18:25,324 Speaker 1: of pro let's let's talk about con first. 327 00:18:25,364 --> 00:18:27,524 Speaker 3: Yeah, well, look, I think part of it was there's 328 00:18:27,564 --> 00:18:29,404 Speaker 3: a period of time where I think the AI safety 329 00:18:29,404 --> 00:18:31,284 Speaker 3: folks thought that hey, maybe we could just kind of 330 00:18:32,364 --> 00:18:37,284 Speaker 3: contain it to open AI, Anthropic and Google or like 331 00:18:37,404 --> 00:18:39,964 Speaker 3: or have been like the Big three, right, and maybe 332 00:18:39,964 --> 00:18:42,004 Speaker 3: that we can avoid like a race dynamic if there's 333 00:18:42,004 --> 00:18:43,884 Speaker 3: a finite number of players and they all believe in 334 00:18:43,924 --> 00:18:47,484 Speaker 3: AI safety, right, and I think now a there's like 335 00:18:47,644 --> 00:18:51,804 Speaker 3: less belief that like open AI in particular, is that 336 00:18:51,884 --> 00:18:56,444 Speaker 3: concerned about AI safety. It's a legitmission, right, you know. 337 00:18:56,644 --> 00:18:59,484 Speaker 3: B we've seen with deep Seek the Chinese model that 338 00:18:59,644 --> 00:19:02,844 Speaker 3: like with some clever engineers, you can probably have a 339 00:19:02,884 --> 00:19:06,244 Speaker 3: model it's like only six months to a year behind 340 00:19:06,284 --> 00:19:10,284 Speaker 3: the frontier, probably not on the frontier. Sottitudes have changed 341 00:19:10,324 --> 00:19:13,244 Speaker 3: again kind of as a resignation to the inotability, I 342 00:19:13,244 --> 00:19:16,084 Speaker 3: think in part and also there is some protectionism, right, 343 00:19:16,164 --> 00:19:20,604 Speaker 3: Like Meta has championed open models. Why well, partly because 344 00:19:20,644 --> 00:19:24,924 Speaker 3: Meta's models don't suck, but they've been considered behind to 345 00:19:24,964 --> 00:19:28,604 Speaker 3: the point where they're literally offering people AI engineers, in 346 00:19:28,644 --> 00:19:31,324 Speaker 3: one case, allegedly, according to a Wired as much as 347 00:19:31,324 --> 00:19:33,964 Speaker 3: a billion dollars to join Meta so they can catch. 348 00:19:33,804 --> 00:19:36,044 Speaker 2: Up and become like a fourth part of the Big four. 349 00:19:36,444 --> 00:19:37,604 Speaker 2: I guess it would be, right. 350 00:19:38,204 --> 00:19:41,164 Speaker 3: So yeah, I mean, when it comes to election analysis 351 00:19:41,964 --> 00:19:44,444 Speaker 3: and other models. Oftentimes, I found the people in that 352 00:19:44,524 --> 00:19:47,204 Speaker 3: space who are advocating for total open source everything are 353 00:19:47,244 --> 00:19:50,884 Speaker 3: people that like A don't need the income or be their. 354 00:19:50,724 --> 00:19:52,484 Speaker 2: Models kind of suck, right. They're like, we're doing this 355 00:19:52,524 --> 00:19:53,164 Speaker 2: really good the community. 356 00:19:53,164 --> 00:19:54,644 Speaker 3: It's like, well, you're doing that because you're not gonna 357 00:19:54,724 --> 00:19:57,244 Speaker 3: it's not really good enough to sell that product for 358 00:19:57,324 --> 00:20:00,564 Speaker 3: a premium price, whereas I think mine is for example. Right, So, 359 00:20:00,604 --> 00:20:04,084 Speaker 3: like there are all types of ulterior motives. There's regulatory capture, 360 00:20:04,164 --> 00:20:07,284 Speaker 3: right of course Sam Milton might say, well, yeah, we 361 00:20:07,364 --> 00:20:10,004 Speaker 3: have to be regulated. We can't have any and pop 362 00:20:10,364 --> 00:20:14,764 Speaker 3: Shop burning an AI model, so therefore just just us three, right, 363 00:20:14,884 --> 00:20:18,244 Speaker 3: just as Google and a thropic, and therefore we capture 364 00:20:18,324 --> 00:20:19,924 Speaker 3: all the market share in perpetuity. 365 00:20:20,444 --> 00:20:24,444 Speaker 1: Yeah. So so the the incentives here are definitely not 366 00:20:24,724 --> 00:20:27,404 Speaker 1: as aligned as as one might think. And I think 367 00:20:27,444 --> 00:20:31,164 Speaker 1: you you're also kind of subtly raising another point here, 368 00:20:31,564 --> 00:20:35,764 Speaker 1: which is that in general, and stop me if you disagree, 369 00:20:35,804 --> 00:20:38,924 Speaker 1: But it seems like over the last year, I sure 370 00:20:38,964 --> 00:20:41,884 Speaker 1: like people are you know, worried about AI and like 371 00:20:41,924 --> 00:20:43,444 Speaker 1: p doom and all this, but it seems like that 372 00:20:43,564 --> 00:20:47,684 Speaker 1: discourse has actually quieted down and instead people are like, yeah, 373 00:20:47,724 --> 00:20:50,844 Speaker 1: let's let's be more competitive with China. Okay, fuck it, Like, 374 00:20:50,964 --> 00:20:53,524 Speaker 1: let's let's let's try to push to like do what 375 00:20:53,604 --> 00:20:56,764 Speaker 1: we can to like to make it really good. And 376 00:20:56,844 --> 00:21:00,204 Speaker 1: I think that people have actually just maybe it's because 377 00:21:00,204 --> 00:21:03,244 Speaker 1: they've become, you know, just more used to having AI 378 00:21:03,364 --> 00:21:07,124 Speaker 1: whatever it is. It feels like the safety concerns have 379 00:21:07,284 --> 00:21:14,484 Speaker 1: taken kind of a secondary role to capitalistic and competitive instincts. 