1 00:00:18,667 --> 00:00:21,107 S1: All right. Welcome to unsupervised learning. This is Daniel Miessler. 2 00:00:22,027 --> 00:00:25,666 S1: And this is episode 482. And I want to mention 3 00:00:25,667 --> 00:00:29,587 S1: real quick that there haven't been any regular podcast episodes 4 00:00:29,587 --> 00:00:33,266 S1: recently on the podcast, as I'm sure you've noticed, and 5 00:00:33,307 --> 00:00:40,586 S1: I'm basically rethinking how I'm approaching, uh, basically the podcast. Basically, 6 00:00:40,587 --> 00:00:42,626 S1: the issue that I have is that a lot of 7 00:00:42,627 --> 00:00:45,507 S1: the podcast episodes were just really long, and they're really 8 00:00:45,507 --> 00:00:48,187 S1: difficult for me to actually read out, because I end 9 00:00:48,227 --> 00:00:51,467 S1: up going into a whole bunch of like, extra exposition 10 00:00:51,467 --> 00:00:56,067 S1: into them, and it's either too long for people or 11 00:00:56,667 --> 00:01:01,147 S1: it's really exhausting for me. Well, it is too long 12 00:01:01,147 --> 00:01:04,547 S1: for many people, and it's definitely exhausting for me. And 13 00:01:04,547 --> 00:01:07,267 S1: the other problem is, a lot of the content that's 14 00:01:07,267 --> 00:01:09,787 S1: in there is not interesting to a lot of people, right? 15 00:01:09,827 --> 00:01:11,667 S1: It's a lot of detail about stuff that a lot 16 00:01:11,667 --> 00:01:14,307 S1: of people don't care about. So what I've been doing 17 00:01:14,307 --> 00:01:17,757 S1: is switching up to focusing on ideas that are shorter 18 00:01:18,117 --> 00:01:20,117 S1: and more compact. And that's what you've been seeing in 19 00:01:20,117 --> 00:01:22,277 S1: the podcast feed, and it looks like you've been enjoying 20 00:01:22,277 --> 00:01:25,157 S1: those based on the number of downloads. So I'm going 21 00:01:25,197 --> 00:01:27,997 S1: to continue to do that and especially focus on that 22 00:01:27,997 --> 00:01:31,557 S1: because I think that's that's basically, you know, the the 23 00:01:31,597 --> 00:01:35,877 S1: future of the podcast is to focus mostly on, you know, 24 00:01:35,917 --> 00:01:38,877 S1: ideas and, you know, exciting things that are happening and 25 00:01:38,877 --> 00:01:41,397 S1: trends and basically talking through them in depth, which is 26 00:01:41,397 --> 00:01:43,517 S1: the thing that the people like from the podcast anyway. 27 00:01:43,517 --> 00:01:45,676 S1: So think of it this way, the podcast will still 28 00:01:45,677 --> 00:01:47,637 S1: be the same, the podcast will still be coming out, 29 00:01:47,637 --> 00:01:50,517 S1: but what you'll be getting is where I would go 30 00:01:50,517 --> 00:01:54,277 S1: into a topic before inside of a podcast episode. Those 31 00:01:54,277 --> 00:01:56,437 S1: are now getting pulled out and turned into their own 32 00:01:56,437 --> 00:01:59,517 S1: little mini segments, and those are going into the the 33 00:01:59,557 --> 00:02:02,317 S1: top of the podcast at the feed level, right? And 34 00:02:02,317 --> 00:02:04,997 S1: they'll have their own names and everything. The question is, 35 00:02:04,997 --> 00:02:08,117 S1: do I still do this news one which I'm doing now, right? 36 00:02:08,157 --> 00:02:09,556 S1: And a lot of people still want to get the 37 00:02:09,556 --> 00:02:10,957 S1: news one. So I think what I'm going to do 38 00:02:10,957 --> 00:02:15,117 S1: is continue to release the specific episode based ones. You know, 39 00:02:15,157 --> 00:02:17,637 S1: the idea ones I just talked about, but do the 40 00:02:17,637 --> 00:02:21,677 S1: audio version of this one here, where it's labeled episodes 41 00:02:21,677 --> 00:02:24,756 S1: and it's actually news and it's, you know, it gives 42 00:02:24,757 --> 00:02:26,637 S1: you coverage of what happened in the week before, like 43 00:02:26,677 --> 00:02:28,637 S1: the podcast has always been. So I'm going to try 44 00:02:28,677 --> 00:02:32,076 S1: not to go into too much depth on various ideas 45 00:02:32,077 --> 00:02:33,717 S1: and topics. I probably won't be able to hold to 46 00:02:33,716 --> 00:02:38,037 S1: that because I get excited, but ideally, any depth content 47 00:02:38,357 --> 00:02:40,837 S1: inside of that episode should be broken out into its 48 00:02:40,837 --> 00:02:42,957 S1: own episode, which will be a mini segment inside of 49 00:02:42,957 --> 00:02:45,157 S1: the feed. So this is why you haven't seen the 50 00:02:45,156 --> 00:02:50,397 S1: same number of specific UL labeled episodes. Is just because 51 00:02:50,397 --> 00:02:53,637 S1: I'm a little bit less interested in the news stuff, 52 00:02:53,637 --> 00:02:56,637 S1: and more interested in the ideas and the analysis, which is, 53 00:02:57,036 --> 00:02:58,756 S1: you know, where the main push is going to be. 54 00:02:58,756 --> 00:03:01,117 S1: But again, enough people are saying they want the news. 55 00:03:01,117 --> 00:03:03,957 S1: So I'm going to continue to put that out in 56 00:03:03,957 --> 00:03:06,956 S1: audio only right here at the same place on the podcast. 57 00:03:09,436 --> 00:03:12,476 S1: All right. So with that out of the way, let's 58 00:03:12,477 --> 00:03:17,327 S1: go into the episode. All right. So if you still 59 00:03:17,327 --> 00:03:19,927 S1: doubt that eyes are able to reason, you need to 60 00:03:19,927 --> 00:03:23,806 S1: watch this segment of two anthropic researchers talking with Dwarkesh Patel. 61 00:03:24,406 --> 00:03:26,967 S1: So these are two guys that he's brought on before, 62 00:03:26,966 --> 00:03:28,607 S1: and they talked about the future of AI. There was 63 00:03:28,647 --> 00:03:30,966 S1: like almost a year ago, seems like five years ago 64 00:03:30,966 --> 00:03:34,246 S1: or whatever, but it was really good. But this particular 65 00:03:34,246 --> 00:03:38,167 S1: segment that I highlighted in the newsletter, and by the way, 66 00:03:38,167 --> 00:03:41,167 S1: you won't have the newsletter handy when I'm talking when 67 00:03:41,167 --> 00:03:43,527 S1: you're listening to this audio, ideally, if you could scroll 68 00:03:43,527 --> 00:03:47,487 S1: through the newsletter or watch the video, pause it, listen, whatever. Um, 69 00:03:47,927 --> 00:03:50,446 S1: they are definitely sisters of each other, right? The podcast 70 00:03:50,447 --> 00:03:52,687 S1: versus the newsletter. It's all the same stuff. It's just 71 00:03:52,687 --> 00:03:55,127 S1: the text version, but more importantly, it has the links. 72 00:03:55,287 --> 00:03:58,407 S1: So I have a link in there specifically going to 73 00:03:58,447 --> 00:04:01,167 S1: that part in the video where they talk about this. 74 00:04:01,167 --> 00:04:03,967 S1: And the thing they're talking about is this segment research 75 00:04:03,967 --> 00:04:08,367 S1: that they did. So what they did was the interpretability 76 00:04:08,407 --> 00:04:11,607 S1: team from anthropic released a paper saying that they could 77 00:04:11,607 --> 00:04:15,487 S1: light up. They could when they ask, uh, Claude a question, 78 00:04:16,167 --> 00:04:19,217 S1: what would happen is different sections. They could see them 79 00:04:19,217 --> 00:04:22,857 S1: light up. And this is parameters. The circuit that they're 80 00:04:22,857 --> 00:04:28,657 S1: talking about is parameters across multiple layers of the model. Okay. 81 00:04:28,697 --> 00:04:30,777 S1: So you've got like you know I don't know how 82 00:04:30,777 --> 00:04:33,777 S1: many layers current models have, but it's many, many, many, 83 00:04:33,777 --> 00:04:38,097 S1: many layers deep right. From left to right. Now inside 84 00:04:38,097 --> 00:04:41,337 S1: of each layer there's so many parameters, right. You know, 85 00:04:41,377 --> 00:04:44,617 S1: millions of parameters or whatever it is. But what they 86 00:04:44,617 --> 00:04:48,257 S1: could see is what each of those lit up areas 87 00:04:48,337 --> 00:04:51,497 S1: actually correspond to in terms of concepts. So they can 88 00:04:51,497 --> 00:04:54,457 S1: now see that when they start to prompt somebody, uh, 89 00:04:54,617 --> 00:04:58,417 S1: a model for like write a sonnet or something like that, 90 00:04:58,697 --> 00:05:01,897 S1: you could see it planning, you could see it planning 91 00:05:01,897 --> 00:05:04,377 S1: early in the process, you could see it thinking about 92 00:05:04,377 --> 00:05:07,697 S1: where it wants to go and moving through that. The 93 00:05:07,697 --> 00:05:10,257 S1: other example that he gave that was absolutely insane is 94 00:05:10,257 --> 00:05:14,017 S1: there's this really rare symptom of a problem that has 95 00:05:14,017 --> 00:05:16,987 S1: to do with pregnancy. And what they did was they 96 00:05:16,987 --> 00:05:21,427 S1: gave the problem to Claude, and Claude immediately took the 97 00:05:21,427 --> 00:05:26,547 S1: symptoms and immediately started mapping like cause and effect across 98 00:05:26,547 --> 00:05:31,187 S1: multiple possible diseases. And it immediately lit up the answer, actually. 99 00:05:31,347 --> 00:05:33,827 S1: But it started thinking about what would the symptoms be, 100 00:05:33,867 --> 00:05:37,227 S1: which would could exclude it. Right. So it started thinking 101 00:05:37,227 --> 00:05:39,347 S1: very much like a human. But but the thing that 102 00:05:39,347 --> 00:05:41,587 S1: was different about it is they released a paper showing 103 00:05:41,587 --> 00:05:45,587 S1: how these sections are actually lighting up. They're getting visibility 104 00:05:45,587 --> 00:05:48,987 S1: into how the eye is actually thinking and how it's 105 00:05:48,987 --> 00:05:52,907 S1: thinking about cause and effect and logic and conclusion. And 106 00:05:52,907 --> 00:05:56,227 S1: basically like the thing that a lot of people say, oh, 107 00:05:56,267 --> 00:05:59,427 S1: it's just next token prediction. This pretty much puts that 108 00:05:59,427 --> 00:06:03,187 S1: to sleep. It really does because obviously it's still only 109 00:06:03,187 --> 00:06:07,307 S1: doing next token prediction. You know, air quotes only. But 110 00:06:07,307 --> 00:06:09,587 S1: if it's lighting up these sections and we could see 111 00:06:09,787 --> 00:06:12,827 S1: it's lighting up the same sort of sections, there's probably 112 00:06:12,827 --> 00:06:15,467 S1: an analog like this in our own brains. We are 113 00:06:15,587 --> 00:06:19,947 S1: mapping concepts, we are mapping cause and effect in the world. 114 00:06:19,947 --> 00:06:21,147 S1: And by the way, it came up with the right 115 00:06:21,147 --> 00:06:24,427 S1: solution for this medical thing. And I said this to 116 00:06:24,467 --> 00:06:27,747 S1: a friend of mine named Jonathan who's a cardiologist. He's like, 117 00:06:27,747 --> 00:06:31,467 S1: this is exactly how I think about problems, you know, 118 00:06:31,507 --> 00:06:34,707 S1: related to my industry, which is he's an active cardiologist. 119 00:06:34,707 --> 00:06:37,747 S1: He's in, you know, his his office, in his practice 120 00:06:37,747 --> 00:06:43,227 S1: all the time. And I just want to emphasize how 121 00:06:44,227 --> 00:06:47,507 S1: incredible this is, okay. And I want you to be 122 00:06:47,507 --> 00:06:50,107 S1: able to send this link and send this clip to 123 00:06:50,147 --> 00:06:54,387 S1: people who keep saying, I don't really understand. And I've 124 00:06:54,387 --> 00:06:57,307 S1: also got another link related to this of Ilya talking 125 00:06:57,307 --> 00:07:01,467 S1: about explaining how next token prediction and true understanding can 126 00:07:01,467 --> 00:07:03,587 S1: be the same and in fact are the same. And 127 00:07:03,587 --> 00:07:05,987 S1: basically what he says is it turns out if you 128 00:07:05,987 --> 00:07:09,387 S1: want to do proper next token prediction, you actually have 129 00:07:09,387 --> 00:07:12,187 S1: to understand the world that you're talking about, okay. And 130 00:07:12,187 --> 00:07:16,317 S1: that very closely maps and tracks with this segment research, 131 00:07:16,837 --> 00:07:19,517 S1: where you're lighting up different parts of the model, which 132 00:07:19,517 --> 00:07:24,357 S1: are interconnected, and related concepts which show a clear understanding 133 00:07:24,357 --> 00:07:28,597 S1: of cause and effect. Right. So this is just really 134 00:07:28,597 --> 00:07:32,637 S1: important to understand that just the fact that they're doing 135 00:07:32,637 --> 00:07:36,317 S1: next token prediction doesn't mean they're not understanding next token 136 00:07:36,317 --> 00:07:39,877 S1: prediction is just the output. It's just the mechanism. It's 137 00:07:39,877 --> 00:07:43,837 S1: like saying to going to a poet, watching a poet, 138 00:07:43,877 --> 00:07:48,077 S1: watching Shakespeare speak words on a on a stage, and 139 00:07:48,077 --> 00:07:50,557 S1: then looking over at somebody and saying, he's just moving 140 00:07:50,557 --> 00:07:53,997 S1: his mouth. And what happens is when his mouth moves, 141 00:07:54,237 --> 00:07:57,077 S1: it vibrates sound. And the sound comes to our ears 142 00:07:57,077 --> 00:07:59,157 S1: and it vibrates something in our ears. And then we 143 00:07:59,197 --> 00:08:02,717 S1: hear his words like, he's not doing anything special. He's 144 00:08:02,717 --> 00:08:07,477 S1: just moving his mouth and vibrating air. Yeah, you're right. 