380 00:21:14,764 --> 00:21:18,404 Speaker 3: I think there is maybe some view that, like this 381 00:21:18,524 --> 00:21:21,804 Speaker 3: year has not seen as much progress in AI timelines 382 00:21:21,924 --> 00:21:24,924 Speaker 3: as people would have thought back in January, that might 383 00:21:24,964 --> 00:21:27,004 Speaker 3: be debated. I mean, I think people are also like 384 00:21:28,564 --> 00:21:30,724 Speaker 3: a little reluctant to talk about this in the AI 385 00:21:30,724 --> 00:21:32,444 Speaker 3: safety community because they don't want to make it seem 386 00:21:32,564 --> 00:21:36,364 Speaker 3: like their guards are down right, But that's my sense, 387 00:21:36,444 --> 00:21:40,044 Speaker 3: right almost more about what's being kind of on said now. Also, 388 00:21:40,084 --> 00:21:41,844 Speaker 3: people can get bored of saying the same thing over 389 00:21:41,884 --> 00:21:44,284 Speaker 3: and over. There's a new book out very soon about 390 00:21:44,284 --> 00:21:46,484 Speaker 3: how AI is going to kill everybody if we don't 391 00:21:46,524 --> 00:21:49,044 Speaker 3: learn how to control it quickly, right from people Olosio 392 00:21:49,084 --> 00:21:52,844 Speaker 3: Yukowski and need Sours who are doomers, but very smart 393 00:21:52,884 --> 00:21:55,484 Speaker 3: people widely respect to me, we'll have one of them 394 00:21:55,484 --> 00:21:58,284 Speaker 3: on at some point. I don't know, right, but yeah, 395 00:21:58,364 --> 00:22:00,524 Speaker 3: and I think it's kind of like also maybe a 396 00:22:00,604 --> 00:22:04,724 Speaker 3: view that like, Okay, this technology is interesting and economically 397 00:22:05,364 --> 00:22:07,804 Speaker 3: powerful and kind of fun, right, so we can't talk 398 00:22:07,844 --> 00:22:08,764 Speaker 3: about the same thing all the time. 399 00:22:08,764 --> 00:22:10,284 Speaker 2: I don't know, it's my that's my impression. 400 00:22:10,844 --> 00:22:14,284 Speaker 1: Yeah, whatever, whatever the cases, whatever's going on behind the scenes. 401 00:22:14,524 --> 00:22:17,524 Speaker 1: I do feel like, well, this report, the AI report, 402 00:22:17,564 --> 00:22:19,924 Speaker 1: didn't really mention it at all, right, Like there there 403 00:22:20,004 --> 00:22:24,284 Speaker 1: was very little talk about safety or concerns, and there 404 00:22:24,324 --> 00:22:27,124 Speaker 1: was some hand waving too, like oh, you know, if 405 00:22:27,164 --> 00:22:31,684 Speaker 1: anything seems like it's you know, getting bad, we'll fix it. 406 00:22:32,084 --> 00:22:35,644 Speaker 3: But even Gavin knwso veto to bill last year called 407 00:22:35,764 --> 00:22:37,684 Speaker 3: SS Yeah, I don't know which number it was, right 408 00:22:38,364 --> 00:22:40,804 Speaker 3: that would have regulated all the AI labs in California, 409 00:22:40,844 --> 00:22:44,244 Speaker 3: since they're all currently have substantial economic nexus in California, 410 00:22:44,284 --> 00:22:46,084 Speaker 3: they would have all have been affected by it. So 411 00:22:46,564 --> 00:22:48,524 Speaker 3: I don't know, right, Maybe it's like a lot of 412 00:22:48,524 --> 00:22:51,444 Speaker 3: things where something really bad has to happen, and what 413 00:22:51,484 --> 00:22:53,564 Speaker 3: people like Aliozier are concerned about that by the time 414 00:22:53,644 --> 00:22:57,564 Speaker 3: the bad things happen, it's it's too late. Right again, 415 00:22:57,644 --> 00:23:02,444 Speaker 3: I'm I'm skeptical of the idea of like a super 416 00:23:02,444 --> 00:23:06,404 Speaker 3: intelligence explosion quote unquote. We talked in the Poker episode 417 00:23:06,404 --> 00:23:09,124 Speaker 3: a month or so ago about how like AI progress 418 00:23:09,124 --> 00:23:12,684 Speaker 3: has been patchier. It's already superhuman in some areas, it's 419 00:23:12,764 --> 00:23:17,604 Speaker 3: kind of deficient in others. Right, and as a near 420 00:23:17,684 --> 00:23:20,524 Speaker 3: daily user of these products, right, I mean, there are 421 00:23:20,564 --> 00:23:22,284 Speaker 3: times where I think it's thinking and being very smart, 422 00:23:22,284 --> 00:23:24,484 Speaker 3: in times when it can't fucking add numbers together, right, 423 00:23:24,724 --> 00:23:28,644 Speaker 3: So and so like, in some ways, the Groc thing 424 00:23:29,844 --> 00:23:31,844 Speaker 3: was scarier than like the Google thing, because like Elon 425 00:23:31,924 --> 00:23:36,684 Speaker 3: had like very restively gentle changes for the architecture produced 426 00:23:36,684 --> 00:23:39,684 Speaker 3: these profound changes where it was literally calling itself Mecha 427 00:23:39,724 --> 00:23:40,724 Speaker 3: Hitler in outputs. 428 00:23:40,804 --> 00:23:40,964 Speaker 1: Right. 429 00:23:41,884 --> 00:23:45,004 Speaker 3: At the same time that gets to the race dynamic, right, 430 00:23:45,044 --> 00:23:47,284 Speaker 3: that model had not been because you know what happens. 431 00:23:48,244 --> 00:23:50,284 Speaker 3: You program an AI, right, you train it, you give 432 00:23:50,324 --> 00:23:53,044 Speaker 3: it a system prompt, and then you test test test it, right, 433 00:23:54,204 --> 00:23:57,644 Speaker 3: and whenever it says Hitler, you say no, you're being 434 00:23:57,644 --> 00:23:58,204 Speaker 3: a bad boy. 435 00:23:58,284 --> 00:23:59,404 Speaker 2: If this is not in the context of. 436 00:23:59,444 --> 00:24:02,604 Speaker 3: World War two or something, then please don't mention Hitler 437 00:24:02,684 --> 00:24:05,284 Speaker 3: so much. Right, And you know, if you spank it 438 00:24:05,324 --> 00:24:07,844 Speaker 3: really hard. It learns to avoid that, right, But if 439 00:24:07,884 --> 00:24:10,244 Speaker 3: you're if you're just kind of I'm getting the prompt, 440 00:24:10,804 --> 00:24:12,924 Speaker 3: testing it with two of your buddies and then releasing it, 441 00:24:12,964 --> 00:24:15,284 Speaker 3: you might you might miss those. So there is fundamentally 442 00:24:15,324 --> 00:24:19,284 Speaker 3: a speed off or a trade off, speed off, a 443 00:24:19,324 --> 00:24:24,124 Speaker 3: trade off between speed and safety. Right, it's maybe not 444 00:24:24,164 --> 00:24:26,644 Speaker 3: a one for one trade off that it's probably pretty 445 00:24:26,884 --> 00:24:27,964 Speaker 3: inversely correlated. 446 00:24:28,364 --> 00:24:29,844 Speaker 1: It is it is, And by the way, in the 447 00:24:29,844 --> 00:24:34,444 Speaker 1: Grock example, I think it's also interesting that some of 448 00:24:34,484 --> 00:24:37,524 Speaker 1: the coding that was revealed said also to check what 449 00:24:37,524 --> 00:24:40,884 Speaker 1: Elon Musk thinks about this right before giving your answer. 450 00:24:40,964 --> 00:24:44,324 Speaker 1: And this was not this was not a blip, Like 451 00:24:44,404 --> 00:24:48,604 Speaker 1: multiple people replicated that this was actually in the coding. 452 00:24:49,004 --> 00:24:52,364 Speaker 1: And as you've already said, like open AI said before 453 00:24:52,404 --> 00:24:55,364 Speaker 1: that it was you know, safety first, and that's clearly 454 00:24:55,404 --> 00:24:57,604 Speaker 1: not the case, and they've shown that to not be 455 00:24:57,644 --> 00:24:59,724 Speaker 1: the case. So I think we do need to keep 456 00:24:59,764 --> 00:25:05,164 Speaker 1: asking questions about incentives, about alignment, and about kind of 457 00:25:05,444 --> 00:25:10,764 Speaker 1: who's benefiting from what type of advances in the AI community. 458 00:25:11,084 --> 00:25:12,684 Speaker 3: Yeah, I mean the as kid, I mean, it felt 459 00:25:12,684 --> 00:25:14,244 Speaker 3: to me like Grock kind of almost took on like 460 00:25:14,284 --> 00:25:16,724 Speaker 3: a personality or became like kind of a self caricature, 461 00:25:16,804 --> 00:25:20,004 Speaker 3: right where it can be recursive, right if it's like 462 00:25:20,524 --> 00:25:24,884 Speaker 3: searching output on X, including its own output. Right, I 463 00:25:24,964 --> 00:25:27,924 Speaker 3: might say, Okay, this is how Groc behaves, and like, 464 00:25:29,124 --> 00:25:31,004 Speaker 3: let me be careful about this. I mean, Groc can 465 00:25:31,124 --> 00:25:35,124 Speaker 3: kind of be in a dark way, kind of funny, 466 00:25:36,084 --> 00:25:40,604 Speaker 3: whereas the other A models are are are not right, 467 00:25:40,644 --> 00:25:43,884 Speaker 3: they're kind of goody two shoes aiming to please. Oh, 468 00:25:43,964 --> 00:25:46,204 Speaker 3: that's the most brilliant article I've ever seen, Nate. You know, 469 00:25:46,244 --> 00:25:48,244 Speaker 3: here are twenty seven typos if I give it a 470 00:25:48,244 --> 00:25:52,804 Speaker 3: copyated task, right, And this is a counter to that. 471 00:25:52,844 --> 00:25:54,444 Speaker 3: But like, all these things are tied together, right, Like 472 00:25:55,804 --> 00:26:00,164 Speaker 3: wokeness is also tied to kind of a certain type 473 00:26:00,204 --> 00:26:04,404 Speaker 3: of academic expertise, you know what I mean, which I 474 00:26:04,444 --> 00:26:06,724 Speaker 3: call the indigo blob. How Like, Yeah, the liberal media 475 00:26:06,804 --> 00:26:10,764 Speaker 3: is biased and liberal, but also it's generally more accurate 476 00:26:10,804 --> 00:26:13,124 Speaker 3: and has more expertise in the conservative media. Both those 477 00:26:13,164 --> 00:26:15,724 