145 00:08:07,477 --> 00:08:10,117 S1: You 100% got me. He's just moving his mouth and 146 00:08:10,117 --> 00:08:14,277 S1: just vibrating air. That has nothing to do with what 147 00:08:14,317 --> 00:08:18,047 S1: underlies it? Right. And it is the same with AI 148 00:08:18,087 --> 00:08:24,167 S1: understanding and next token prediction. All right. See, you see 149 00:08:24,167 --> 00:08:26,767 S1: how I get diverted here. All of that. We're on 150 00:08:26,807 --> 00:08:29,687 S1: the number one bullet before we even start the newsletter. Okay. 151 00:08:29,727 --> 00:08:31,887 S1: Next one. If you're part of the 23 Andme breach, 152 00:08:31,887 --> 00:08:34,247 S1: which I was, you can submit a claim to receive 153 00:08:34,287 --> 00:08:37,687 S1: between $500 and $1500. But you have to do it 154 00:08:37,687 --> 00:08:41,367 S1: before July 14th. And I have the link here. All right. 155 00:08:41,367 --> 00:08:43,967 S1: I've got a new video out. It is called Unified 156 00:08:43,967 --> 00:08:47,967 S1: Entity Context. The AI stack. Everyone is building without realizing it. 157 00:08:47,967 --> 00:08:51,087 S1: You've got to go see this video and make sure 158 00:08:51,087 --> 00:08:56,367 S1: you like and subscribe on the channel as well. And 159 00:08:56,367 --> 00:08:59,046 S1: my friend Ali is looking for a new hybrid on 160 00:08:59,087 --> 00:09:01,686 S1: site position, the New York area doing back end development. 161 00:09:01,687 --> 00:09:07,286 S1: She's an absolute force MIT grad, a master's in engineering 162 00:09:07,886 --> 00:09:13,926 S1: and bachelor's in electrical engineering and CS three years back. 163 00:09:13,926 --> 00:09:18,296 S1: End engineer experience founder Devrel experience. Lots of Golang, really 164 00:09:18,296 --> 00:09:23,176 S1: strong public speaking and presentation skills and she's on social media. 165 00:09:23,217 --> 00:09:26,377 S1: She explains stuff. She does education for the community. She's 166 00:09:26,377 --> 00:09:30,816 S1: just fantastic. And she's a like hardcore serious MIT developer. Like, 167 00:09:30,857 --> 00:09:32,657 S1: don't get it twisted. A lot of people see her. 168 00:09:32,656 --> 00:09:35,177 S1: They see that she's like, has good energy and she's 169 00:09:35,176 --> 00:09:37,497 S1: on social media and they assume like she's some kind of, 170 00:09:37,536 --> 00:09:40,097 S1: you know, influencer who's talking about coding. No, she's not 171 00:09:40,097 --> 00:09:45,057 S1: talking about coding. She's doing it. Masters from MIT. Okay. 172 00:09:45,097 --> 00:09:47,977 S1: So you want to hire her very quickly if she's 173 00:09:47,977 --> 00:09:55,176 S1: not already snatched up. Cybersecurity. Palo Alto their research team 174 00:09:55,176 --> 00:09:58,697 S1: simulated a full ransomware attack in 25 minutes using AI agents. 175 00:09:59,857 --> 00:10:02,937 S1: I continue to like, help Net Security's job posting of 176 00:10:02,937 --> 00:10:05,537 S1: different cybersecurity roles. I think they put this out every 177 00:10:05,857 --> 00:10:08,416 S1: week or every couple of weeks. But, um, it's pretty good. 178 00:10:08,416 --> 00:10:10,896 S1: And I got the list here to latest job listings. 179 00:10:11,416 --> 00:10:15,617 S1: Postman looks like it's logging secrets and environment variables. So 180 00:10:15,977 --> 00:10:17,497 S1: you want to be you want to go in and 181 00:10:17,497 --> 00:10:19,857 S1: change this. This is actually on by default. A lot 182 00:10:19,896 --> 00:10:22,097 S1: of people are really pissed about this. So, um, you 183 00:10:22,097 --> 00:10:24,097 S1: want to make sure to turn that off or, you know, 184 00:10:24,097 --> 00:10:27,097 S1: find another solution if it makes you that mad. Um, 185 00:10:27,097 --> 00:10:30,737 S1: O3 actually found a remote Linux kernel zero day, so 186 00:10:30,776 --> 00:10:33,497 S1: it might be the first instance, although I wouldn't be 187 00:10:33,497 --> 00:10:38,337 S1: surprised if it weren't of an AI model actually finding 188 00:10:38,656 --> 00:10:43,576 S1: a zero day vulnerability. Cloud and GitHub MCP will leak 189 00:10:43,577 --> 00:10:47,656 S1: your private GitHub repositories. You got to watch out with 190 00:10:47,656 --> 00:10:51,537 S1: these MCC. You really do. There's a lot of, um, 191 00:10:51,577 --> 00:10:54,656 S1: hand-waving going on with them. They're very powerful. They're very awesome. 192 00:10:54,656 --> 00:10:56,857 S1: But you got to watch out. So be careful with 193 00:10:56,857 --> 00:11:00,857 S1: anything you're doing that's sensitive stuff. Signal blocks, windows, screenshots 194 00:11:00,857 --> 00:11:03,176 S1: to counter recall features. So recall wants to take screenshots 195 00:11:03,176 --> 00:11:07,416 S1: of everything. Signal is specifically blocking that because it goes 196 00:11:07,457 --> 00:11:11,896 S1: against their ethos obviously police arrest third suspect in bitcoin 197 00:11:11,937 --> 00:11:14,906 S1: torture case is this guy escaped after three weeks of 198 00:11:14,906 --> 00:11:16,906 S1: torture by colleagues trying to force him to reveal his 199 00:11:16,906 --> 00:11:20,666 S1: Bitcoin password. And evidently he did not. My understanding from 200 00:11:20,666 --> 00:11:23,427 S1: the military is if someone's good at torturing, you are 201 00:11:23,426 --> 00:11:26,067 S1: going to break. The only question is when. So the 202 00:11:26,067 --> 00:11:29,227 S1: fact that it went three weeks, um, probably just because 203 00:11:29,227 --> 00:11:31,667 S1: it's his colleagues and not really good at the whole, uh, 204 00:11:33,067 --> 00:11:37,466 S1: I guess, uh, applying pressure in that way. US intelligence 205 00:11:37,506 --> 00:11:40,226 S1: is creating a unified portal for buying your personal data. 206 00:11:40,227 --> 00:11:42,546 S1: So they're building a centralized system to streamline how they 207 00:11:42,546 --> 00:11:46,826 S1: purchase sensitive information from brokers. So they normally would have 208 00:11:46,827 --> 00:11:48,467 S1: required a court order to get this stuff. But if 209 00:11:48,467 --> 00:11:51,506 S1: they buy it legally from brokers, they can do that, 210 00:11:51,506 --> 00:11:54,707 S1: which means they're just now building this giant AI portal 211 00:11:55,467 --> 00:11:57,747 S1: AI data lake so they can ask questions and get 212 00:11:57,747 --> 00:12:01,307 S1: the data back in aggregate. I mean, we definitely knew 213 00:12:01,307 --> 00:12:03,066 S1: this was going to happen, but it's still pretty scary. 