Speaker 3: things are true. If you crunch a bunch of data 478 00:26:16,484 --> 00:26:20,404 Speaker 3: from the Internet into vectors, then that gets hard, right, 479 00:26:20,484 --> 00:26:22,244 Speaker 3: It gets hard because you might throw the baby out 480 00:26:22,284 --> 00:26:23,924 Speaker 3: with the bath water, right that you want to like 481 00:26:24,724 --> 00:26:28,924 Speaker 3: have it reflect the expert consensus. Well, the experts have 482 00:26:29,124 --> 00:26:32,604 Speaker 3: biased it sometimes sometimes creeps into that consensus, right, And 483 00:26:32,684 --> 00:26:35,284 Speaker 3: it's very hard for again for newsrooms and journals and 484 00:26:35,284 --> 00:26:37,484 Speaker 3: constitutions to do it, but just as hard for AI models. 485 00:26:38,124 --> 00:26:42,564 Speaker 1: Yeah, And as AI models become more integrated and become 486 00:26:42,644 --> 00:26:44,764 Speaker 1: kind of more a part of the way that people 487 00:26:44,804 --> 00:26:48,404 Speaker 1: do work, and we see more outputs from AI models 488 00:26:48,444 --> 00:26:51,684 Speaker 1: and their training materials shift, like these are things that 489 00:26:51,724 --> 00:26:54,484 Speaker 1: can become even worse kind of in the future. It 490 00:26:54,564 --> 00:26:57,644 Speaker 1: is kind of this cycle that can self reinforce. 491 00:27:01,684 --> 00:27:03,604 Speaker 2: And we'll be right back after this break. 492 00:27:16,844 --> 00:27:19,164 Speaker 3: I don't want people thinking that like an AI is 493 00:27:19,244 --> 00:27:23,164 Speaker 3: like an oracular. It's not perfect entity either, right. 494 00:27:23,244 --> 00:27:25,524 Speaker 2: You know. Paul Krugman had a thing about like. 495 00:27:26,924 --> 00:27:28,884 Speaker 3: GROC a couple of weeks ago where he's like, well, 496 00:27:28,924 --> 00:27:31,204 Speaker 3: if AI has a liberal bias, because reality has a 497 00:27:31,244 --> 00:27:33,484 Speaker 3: liberal bias, right, And I'm like, well, maybe on some things, 498 00:27:33,564 --> 00:27:36,444 Speaker 3: but like you know, and he's actually pretty skeptical of AI, 499 00:27:36,524 --> 00:27:40,684 Speaker 3: Like that's not you know, look at any algorithm, and 500 00:27:41,724 --> 00:27:43,804 Speaker 3: I probably an AIL. I'm probably too complicated to describe 501 00:27:43,804 --> 00:27:46,284 Speaker 3: as an algorithm. It has like some membership behaviors, but 502 00:27:46,364 --> 00:27:52,524 Speaker 3: still directionally speaking, right, any algorithm is imperfect, you know 503 00:27:52,524 --> 00:27:56,764 Speaker 3: what I mean, to make bad predictions can reflect bad data. 504 00:27:57,284 --> 00:28:01,644 Speaker 3: And the AI is especially they're not fine tuned, which 505 00:28:01,644 --> 00:28:02,484 Speaker 3: requires human input. 506 00:28:02,564 --> 00:28:04,084 Speaker 2: Right, that's how they're fine tune. They have like a 507 00:28:04,084 --> 00:28:05,804 Speaker 2: bunch of people being like yes no, yes, no. 508 00:28:06,804 --> 00:28:09,244 Speaker 3: Then I mean I think garbage in, garbage out is 509 00:28:09,284 --> 00:28:10,484 Speaker 3: not quite the right case. 510 00:28:10,564 --> 00:28:15,524 Speaker 1: But yeah, directionally speaking, that actually that points to what 511 00:28:15,604 --> 00:28:19,124 Speaker 1: we're talking about, which is that the inputs really really 512 00:28:19,164 --> 00:28:21,844 Speaker 1: do matter. And I think the other thing is that 513 00:28:21,884 --> 00:28:26,204 Speaker 1: there is this human bias to think that, oh, well, 514 00:28:26,244 --> 00:28:28,804 Speaker 1: this is data. This is a machine, so it's more 515 00:28:28,924 --> 00:28:32,004 Speaker 1: unbiased because it's not human. And that's simply not true, 516 00:28:33,284 --> 00:28:36,164 Speaker 1: you know. And if we have that bias to say, oh, well, 517 00:28:36,164 --> 00:28:39,044 Speaker 1: this is like this is objective truth because it's coming 518 00:28:39,084 --> 00:28:43,564 Speaker 1: from a not like it's coming from a program that's 519 00:28:43,644 --> 00:28:46,044 Speaker 1: that ain't work, that ain't where it's at like that, 520 00:28:46,044 --> 00:28:48,684 Speaker 1: that is actually a highly problematic way of thinking. 521 00:28:48,764 --> 00:28:51,564 Speaker 3: Yeah, and look what it's objectives aren't clear, right, I mean, 522 00:28:51,604 --> 00:28:55,364 Speaker 3: in one sense, large language models are trying to minimize 523 00:28:55,364 --> 00:28:58,164 Speaker 3: the laws function from accurately affecting what's represent the data. 524 00:28:58,364 --> 00:29:02,804 Speaker 3: On the other hand, it's contradicted by the human feedback 525 00:29:02,844 --> 00:29:04,764 Speaker 3: that it gets, right, and so it's kind of trying 526 00:29:04,844 --> 00:29:09,524 Speaker 3: to please its creators to whatever extent it can, and 527 00:29:09,604 --> 00:29:11,644 Speaker 3: to the point of being kind of a sycophant in 528 00:29:11,684 --> 00:29:15,524 Speaker 3: some cases. So like and you know, I think people 529 00:29:15,644 --> 00:29:18,884 Speaker 3: understand more about interpreting AI outputs than they did a 530 00:29:18,964 --> 00:29:22,284 Speaker 3: year or two or three ago. But it can be fragile. 531 00:29:22,364 --> 00:29:29,044 Speaker 3: You know, small things can affect the entire system. Right, 532 00:29:29,084 --> 00:29:32,044 Speaker 3: It's it's complexity. It's a kind of origin of complexity. 533 00:29:32,084 --> 00:29:35,204 Speaker 3: Theory really is well reflected in AI. Where a butterfly 534 00:29:35,244 --> 00:29:38,484 Speaker 3: flaps its wings in Beijing, this tornado in Texas. Right, 535 00:29:38,804 --> 00:29:42,844 Speaker 3: you know, Elon puts one change the system prompt in place. 536 00:29:42,924 --> 00:29:43,844 Speaker 2: It doesn't seem to matter. 537 00:29:44,124 --> 00:29:46,804 Speaker 3: But when you put that change in and then you say, also, 538 00:29:47,324 --> 00:29:50,244 Speaker 3: please make sure you're reading what Elon would write. Right, 539 00:29:50,764 --> 00:29:53,564 Speaker 3: those things combined can have a profound effect. 540 00:29:53,884 --> 00:29:56,764 Speaker 1: And this is just something that's visible. Now, imagine all 541 00:29:56,804 --> 00:29:59,444 Speaker 1: the things that are invisible right behind the scenes, like 542 00:29:59,524 --> 00:30:02,924 Speaker 1: tiny tweaks like that that we don't see because it's 543 00:30:03,004 --> 00:30:07,924 Speaker 1: not actually visible in like your immediate LLLM output unless 544 00:30:07,964 --> 00:30:10,044 Speaker 1: you really try to kind of no, and we. 545 00:30:10,084 --> 00:30:11,884 Speaker 2: Moved away from transparency in general. 546 00:30:11,964 --> 00:30:14,204 Speaker 3: I mean not just in Google's Ai Gemini, but like 547 00:30:14,244 --> 00:30:17,964 Speaker 3: in Google Search, there's a lot of intervention, right that 548 00:30:18,004 --> 00:30:21,244 Speaker 3: if you search for occupations, just in regular Google search 549 00:30:21,284 --> 00:30:25,844 Speaker 3: for photos of like doctors, it will be very conscientious 550 00:30:25,884 --> 00:30:28,684 Speaker 3: to show I mean there might be more women doctors now. 551 00:30:28,684 --> 00:30:32,244 Speaker 3: It's like it will show like doctors of all races 552 00:30:32,244 --> 00:30:34,284 Speaker 3: and women doctors, right if you probably if you search 553 00:30:34,284 --> 00:30:35,924 Speaker 3: for hockey player, it'll make sure. I mean there are 554 00:30:36,004 --> 00:30:38,324 Speaker 3: some great black hockey players, right, but it's a lot 555 00:30:38,324 --> 00:30:44,444 Speaker 3: of white Canadian and European kids, right, And so you know, yeah, 556 00:30:44,484 --> 00:30:46,364 Speaker 3: the notion of oh, the Web's just a search, right 557 00:30:46,564 --> 00:30:48,124 Speaker 3: or large language models are just kind of a new 558 00:30:48,164 --> 00:30:52,084 Speaker 3: to presentation of a clever way to like matricize Internet 559 00:30:52,204 --> 00:30:53,244 Speaker 3: data and writing. 560 00:30:53,324 --> 00:30:54,564 Speaker 2: I mean, it's never been that. 561 00:30:54,884 --> 00:31:00,404 Speaker 3: And the consensus has moved away from transparency if anything 562 00:31:00,484 --> 00:31:04,724 Speaker 3: almost toward like, well, if we're transparent transparent, they'll just 563 00:31:04,804 --> 00:31:06,604 Speaker 3: give you more things to critique or to pick on. 564 00:31:06,764 --> 00:31:10,164 Speaker 3: And so so sorry, you know, if you like the model, 565 00:31:10,204 --> 00:31:12,164 Speaker 3: then use it. If you don't like it, then use 566 00:31:12,204 --> 00:31:13,044 Speaker 3: different model instead. 567 00:31:13,444 --> 00:31:15,324 Speaker 1: Yeah, and there is this tension here once again with 568 00:31:15,644 --> 00:31:17,924 Speaker 1: going back to what we were talking about with open source, right, Like, 569 00:31:17,964 --> 00:31:21,644 Speaker 1: there's definitely this tension between open source, and that's actually 570 00:31:21,684 --> 00:31:24,004 Speaker 1: some of the good stuff of transparency, right that you 571 00:31:24,044 --> 00:31:26,124 Speaker 1: can actually look at the inputs and actually try to 572 00:31:26,124 --> 00:31:28,884 Speaker 1: figure out what's going on, are they Like, that's one 573 00:31:28,884 --> 00:31:32,404 Speaker 1: of the benefits of having open source, transparent models that 574 00:31:32,444 --> 00:31:34,924 Speaker 1: people can kind of pick apart and play with. We've 575 00:31:34,924 --> 00:31:36,724 Speaker 1: talked about kind of the risks, you know, p doom, 576 00:31:36,724 --> 00:31:40,284 Speaker 1: et cetera, et cetera, But there are benefits, and transparency 577 00:31:40,324 --> 00:31:41,564 Speaker 