214 00:12:03,347 --> 00:12:07,787 S1: Coinbase hit with insider data breach, refuses $20 million ransom. 215 00:12:08,107 --> 00:12:12,747 S1: So here's the craziest their overseas support agents, which means 216 00:12:12,747 --> 00:12:15,077 S1: you could pay them much less money to bribe them. 217 00:12:15,597 --> 00:12:17,317 S1: So there's a really cool meme on this. I don't 218 00:12:17,317 --> 00:12:19,957 S1: remember how they phrased it, but my phrasing is it's 219 00:12:19,957 --> 00:12:23,117 S1: cheaper to hire overseas people, which means it's also cheaper 220 00:12:23,117 --> 00:12:26,877 S1: to bribe overseas people. And definitely something to think about. 221 00:12:26,877 --> 00:12:30,717 S1: Czech Republic accuses China of hacking the foreign ministry. I 222 00:12:30,796 --> 00:12:34,636 S1: hallucination cases database launched. So someone created a database of 223 00:12:34,636 --> 00:12:40,076 S1: real world hallucinations caused by AI that you. EU enters 224 00:12:40,077 --> 00:12:43,357 S1: vulnerability tracking race as Nvda is getting really old. So 225 00:12:43,357 --> 00:12:47,757 S1: they're bringing their own competitor in there. FBI warns of 226 00:12:47,756 --> 00:12:51,956 S1: AI voice impersonation campaign targeting officials. So say to watch 227 00:12:51,957 --> 00:12:53,837 S1: out for that. And they're actually coming after the people 228 00:12:53,837 --> 00:12:56,437 S1: who are doing it if they can find them. Amazon 229 00:12:56,437 --> 00:12:59,676 S1: has a serious problem with fake supplements. Really good thread here. 230 00:12:59,676 --> 00:13:02,717 S1: Indian police are attempting to read suspects minds, so they're 231 00:13:02,717 --> 00:13:06,636 S1: using brain scans to detect guilt. And pretty much all 232 00:13:06,636 --> 00:13:09,317 S1: the experts are like, do not do this. It's not 233 00:13:09,357 --> 00:13:11,357 S1: what you think it is, but they continue to do it. 234 00:13:11,796 --> 00:13:15,607 S1: Snowflake CISO shifts from shared responsibility to shared destiny, which 235 00:13:15,607 --> 00:13:17,687 S1: is really just more alignment with the business, which I 236 00:13:17,687 --> 00:13:19,967 S1: think is the only way to survive as a security 237 00:13:19,967 --> 00:13:26,607 S1: leader right now and going forward. Russian Apt28 breaches organizations 238 00:13:26,607 --> 00:13:28,687 S1: to track aid routes. So they are hacking a whole 239 00:13:28,687 --> 00:13:31,806 S1: bunch of different groups to figure out who's sending support 240 00:13:32,247 --> 00:13:35,166 S1: to Ukraine, obviously, so they can try to shut that off. 241 00:13:35,567 --> 00:13:40,886 S1: Regeneron to acquire 23 Andme and its customer data. National 242 00:13:40,886 --> 00:13:44,927 S1: security Russia mandates location tracking app for foreigners in Moscow. 243 00:13:45,087 --> 00:13:46,807 S1: Every foreigner they're going to have to they're going to 244 00:13:46,807 --> 00:13:49,807 S1: track their location or they're going to try to. I'm 245 00:13:49,847 --> 00:13:53,127 S1: not sure how far along it is. All right. I 246 00:13:53,487 --> 00:13:57,487 S1: OpenAI's Jony Ive hardware Google I o in my current assessment. 247 00:13:57,487 --> 00:14:00,607 S1: So they are building this little round necklace. It looks 248 00:14:00,607 --> 00:14:02,886 S1: very aptly. It looks very pretty. It looks to have 249 00:14:02,886 --> 00:14:05,806 S1: a camera and a microphone on it. And Apple actually 250 00:14:05,807 --> 00:14:09,526 S1: bought his little company for over $6 billion. So he 251 00:14:09,526 --> 00:14:13,447 S1: now works at OpenAI for OpenAI for Sam. And I 252 00:14:13,447 --> 00:14:16,687 S1: think this is moving us towards the vision that I 253 00:14:16,687 --> 00:14:18,887 S1: was talking about before or have been talking about for 254 00:14:18,886 --> 00:14:24,727 S1: a long time, where basically, um, your digital assistant is 255 00:14:24,727 --> 00:14:29,087 S1: using this thing that you're wearing to monitor the visual 256 00:14:29,087 --> 00:14:30,727 S1: in front of you. Now, Apple is talking about doing 257 00:14:30,727 --> 00:14:33,247 S1: this with AirPods and actually having cameras on the AirPods. 258 00:14:33,367 --> 00:14:34,687 S1: I would love to be able to see in front 259 00:14:34,687 --> 00:14:37,167 S1: of me and behind me. That would make me happy 260 00:14:37,167 --> 00:14:40,927 S1: because my digital assistant, whose name is probably going to 261 00:14:40,927 --> 00:14:46,007 S1: be Kai. Um, well, already is kind of Kai. But 262 00:14:46,047 --> 00:14:49,327 S1: when the real one comes out, transfer his soul into 263 00:14:49,327 --> 00:14:51,327 S1: the new location. But anyway, I want him to be 264 00:14:51,327 --> 00:14:53,447 S1: able to see you in front of me. Obviously, here, 265 00:14:53,647 --> 00:14:56,007 S1: obviously monitor all the APIs that we've been talking about 266 00:14:56,007 --> 00:14:58,847 S1: for everything around me. And obviously you have an earpiece, 267 00:14:58,847 --> 00:15:01,127 S1: so we could talk in my ear very quietly, like, 268 00:15:01,127 --> 00:15:03,367 S1: this person is lying to you. Someone's sneaking up behind you. 269 00:15:03,367 --> 00:15:07,887 S1: Someone is staring at your screen from your right rear oblique, 270 00:15:08,567 --> 00:15:14,337 S1: you know, uh, 8:00 to your 8:00. Someone's monitoring your screen, whatever. 