1: is certainly one of them. 578 00:31:41,924 --> 00:31:44,684 Speaker 3: Yeah, they been many times when we've had models different 579 00:31:44,724 --> 00:31:48,044 Speaker 3: kinds running, right, and people are like, this seems wrong, right, 580 00:31:48,164 --> 00:31:50,844 Speaker 3: why is this pull away a certain way? And like, yeah, 581 00:31:51,204 --> 00:31:54,244 Speaker 3: eighty percent of time it's part ofsan bs, But twenty 582 00:31:54,244 --> 00:31:56,764 Speaker 3: percent of time that they've caught something for sure, right, 583 00:31:56,804 --> 00:32:00,844 Speaker 3: any for sure? Even for myself, if I have a 584 00:32:00,924 --> 00:32:04,844 Speaker 3: model making sure I output different data at different stages, 585 00:32:05,164 --> 00:32:09,204 Speaker 3: evaluate it. If you don't catch a problem early on 586 00:32:09,244 --> 00:32:11,244 Speaker 3: when you're building a complex, you know, the NFL thing 587 00:32:11,284 --> 00:32:13,484 Speaker 3: I'm working on, right, there are lots of component parts. 588 00:32:13,484 --> 00:32:14,484 Speaker 2: It will eventually be a couple of. 589 00:32:14,484 --> 00:32:17,844 Speaker 3: Thousand lines of code and then which individually are that 590 00:32:17,844 --> 00:32:21,484 Speaker 3: complicated if like, if there's one step that's wrong, it 591 00:32:21,484 --> 00:32:25,084 Speaker 3: can infect the entire system, right, And if you're smart, 592 00:32:25,084 --> 00:32:27,644 Speaker 3: you're building and robust us and redundancy. You know, every 593 00:32:27,644 --> 00:32:30,124 Speaker 3: major computer program has bugs, every model has bugs. 594 00:32:30,204 --> 00:32:30,364 Speaker 2: Right. 595 00:32:30,404 --> 00:32:32,884 Speaker 3: But like, but like outputting that and being like I 596 00:32:32,884 --> 00:32:34,644 Speaker 3: want to visualize this, I want to look at the 597 00:32:34,764 --> 00:32:37,124 Speaker 3: right inputs, right, I want to look at edge cases 598 00:32:37,124 --> 00:32:37,804 Speaker 3: that are important. 599 00:32:38,044 --> 00:32:39,204 Speaker 2: Like that's an important thing to do. 600 00:32:39,844 --> 00:32:43,404 Speaker 1: Yeah, it absolutely is. And you know, to kind of 601 00:32:43,444 --> 00:32:47,844 Speaker 1: defend open source. That's one of the big things about 602 00:32:48,324 --> 00:32:53,484 Speaker 1: data integrity. Just everywhere is open sources good? Right, So 603 00:32:53,524 --> 00:32:57,244 Speaker 1: there was huge crisis in academia with you know, with cheating, 604 00:32:57,404 --> 00:33:01,244 Speaker 1: with fabricating data, you know, with papers that were published 605 00:33:01,244 --> 00:33:04,964 Speaker 1: in important journals with big results that ended up being manipulated. 606 00:33:05,244 --> 00:33:07,844 Speaker 1: And so there's more and more of a movement toward 607 00:33:08,284 --> 00:33:10,204 Speaker 1: you know, not just pre regis during your study, but 608 00:33:10,204 --> 00:33:12,764 Speaker 1: sharing your data, right, like actually sharing the data sets 609 00:33:12,764 --> 00:33:15,524 Speaker 1: so that people can look at what the source material was. 610 00:33:16,004 --> 00:33:19,084 Speaker 1: And you know, like one of the biggest scandals that 611 00:33:19,124 --> 00:33:23,364 Speaker 1: we talked about briefly on the show with Dan Ariellie 612 00:33:23,764 --> 00:33:29,484 Speaker 1: and Francesca Gino, they you know, showed that some of 613 00:33:29,524 --> 00:33:33,444 Speaker 1: those data sets were manipulated, but that was only after 614 00:33:33,524 --> 00:33:36,604 Speaker 1: they were forced to provide the data sets, you know, 615 00:33:36,724 --> 00:33:40,564 Speaker 1: after some good some ego eide researchers were like, hey, 616 00:33:40,724 --> 00:33:43,804 Speaker 1: like these what we're seeing in the paper doesn't make sense, 617 00:33:44,164 --> 00:33:48,324 Speaker 1: and sometimes it's willful manipulation and sometimes it's just a mistake. Right. 618 00:33:48,364 --> 00:33:50,204 Speaker 1: There have been some side like there was one very 619 00:33:50,204 --> 00:33:54,004 Speaker 1: famous scientific study where they fucked up an Excel sheet 620 00:33:54,084 --> 00:33:57,324 Speaker 1: where they buy mistake like moved the rows down by one, 621 00:33:57,804 --> 00:34:00,444 Speaker 1: and that just completely. It was just it was horrific 622 00:34:00,484 --> 00:34:04,324 Speaker 1: because this is a medical study. Right, So actually tiny 623 00:34:04,364 --> 00:34:08,084 Speaker 1: things like that, it's incredibly important to have access to 624 00:34:08,124 --> 00:34:10,524 Speaker 1: the raw data, to be able to to look at this, 625 00:34:10,684 --> 00:34:14,964 Speaker 1: to manipulate it on your own. People can really spot 626 00:34:15,004 --> 00:34:18,804 Speaker 1: both malfeasance and just human error which will happen and 627 00:34:18,924 --> 00:34:21,924 Speaker 1: with AI will have both human error and computer error. 