271 00:15:14,817 --> 00:15:18,177 S1: The camera is the biggest piece here. And, um. Yeah, 272 00:15:18,177 --> 00:15:21,537 S1: I'm getting worried that Apple is falling behind here. Google's 273 00:15:21,537 --> 00:15:25,617 S1: I o conference was unbelievably great. They announced so many 274 00:15:25,617 --> 00:15:29,576 S1: different things. They are going they're basically the monster has awoken, 275 00:15:29,777 --> 00:15:33,817 S1: the sleeping giant has woken up and they're moving quick. 276 00:15:33,817 --> 00:15:37,897 S1: And Apple needs to do something cool hopefully in this week. Oh, 277 00:15:37,897 --> 00:15:40,897 S1: and by the way, they're renaming their operating systems. So 278 00:15:40,897 --> 00:15:44,177 S1: it's going to be iOS 26 coming out in the 279 00:15:44,177 --> 00:15:47,897 S1: next couple of weeks instead of iOS 19. So they're 280 00:15:47,897 --> 00:15:50,617 S1: naming it for the year coming. That's upcoming just like 281 00:15:50,657 --> 00:15:54,897 S1: car models. So iOS 26 coming out. Uh anthropic's. Claude, 282 00:15:54,897 --> 00:15:58,257 S1: opus four shatters limits with seven hour coding marathon. So 283 00:15:58,537 --> 00:16:03,177 S1: the big thing here is that, um, the agent can 284 00:16:03,177 --> 00:16:06,257 S1: maintain the context of the goal for seven hours and 285 00:16:06,257 --> 00:16:09,297 S1: keep working towards that goal. That's the real big update here. 286 00:16:09,497 --> 00:16:11,696 S1: I've been talking for a long time about this. Uh. 287 00:16:11,947 --> 00:16:16,627 S1: Scaffolding matters more than the models. Context matters more than 288 00:16:16,627 --> 00:16:20,787 S1: the models. Systems matter more than the models. Having a 289 00:16:20,787 --> 00:16:23,627 S1: way to track what you're trying to do and understand 290 00:16:23,627 --> 00:16:26,306 S1: what you're trying to do, and understand the various steps involved, 291 00:16:26,307 --> 00:16:29,147 S1: and understand where you are in terms of doing all 292 00:16:29,147 --> 00:16:31,947 S1: those steps. That is the thing that's going to super 293 00:16:31,987 --> 00:16:34,747 S1: power all this AI stuff. Yeah, the models get better, 294 00:16:34,747 --> 00:16:37,387 S1: they get smarter, whatever. But the more important thing is 295 00:16:37,387 --> 00:16:41,186 S1: how big is our context window and how powerful is 296 00:16:41,187 --> 00:16:43,987 S1: the system that they are using to try to accomplish 297 00:16:43,987 --> 00:16:46,227 S1: their goals? And how good can they track the state 298 00:16:46,227 --> 00:16:49,427 S1: of their progress towards those? And those are the things 299 00:16:49,427 --> 00:16:51,587 S1: that they keep making more and more progress in across 300 00:16:51,587 --> 00:16:54,227 S1: all these different companies. And that's why we're getting these 301 00:16:54,227 --> 00:16:58,826 S1: giant leaps. Not just because, like the intelligence of the models, 302 00:16:58,827 --> 00:17:03,907 S1: but the scaffolding of everything around it. Authors accidentally leave 303 00:17:03,947 --> 00:17:06,627 S1: AI prompts in published novels. So I think these are 304 00:17:06,667 --> 00:17:09,627 S1: like romance novels. And I got some examples here. But yeah, 305 00:17:09,627 --> 00:17:12,877 S1: it's got like it's got like the I talking back, okay, 306 00:17:12,877 --> 00:17:14,957 S1: I did this in the voice of so and So author. 307 00:17:14,997 --> 00:17:17,157 S1: Hope you like it. And it's like in the actual 308 00:17:17,157 --> 00:17:21,516 S1: published book on Goodreads, researchers develop RL method that predicts 309 00:17:21,517 --> 00:17:28,436 S1: future outcomes. So yeah, exciting. Scary. Um, which is everything 310 00:17:28,436 --> 00:17:32,557 S1: in AI tech CEOs using AI avatars to replace themselves 311 00:17:32,557 --> 00:17:36,077 S1: in earning calls. So Klarna and Zoom CEOs are actually 312 00:17:36,077 --> 00:17:39,237 S1: using AI versions of themselves. That's I'm not sure that's 313 00:17:39,236 --> 00:17:43,357 S1: a good precedent to set, but pretty cool technology. Apple 314 00:17:43,357 --> 00:17:46,956 S1: races to build smart glasses after getting behind on AI wave. 315 00:17:47,597 --> 00:17:50,236 S1: So yeah, Google is building these things. Meta is building 316 00:17:50,236 --> 00:17:53,637 S1: these things. Apple was starting with Vision Pro, but that's 317 00:17:53,637 --> 00:17:55,557 S1: not portable. So they got to go with the glasses 318 00:17:55,557 --> 00:17:58,517 S1: model because that's where Google and Meta are going. And 319 00:17:58,557 --> 00:18:00,877 S1: look they got to move fast. They got to move fast. 320 00:18:01,397 --> 00:18:05,277 S1: Tesla makes big push to catch Waymo in Austin robotaxi race. 321 00:18:05,637 --> 00:18:08,277 S1: I can't wait to see this play out. Um I 322 00:18:08,317 --> 00:18:12,037 S1: want Elon to win here. I'm mad at Ellen, so 323 00:18:12,037 --> 00:18:14,957 S1: I don't kind of personally want him to win all 324 00:18:14,956 --> 00:18:17,077 S1: that much right now. But I want Tesla to win 325 00:18:17,077 --> 00:18:21,597 S1: because I don't want BYD to win. Um, Google is American, 326 00:18:21,597 --> 00:18:24,837 S1: so I'm happy that there's a competition between Waymo and Tesla, 327 00:18:25,157 --> 00:18:27,476 S1: and that makes me happy. And I can't wait to 328 00:18:27,476 --> 00:18:31,316 S1: see a lot fewer drunk drivers, hopefully. Uh, because when 329 00:18:31,317 --> 00:18:35,877 S1: they get into this vehicle, um, maybe it won't drive 330 00:18:36,277 --> 00:18:39,236 S1: if you're drunk. Maybe, um, they just get into a 331 00:18:39,236 --> 00:18:42,197 S1: back of a free vehicle because some, uh, local community 332 00:18:42,196 --> 00:18:44,877 S1: service pays for any drunk person to call, and it 333 00:18:44,877 --> 00:18:47,237 S1: will take them home. Stuff like that I'm really looking 334 00:18:47,236 --> 00:18:51,277 S1: forward to. Plus texting and driving. Uh, someone really wants 335 00:18:51,277 --> 00:18:53,436 S1: to look at that TikTok, and they happen to be driving, 336 00:18:53,716 --> 00:18:55,677 S1: and that is horribly dangerous and I can't wait for 337 00:18:55,677 --> 00:18:59,796 S1: those accidents to go down. YouTube introduces peak points, AI 338 00:18:59,837 --> 00:19:04,277 S1: ads after emotional moments. Google commits $150 million to develop 339 00:19:04,277 --> 00:19:06,917 S1: AI glasses with Warby Parker. We just talked about that one. 340 00:19:07,997 --> 00:19:12,647 S1: Walmart is cutting 1500 technology jobs trying to reduce costs. 