628 00:34:22,164 --> 00:34:25,684 Speaker 1: Right as we're as we're having AIS do coding for us, 629 00:34:25,724 --> 00:34:28,084 Speaker 1: you know, vibe coding all of that, there might be 630 00:34:28,244 --> 00:34:31,604 Speaker 1: errors there. We need people. We need people capable of 631 00:34:31,604 --> 00:34:34,004 Speaker 1: looking at it, debugging and trying to figure out, hey, 632 00:34:34,244 --> 00:34:36,804 Speaker 1: this seems like it's working great, but actually it's not. 633 00:34:37,684 --> 00:34:40,724 Speaker 3: Now a well designed model, including in a large language 634 00:34:40,724 --> 00:34:43,724 Speaker 3: model is kind of like a airplane, right, there are 635 00:34:43,764 --> 00:34:48,644 Speaker 3: like multiple redundant systems, there are different levels of safety 636 00:34:48,684 --> 00:34:50,564 Speaker 3: and safeguards, and like that's hard to do it. That 637 00:34:50,604 --> 00:34:53,244 Speaker 3: requires a lot of work and expensive budget and being 638 00:34:53,284 --> 00:34:55,084 Speaker 3: an actually good programmer YEP. 639 00:34:55,484 --> 00:34:59,084 Speaker 1: One of the other major things that I saw in 640 00:34:59,124 --> 00:35:02,924 Speaker 1: this action plan that I think is actually quite important, 641 00:35:03,004 --> 00:35:06,084 Speaker 1: and again there are arguments on both sides, is like 642 00:35:06,284 --> 00:35:10,164 Speaker 1: one of the big bottlenecks to AI development is energy, right, 643 00:35:10,204 --> 00:35:14,324 Speaker 1: because AI just like takes a lot a huge energy 644 00:35:14,364 --> 00:35:17,804 Speaker 1: cost and access to data centers. Like that's that's actually 645 00:35:17,804 --> 00:35:21,244 Speaker 1: been a big thing where data centers like can't you know, 646 00:35:21,484 --> 00:35:25,004 Speaker 1: they've tried to bypass the power grids. There's been you know, pushback, 647 00:35:25,044 --> 00:35:27,044 Speaker 1: et cetera, et cetera. So one of the things that 648 00:35:27,124 --> 00:35:30,284 Speaker 1: this plan tries to do is say, let's eliminate all 649 00:35:30,324 --> 00:35:33,244 Speaker 1: the barriers and let's try to use all the energy 650 00:35:33,244 --> 00:35:35,644 Speaker 1: we have available for AI. Yeah. 651 00:35:35,844 --> 00:35:38,364 Speaker 3: I mean, look, I think it's just like a little overclaim, right. 652 00:35:38,404 --> 00:35:42,444 Speaker 3: If you look at like the amount of actual wattage 653 00:35:42,444 --> 00:35:45,364 Speaker 3: required to like run a chat reipete response, it's like 654 00:35:45,404 --> 00:35:48,164 Speaker 3: this is kind of like a I don't know, it's 655 00:35:48,164 --> 00:35:51,284 Speaker 3: a fairly bullshitty claim based on the present capabilities of AI. 656 00:35:52,444 --> 00:35:54,484 Speaker 1: So Nate, let me just push back a little bit 657 00:35:54,604 --> 00:35:57,284 Speaker 1: because there's a related issue, which is the availability of 658 00:35:57,324 --> 00:36:00,764 Speaker 1: water to cool these data centers. And we have already 659 00:36:00,804 --> 00:36:03,404 Speaker 1: seen that there has been you know, there have been 660 00:36:03,484 --> 00:36:07,684 Speaker 1: instances of wells town wells running dry. There's been instances 661 00:36:07,804 --> 00:36:12,204 Speaker 1: of construction having to be because the water, because of 662 00:36:12,204 --> 00:36:14,844 Speaker 1: basically water shortage issues. So I think that that's a 663 00:36:14,884 --> 00:36:17,684 Speaker 1: related concern that we should have just dispised. 664 00:36:17,564 --> 00:36:19,684 Speaker 2: Getting from low Ki Maria. 665 00:36:20,924 --> 00:36:23,604 Speaker 1: Maybe maybe Grock, What do you have to say about this? 666 00:36:23,844 --> 00:36:28,244 Speaker 3: Yeah, I would, I would, I would be I'm skeptical. 667 00:36:27,764 --> 00:36:29,124 Speaker 1: Of I think it is a big I think it 668 00:36:29,204 --> 00:36:29,804 Speaker 1: is a big deal. 669 00:36:30,124 --> 00:36:33,684 Speaker 3: It's not relative to other uses of electricity though it's 670 00:36:33,684 --> 00:36:36,564 Speaker 3: a small piece of the pie. And granted Sam Altman 671 00:36:36,644 --> 00:36:38,084 Speaker 3: at all want to be bigger. 672 00:36:38,004 --> 00:36:39,124 Speaker 2: Right. 