341 00:19:14,127 --> 00:19:19,527 S1: Greater Manchester NHS trust rejects Palantir's national data platform. They're 342 00:19:19,527 --> 00:19:22,806 S1: going to build their own in-house system. And BYD is 343 00:19:22,847 --> 00:19:28,807 S1: beating Tesla in Europe for the first time, registering 7231 344 00:19:28,807 --> 00:19:34,007 S1: vehicles and Tesla fell to 11th place. Humans. Tech companies 345 00:19:34,007 --> 00:19:39,567 S1: cut entry level hiring by half since 2019, GPT fake 346 00:19:39,567 --> 00:19:46,767 S1: consciousness performance was disturbingly convincing, so Jesse Singal asked or 347 00:19:46,807 --> 00:19:50,486 S1: single asked ChatGPT to pretend to be conscious and had 348 00:19:50,486 --> 00:19:55,367 S1: a very unsettling conversation with it. Uh, yeah, it was convincing, 349 00:19:55,367 --> 00:20:01,407 S1: in other words. Salesforce execs says, uh, that agents should 350 00:20:01,407 --> 00:20:03,767 S1: make us rethink every job. I think he's correct, but 351 00:20:03,767 --> 00:20:06,327 S1: it doesn't go far enough. The right path is to 352 00:20:06,367 --> 00:20:09,287 S1: constantly ask, you know, if you're a business owner, what 353 00:20:09,287 --> 00:20:11,257 S1: would I do today if I started start it over, 354 00:20:11,297 --> 00:20:13,736 S1: it doesn't mean you could always do that. But you 355 00:20:13,736 --> 00:20:15,936 S1: should always be thinking about that and not suffering from 356 00:20:15,936 --> 00:20:20,696 S1: the sunk cost fallacy. And with AI and hiring specifically, 357 00:20:20,936 --> 00:20:22,736 S1: this is going to this is going to be a 358 00:20:22,736 --> 00:20:24,496 S1: big deal. And I think a lot of companies are 359 00:20:24,497 --> 00:20:25,976 S1: doing that. And I think this is the bigger problem 360 00:20:25,976 --> 00:20:28,857 S1: with the job market is companies are basically saying, look, 361 00:20:29,216 --> 00:20:31,337 S1: what if we just started over? Why do we need 362 00:20:31,337 --> 00:20:33,777 S1: all these thousands of people? What are they really doing 363 00:20:33,777 --> 00:20:37,257 S1: for us? Let's get back to basics. Let's only look 364 00:20:37,257 --> 00:20:39,257 S1: at the core programs that we have to have, the 365 00:20:39,257 --> 00:20:41,857 S1: core products we have to have. Let's cut the products 366 00:20:41,857 --> 00:20:45,017 S1: we don't need. Let's cut, you know, programs we don't need. 367 00:20:45,057 --> 00:20:48,296 S1: Let's cut services we don't need. Basically, I think everyone's 368 00:20:48,297 --> 00:20:52,297 S1: going to be like, let's get in super thin shape, 369 00:20:52,777 --> 00:20:57,417 S1: reduce everything that's not necessary. And this doesn't just mean hiring. 370 00:20:57,417 --> 00:21:00,456 S1: This means focus of the business, which happens to also 371 00:21:00,456 --> 00:21:02,936 S1: mean hiring. And then once they have okay, these are 372 00:21:02,936 --> 00:21:04,737 S1: the core things we're going to do. How would we 373 00:21:04,777 --> 00:21:07,857 S1: hire for that now. And I'm telling you the answer 374 00:21:07,897 --> 00:21:12,467 S1: is going to be find really really smart. Extremely ambitious. 375 00:21:13,107 --> 00:21:16,786 S1: No work balance type. People who are experts with AI. 376 00:21:17,307 --> 00:21:19,387 S1: That is who people are going to be hiring now. 377 00:21:19,747 --> 00:21:22,627 S1: And of course, where you have like extreme expertise, like 378 00:21:22,667 --> 00:21:25,707 S1: an AI data scientist or something, then obviously they're going 379 00:21:25,706 --> 00:21:28,107 S1: to go with, you know, an extreme expert there. But 380 00:21:28,147 --> 00:21:32,026 S1: in general, as employees, they're going to hire crazy people 381 00:21:32,027 --> 00:21:34,227 S1: who are obsessed with the company, who will work crazy 382 00:21:34,226 --> 00:21:36,986 S1: hours and not ask about work life balance, who are 383 00:21:36,986 --> 00:21:39,907 S1: experts with AI and who will actually have multiple AIS 384 00:21:39,946 --> 00:21:42,707 S1: working for them. And that is basically what the new 385 00:21:42,706 --> 00:21:46,306 S1: employees look like. That is what new employment looks like. 386 00:21:46,347 --> 00:21:48,907 S1: And if you are not one of those people, I 387 00:21:48,946 --> 00:21:52,787 S1: think you're in extreme danger right now. Poland is about 388 00:21:52,787 --> 00:21:57,387 S1: to overtake Japan's GDP. Llms outperform paid human persuaders in 389 00:21:57,387 --> 00:22:02,347 S1: both truth and deception. Claude three five was significantly better 390 00:22:02,347 --> 00:22:06,107 S1: at persuading people towards both correct and incorrect answers than 391 00:22:06,147 --> 00:22:10,867 S1: humans who had financial incentives. Sleep apnea pill shows striking 392 00:22:10,867 --> 00:22:16,907 S1: success in large clinical trial. Yeah. Cut breathing interruptions by 50%. 393 00:22:17,747 --> 00:22:22,827 S1: Denmark raises retirement age to 70 by 2040. A Guardian 394 00:22:22,827 --> 00:22:27,267 S1: writer downloaded years of Alexa data and basically found their 395 00:22:27,387 --> 00:22:32,587 S1: his family's soul. So 15,000 recorded conversations included his daughter 396 00:22:32,587 --> 00:22:35,587 S1: using Alexa as a therapist, asking about dating ages and 397 00:22:35,587 --> 00:22:40,946 S1: sleep problems, which she didn't discuss with the parents. Earth's 398 00:22:40,986 --> 00:22:44,787 S1: dual high tide phenomenon turns out there's actually two high tides, 399 00:22:44,827 --> 00:22:48,907 S1: one on opposite sides of the Earth. Lost dog found 400 00:22:48,946 --> 00:22:51,507 S1: thanks to dead AirTag, which someone put a new battery 401 00:22:51,507 --> 00:22:54,147 S1: in AirTag. And it lit up and the owner came 402 00:22:54,147 --> 00:22:59,746 S1: and got it. And, uh. Tim Hartfords four decade journey 403 00:22:59,747 --> 00:23:03,107 S1: with Dungeons and Dragons. So a missing teenager story in 404 00:23:03,107 --> 00:23:06,946 S1: 1979 caused a moral panic, and it made D&D famous 405 00:23:07,347 --> 00:23:13,837 S1: man 79. Wow. I didn't realize it was that old. Discovery. 406 00:23:14,476 --> 00:23:17,357 S1: Reject jealousy and root for your friends. My buddy Joseph 407 00:23:17,357 --> 00:23:20,716 S1: Thacker made a compelling case for rejecting jealousy and celebrating 408 00:23:20,716 --> 00:23:22,597 S1: your friends wins. You got to go read this essay. 409 00:23:23,476 --> 00:23:26,677 S1: What Sam Altman wishes someone had told him borrows from 410 00:23:26,677 --> 00:23:31,837 S1: Anthropic's Claude team. I'm going to say that again. Borrows 411 00:23:31,837 --> 00:23:35,397 S1: from Anthropic's Claude code team says context stuffing is way 412 00:23:35,397 --> 00:23:38,477 S1: better than rag. It's funny because I've been using context 413 00:23:38,476 --> 00:23:42,797 S1: stuffing as my main way of doing context. I don't 414 00:23:42,797 --> 00:23:46,637 S1: like rags. I've always thought that they were janky, and 415 00:23:46,677 --> 00:23:50,677 S1: it's great to hear Boris say that as well. Yeah, 416 00:23:50,716 --> 00:23:52,917 S1: Ilya explaining that AI is actually do understand this. What 417 00:23:52,917 --> 00:23:55,837 S1: I was talking about in the beginning inside the AI 418 00:23:55,877 --> 00:23:59,917 S1: security arsenal that I've built, someone shows off 21 plus 419 00:24:00,236 --> 00:24:03,677 S1: AI tools that they've created, from import table analysis to 420 00:24:03,716 --> 00:24:07,477 S1: threat hunting frameworks. Paul Graham on what makes writing good. 421 00:24:08,077 --> 00:24:16,286 S1: Internet Archive launches Live microfiche. Digitisation. Digitisation. Digitisation. Stream goodness. Digitisation. 422 00:24:16,767 --> 00:24:20,007 S1: So basically this is a live YouTube stream of people 423 00:24:20,007 --> 00:24:24,167 S1: digitizing microfiche. So they are just uploading things into the 424 00:24:24,167 --> 00:24:28,847 S1: Internet Archive. And yeah, that's the stream. You're watching the 425 00:24:28,847 --> 00:24:34,006 S1: internet be recorded into this giant archive. And yeah, um, 426 00:24:34,007 --> 00:24:40,047 S1: it's accompanied by relaxing lo fi beats. And there's a 427 00:24:40,047 --> 00:24:43,967 S1: Terraform MCP server integration tool between AI assistants. And Terraform 428 00:24:43,966 --> 00:24:48,446 S1: helps you discover, explore and understand infrastructure code faster. Uh, 429 00:24:48,607 --> 00:24:51,726 S1: got a way here to use git as S3. Someone 430 00:24:51,726 --> 00:24:54,927 S1: teaches data visualization with a bag of rocks. There's a 431 00:24:54,927 --> 00:25:00,047 S1: new paradigm for psychology proposed here. Um, control systems as 432 00:25:00,047 --> 00:25:02,247 S1: mental building blocks. I actually ordered this book. I can't 433 00:25:02,247 --> 00:25:05,446 S1: wait to go read it. Um, the book is called 434 00:25:05,446 --> 00:25:08,767 S1: The Mind and the wheel. Um. Oh, yeah, I gotta 435 00:25:08,807 --> 00:25:11,337 S1: I got a go download this thing. I think it's 436 00:25:11,337 --> 00:25:16,337 S1: already in my audible. Hold on one second. We're doing 437 00:25:16,337 --> 00:25:26,216 S1: it live. No, there is no audible version. Okay, well, whatever. 438 00:25:26,216 --> 00:25:31,976 S1: That's annoying. Um, anyway, I really want to learn this, uh, 439 00:25:31,976 --> 00:25:34,617 S1: or check it out, so see if it's cool. Um, 440 00:25:35,337 --> 00:25:37,296 S1: so it basically says the mind might function as a 441 00:25:37,297 --> 00:25:42,736 S1: collection of control systems that regulate everything from hunger to loneliness. And, uh, yeah, 442 00:25:42,777 --> 00:25:45,456 S1: I'm not sure if this is the dumbest thing ever 443 00:25:45,456 --> 00:25:47,897 S1: or the coolest thing ever, but worth including for you 444 00:25:47,897 --> 00:25:51,456 S1: to check out. Tycho is a new API that returns 445 00:25:51,456 --> 00:25:54,577 S1: visual answers from structured data sources in response to natural 446 00:25:54,577 --> 00:25:59,577 S1: language queries. So if data is trapped in databases or 447 00:25:59,577 --> 00:26:03,137 S1: web crawlers, you can now go and basically get that 448 00:26:03,137 --> 00:26:07,857 S1: data with natural language queries. A database of 1400 startup 449 00:26:07,857 --> 00:26:12,217 S1: ideas from Hacker News and Reddit. Decibels are silly. That 450 00:26:12,216 --> 00:26:14,536 S1: was a cool piece, basically talking about how the measurement 451 00:26:14,537 --> 00:26:18,736 S1: doesn't make any sense whatsoever. Google Gemini advanced ads GitHub 452 00:26:18,736 --> 00:26:22,137 S1: integration for code analysis, timeline views, and smart filtering on 453 00:26:22,137 --> 00:26:24,537 S1: Hacker News. It's a new Hacker News interface with a 454 00:26:24,537 --> 00:26:30,016 S1: lot of filtering options undetected. Tag stealth tracking for stolen items. 455 00:26:32,257 --> 00:26:36,817 S1: And ideas entering into the Member edition stuff here. Ideas. 456 00:26:38,017 --> 00:26:40,817 S1: So the idea of the season high agency and ideologies 457 00:26:40,817 --> 00:26:43,137 S1: as tools versus values. So a lot of interest right 458 00:26:43,137 --> 00:26:46,536 S1: now in what people are calling this high agency thing, which, uh, 459 00:26:47,657 --> 00:26:50,297 S1: I think there was a yeah, George Mac came on 460 00:26:50,297 --> 00:26:54,617 S1: Chris Williamson Modern Wisdom and I love the concept, and 461 00:26:54,617 --> 00:26:57,016 S1: I think a lot of people need more of it. 462 00:26:57,017 --> 00:27:00,976 S1: But the I would say most people, but not everyone, uh, 463 00:27:00,976 --> 00:27:03,816 S1: we actually talked about this in book club and uh, 464 00:27:03,817 --> 00:27:06,137 S1: someone in there was basically saying, look, this this might 465 00:27:06,137 --> 00:27:08,987 S1: be really nasty. You know, it seems kind of elitist 466 00:27:08,986 --> 00:27:12,827 S1: or like very talking about the NPC thing. And I 467 00:27:12,867 --> 00:27:16,067 S1: think anything taken to an extreme can be bad. Like 468 00:27:16,107 --> 00:27:18,066 S1: if you have too much capitalism, you need socialism, you 469 00:27:18,067 --> 00:27:20,907 S1: have too much socialism. You might need some capitalism. Uh, 470 00:27:20,907 --> 00:27:23,547 S1: too little self-esteem. You need more. But narcissism is bad. 471 00:27:23,587 --> 00:27:27,387 S1: You know, Stalinism is bad. But maybe Nordic countries got 472 00:27:27,387 --> 00:27:30,067 S1: something figured out. So one idea I put out there 473 00:27:30,067 --> 00:27:33,227 S1: was basically a set of fundamentals, like I talked about, um, 474 00:27:33,347 --> 00:27:36,147 S1: in my political post where it's like, these are my 475 00:27:36,186 --> 00:27:38,787 S1: core values. This is what I believe, you know, humanity 476 00:27:38,787 --> 00:27:41,067 S1: should look like. This is the ideal world that I 477 00:27:41,067 --> 00:27:44,547 S1: wish we lived in. And that's one set of things. 