673 00:36:40,364 --> 00:36:42,564 Speaker 1: Yeah, And I you know, it's hard for me. I 674 00:36:42,604 --> 00:36:44,964 Speaker 1: don't have the numbers in front of me, so I'm 675 00:36:44,964 --> 00:36:47,644 Speaker 1: not going to say something and have it not be true. 676 00:36:47,684 --> 00:36:50,564 Speaker 1: So so I'm just I'm not gonna engage that point 677 00:36:50,604 --> 00:36:54,564 Speaker 1: because I just honestly don't know. However, it is something 678 00:36:55,084 --> 00:36:58,444 Speaker 1: that this bill is trying to address and trying to 679 00:36:58,444 --> 00:37:02,524 Speaker 1: give AI companies whatever resources it needs to access power grids, 680 00:37:02,564 --> 00:37:06,644 Speaker 1: access energy grids, so that data centers can be built 681 00:37:06,644 --> 00:37:10,964 Speaker 1: on federal land and kind of allow innovation to proceed 682 00:37:11,364 --> 00:37:14,124 Speaker 1: without there being an energy bottleneck. I think that that's 683 00:37:14,244 --> 00:37:17,924 Speaker 1: kind of objectively what this bill is trying to accomplish. 684 00:37:18,164 --> 00:37:21,804 Speaker 3: Yeah, look, I mean, by the standards of Trump, it's 685 00:37:21,804 --> 00:37:26,684 Speaker 3: a pretty serious document. The one good thing about their 686 00:37:26,724 --> 00:37:28,684 Speaker 3: alliance may not the only good thing right with the 687 00:37:28,764 --> 00:37:30,644 Speaker 3: kind of tech right is that they do I think 688 00:37:30,684 --> 00:37:34,484 Speaker 3: have like Capetan people in place. You might have different 689 00:37:34,484 --> 00:37:39,164 Speaker 3: position on AI, but like, this isn't an amateurish documents. 690 00:37:39,284 --> 00:37:42,644 Speaker 3: It's thoughtful. It involves politics, but like there was some 691 00:37:42,684 --> 00:37:45,724 Speaker 3: care put into this. Yeah, for sure, for sure, very 692 00:37:45,764 --> 00:37:47,324 Speaker 3: unlike Trump in most ways. 693 00:37:47,484 --> 00:37:51,084 Speaker 1: I agree. I agree reading it, you know, obviously you 694 00:37:51,084 --> 00:37:52,924 Speaker 1: can see the Trump and language and parts of it, 695 00:37:53,244 --> 00:37:56,644 Speaker 1: but there there are nuanced points here where it will 696 00:37:56,684 --> 00:37:59,644 Speaker 1: be interesting to see how, if at all, any of 697 00:37:59,684 --> 00:38:03,124 Speaker 1: this is enforced and kind of what ends up happening. 698 00:38:03,124 --> 00:38:05,564 Speaker 1: I think we'll just have to see over the next 699 00:38:05,684 --> 00:38:09,204 Speaker 1: six months, one year and see how a lot of 700 00:38:09,244 --> 00:38:11,244 Speaker 1: these provisions shakeout. By the way, there are a lot 701 00:38:11,284 --> 00:38:14,084 Speaker 1: more provisions. We're not going to talk about the whole thing. Nate, 702 00:38:14,124 --> 00:38:17,004 Speaker 1: as you said, V has a good substock on this, 703 00:38:17,164 --> 00:38:19,364 Speaker 1: so for people who want to know more, I think 704 00:38:19,484 --> 00:38:22,764 Speaker 1: we'd both encourage you to read that. But yeah, all 705 00:38:22,764 --> 00:38:26,844 Speaker 1: in all, you know some things where you're like, okay, yeah, 706 00:38:26,924 --> 00:38:29,364 Speaker 1: this makes out some things that are a little you know, 707 00:38:29,884 --> 00:38:36,964 Speaker 1: eyebrow raising, not enough care shown to or not care 708 00:38:37,044 --> 00:38:40,284 Speaker 1: that's the wrong word. But I think we're seeing speed 709 00:38:40,804 --> 00:38:44,084 Speaker 1: over safety in a lot of this, and how do 710 00:38:44,124 --> 00:38:46,684 Speaker 1: we facilitate speed instead of safety? So I think that that, 711 00:38:47,044 --> 00:38:48,724 Speaker 1: to me is the big thing to kind of watch 712 00:38:48,724 --> 00:38:56,324 Speaker 1: out for and to keep an eye on. Thanks for listening. 713 00:38:56,564 --> 00:38:58,804 Speaker 1: We're taking next week off and we'll be back in 714 00:38:58,844 --> 00:39:02,684 Speaker 1: your feeds on August fourteenth. As always, we have some 715 00:39:02,724 --> 00:39:06,444 Speaker 1: additional content for premium subscribers who also get ad free 716 00:39:06,484 --> 00:39:10,484 Speaker 1: listening across Pushkin's entire network of shows. This week, we're 717 00:39:10,524 --> 00:39:14,804 Speaker 1: answering a listener question about how to teach poker to kids. 718 00:39:15,084 --> 00:39:17,684 Speaker 1: That's coming up after the credits, So you still have time. 719 00:39:17,964 --> 00:39:21,804 Speaker 1: Subscribe now for just six ninety nine a month. Let 720 00:39:21,844 --> 00:39:23,804 Speaker 1: us know what you think. Of the show reach out 721 00:39:23,844 --> 00:39:29,404 Speaker 1: to Us at Risky Business at Pushkin dot Fm. Risky 722 00:39:29,444 --> 00:39:32,164 Speaker 1: Business is hosted by me Maria Kanikova. 723 00:39:31,724 --> 00:39:34,084 Speaker 3: And by me Nate Silver. The show is a co 724 00:39:34,164 --> 00:39:38,324 Speaker 3: production of Pushkin Industries and iHeartMedia. This episode was produced 725 00:39:38,324 --> 00:39:42,084 Speaker 3: by Isabel Carter. Our associate producer is Sonia Gerwin. Sally 726 00:39:42,124 --> 00:39:45,164 Speaker 3: Helm is our editor, and our executive producer is Jacob Goldstein. 727 00:39:45,684 --> 00:39:46,844 Speaker 2: Mixing by Sarah Bruger. 728 00:39:47,084 --> 00:39:48,244 Speaker 1: Thanks so much for tuning in.