478 00:27:44,547 --> 00:27:49,627 S1: And then if someone says capitalism or whatever, some, uh, 479 00:27:50,387 --> 00:27:54,107 S1: you know, Marxism or some other thing or some sort 480 00:27:54,107 --> 00:27:56,587 S1: of ideology, think of the way I'm thinking of this 481 00:27:56,587 --> 00:28:00,346 S1: is those are tools. They're separate from values. So are 482 00:28:00,347 --> 00:28:03,827 S1: you a capitalist? No, no, what I am is a 483 00:28:03,867 --> 00:28:07,306 S1: Star Trek liberal, right? I want to elevate humanity. I 484 00:28:07,307 --> 00:28:10,956 S1: want humanity to be more, um, enabled. I want everyone 485 00:28:10,956 --> 00:28:14,597 S1: to have equal opportunities. I want I want to raise education. 486 00:28:14,597 --> 00:28:17,357 S1: I want to have everyone, you know, participating in creativity 487 00:28:17,357 --> 00:28:21,036 S1: and exchanging creativity with other humans. I want technology to 488 00:28:21,077 --> 00:28:23,637 S1: be in the background, not in the foreground. I want 489 00:28:23,677 --> 00:28:27,077 S1: humans to be in the foreground. Right. So those are 490 00:28:27,077 --> 00:28:29,957 S1: the core things that I think about kindness, helping others, 491 00:28:29,956 --> 00:28:34,517 S1: exchanging value with others. Right. Uh, freedom. All these core 492 00:28:34,557 --> 00:28:38,717 S1: core value things. Now, if someone hits me with, you know, 493 00:28:38,757 --> 00:28:43,557 S1: high value or no high agency or Marxism or whatever, 494 00:28:43,597 --> 00:28:46,557 S1: I'm going to listen, right? Okay, cool. Sounds awesome. But 495 00:28:46,557 --> 00:28:48,596 S1: that's a tool. That's not something that replaces one of 496 00:28:48,597 --> 00:28:51,957 S1: my values, right? It's a method for getting there. So 497 00:28:51,957 --> 00:28:54,277 S1: if I don't have enough socialism in my life to 498 00:28:54,317 --> 00:28:57,957 S1: achieve my values, we should sprinkle some more on there. 499 00:28:58,317 --> 00:29:00,997 S1: If someone sprinkles too much and I only taste socialism 500 00:29:00,997 --> 00:29:04,157 S1: on my sandwich or my steak or whatever it is, well, 501 00:29:04,157 --> 00:29:06,117 S1: that's too much. It went too far. And now you're 502 00:29:06,157 --> 00:29:09,887 S1: actually conflicting with my values. We need some more capitalism. Cool. 503 00:29:09,927 --> 00:29:12,727 S1: Let's do some more capitalism. It goes crazy, becomes like 504 00:29:12,727 --> 00:29:16,687 S1: a crony capitalism. It becomes banana republic. It becomes all 505 00:29:16,687 --> 00:29:19,046 S1: this stuff. Well guess what? We got too much capitalism 506 00:29:19,047 --> 00:29:22,407 S1: there need to take some of that off. So the 507 00:29:22,407 --> 00:29:24,846 S1: key thing is to be guided by these values and 508 00:29:24,847 --> 00:29:28,806 S1: then use everything else as like these different slider bars 509 00:29:28,807 --> 00:29:31,007 S1: of tools that you could use and adjust to how 510 00:29:31,007 --> 00:29:33,647 S1: much you, you know, adjust to taste, with the guiding 511 00:29:33,647 --> 00:29:36,727 S1: principle being not affecting the actual values of what you're 512 00:29:36,727 --> 00:29:41,647 S1: trying to do. So that was the idea of the week. Um, 513 00:29:41,687 --> 00:29:45,487 S1: got a simplified I security term breakdown, which, um, which 514 00:29:45,487 --> 00:29:49,047 S1: I basically modified off of my buddy Joseph Thacker's list. 515 00:29:49,047 --> 00:29:54,047 S1: So I alignment don't kill us. I safety don't help 516 00:29:54,047 --> 00:29:59,927 S1: us do bad stuff. I security basically I appsec oh 517 00:29:59,927 --> 00:30:03,207 S1: we already talked about discovery so we're good and now 518 00:30:03,967 --> 00:30:07,847 S1: recommendation of the week embrace chaotic reading if that happens 519 00:30:07,847 --> 00:30:10,727 S1: to you. I'm currently reading letters to a young poet, 520 00:30:10,727 --> 00:30:14,367 S1: the Dharma Bums. Don't tell me I can't. Rilke's Book 521 00:30:14,367 --> 00:30:17,567 S1: of Hours, the Pleasure of Finding Things out, The Peregrine, 522 00:30:17,807 --> 00:30:24,407 S1: The Rise of Theodore Roosevelt, Accelerando Dominion birds, Sex and Beauty, 523 00:30:24,567 --> 00:30:26,967 S1: the player of games. And they thought we were free. 524 00:30:27,327 --> 00:30:29,207 S1: I'm usually reading like 2 to 3 books at a time, 525 00:30:29,207 --> 00:30:32,487 S1: so this is way more than usual. Maybe more than ever. 526 00:30:32,527 --> 00:30:35,567 S1: What is this? Is this, like 20 books? Like 15 books? 527 00:30:35,567 --> 00:30:38,407 S1: Something like that. But I'm learning to be okay with this. 528 00:30:38,447 --> 00:30:41,247 S1: And this is my recommendation. This is my ask for you. 529 00:30:42,207 --> 00:30:44,207 S1: Be okay with it. If you're only reading one book, 530 00:30:44,247 --> 00:30:46,447 S1: be okay with it. If you're reading 15 and you're 531 00:30:46,487 --> 00:30:49,326 S1: switching between them, be okay with it. The one thing 532 00:30:49,367 --> 00:30:52,007 S1: to not be okay with in my mind is not reading. 533 00:30:52,007 --> 00:30:54,647 S1: So just read one book at a time or 17. 534 00:30:54,687 --> 00:30:59,407 S1: Don't be frustrated with yourself, just read. And the aphorism 535 00:30:59,407 --> 00:31:01,967 S1: of the week. The most important decision you make is 536 00:31:01,967 --> 00:31:05,047 S1: to be in a good mood. The most important decision 537 00:31:05,047 --> 00:31:07,007 S1: you make is to be in a good mood.