1 00:00:01,129 --> 00:00:04,460 S1: Welcome to Unsupervised Learning, a security, AI and meaning focused 2 00:00:04,460 --> 00:00:07,340 S1: podcast that looks at how best to thrive as humans 3 00:00:07,340 --> 00:00:11,570 S1: in a post AI world. It combines original ideas, analysis, 4 00:00:11,570 --> 00:00:14,810 S1: and mental models to bring not just the news, but 5 00:00:14,810 --> 00:00:22,340 S1: why it matters and how to respond. All right, welcome 6 00:00:22,340 --> 00:00:27,530 S1: to unsupervised learning. This is Daniel Miessler. Okay. Pretty much 7 00:00:27,530 --> 00:00:30,380 S1: heads down on doing talks and courses right now. Uh, 8 00:00:30,380 --> 00:00:32,750 S1: a bunch of essays, a bunch of video content. Feels 9 00:00:32,750 --> 00:00:35,030 S1: like I got a lot of ideas, but feels bad 10 00:00:35,030 --> 00:00:38,840 S1: to be behind. Matt Williams put out a quality introduction 11 00:00:38,840 --> 00:00:41,270 S1: to fabric on his YouTube channel, so that was cool. 12 00:00:41,270 --> 00:00:45,320 S1: Really well done. Video got the augmented course is updated. Uh, 13 00:00:45,320 --> 00:00:49,700 S1: so essentially we're expanding it to four plus hours where 14 00:00:49,700 --> 00:00:52,550 S1: it was only three hours before. So it's a lot 15 00:00:52,550 --> 00:00:55,840 S1: more content. Got a whole section on augmenting AI with 16 00:00:55,840 --> 00:00:59,680 S1: personal context, building your own work life workflows, which is 17 00:00:59,680 --> 00:01:01,840 S1: going to be super cool. We're actually going to do 18 00:01:01,840 --> 00:01:05,080 S1: it live with 1 or 2 people in the class 19 00:01:05,080 --> 00:01:08,320 S1: as well. So it's going to be kind of hands on. 20 00:01:08,319 --> 00:01:12,459 S1: We got a full section on obsidian as well, and uh, yeah, 21 00:01:12,459 --> 00:01:15,039 S1: just a whole bunch of fabric uses lots and lots 22 00:01:15,040 --> 00:01:18,789 S1: of examples, and I really like the cohesive way. It's 23 00:01:18,790 --> 00:01:23,200 S1: kind of pulling together the philosophical versus the technical. So 24 00:01:23,200 --> 00:01:26,410 S1: it's very practical in like, okay, you actually do this, 25 00:01:26,410 --> 00:01:29,950 S1: but it's kind of framed in this philosophical way, or 26 00:01:29,950 --> 00:01:32,020 S1: at least that's what I'm trying to pull off. Okay, 27 00:01:32,020 --> 00:01:35,470 S1: so I think I cracked Trump's popularity. I think unless 28 00:01:35,470 --> 00:01:37,750 S1: the DNC figures this out, it actually doesn't matter who 29 00:01:37,780 --> 00:01:40,300 S1: they run, they have to figure this out. So I 30 00:01:40,390 --> 00:01:41,710 S1: got a post on that. I'm not going to go 31 00:01:41,740 --> 00:01:45,009 S1: into that because that's politics stories. So there's a new 32 00:01:45,010 --> 00:01:48,610 S1: zero day and open SSH that allows remote code execution. 33 00:01:48,610 --> 00:01:51,700 S1: It's a little bit convoluted to sort of attack. And 34 00:01:51,700 --> 00:01:54,190 S1: I think people are figuring that out. One thing to 35 00:01:54,190 --> 00:01:56,200 S1: keep in mind though, is it always gets easier to 36 00:01:56,200 --> 00:01:58,600 S1: attack these things. It never gets harder to attack these 37 00:01:58,600 --> 00:02:01,540 S1: things unless people patch, of course. But even if it's 38 00:02:01,540 --> 00:02:04,480 S1: a little convoluted now, I mean, you still really need 39 00:02:04,480 --> 00:02:07,060 S1: to take a look at what you have exposed and 40 00:02:07,060 --> 00:02:10,450 S1: whether or not your particular SSH stack is vulnerable. There's 41 00:02:10,450 --> 00:02:14,649 S1: a full 10.0 critical vulnerability in juniper network routers that 42 00:02:14,650 --> 00:02:17,650 S1: basically allows you to bypass auth and take full control. 43 00:02:17,650 --> 00:02:23,380 S1: Cvss 10.0 snowflake had a breach, as everyone knows, and 44 00:02:23,380 --> 00:02:28,119 S1: it's expanding with over now 165 victims, including Ticketek and 45 00:02:28,120 --> 00:02:31,500 S1: Advanced Auto Parts and some folks from the Shiny Hunters 46 00:02:31,500 --> 00:02:35,970 S1: are saying that they access snowflake via third party contractors. 47 00:02:35,970 --> 00:02:39,600 S1: And as part of that snowflake incident, Santander's US branch 48 00:02:39,600 --> 00:02:43,739 S1: is notifying over 12,000 people that their personal info is stolen. 49 00:02:43,740 --> 00:02:45,780 S1: So this is one of those things where it's just, like, 50 00:02:45,780 --> 00:02:48,810 S1: contagious because the third party nature of it, it just 51 00:02:48,810 --> 00:02:52,709 S1: keeps spreading. And yeah, like we said above 165 victims 52 00:02:52,710 --> 00:02:56,030 S1: currently read Juliet, a Chinese state sponsored group, has been 53 00:02:56,030 --> 00:03:00,650 S1: exploiting network Edge devices to target Taiwanese government, academic, technology 54 00:03:00,650 --> 00:03:05,240 S1: and diplomatic organizations thanks to tines for sponsoring. And, uh, 55 00:03:05,240 --> 00:03:09,680 S1: if everyone remembers that orange uh, r1 I device. Well, 56 00:03:09,800 --> 00:03:13,309 S1: basically you can you can extract or it was possible 57 00:03:13,310 --> 00:03:18,520 S1: to extract all responses that ever came back from them and, uh, yeah, it's. 58 00:03:18,700 --> 00:03:21,790 S1: Basically nightmare fuel and anything you got back from your 59 00:03:21,790 --> 00:03:25,420 S1: personal AI. Device visible to whoever. I think this is 60 00:03:25,419 --> 00:03:29,740 S1: exactly what most security experts predicted with regard to AI 61 00:03:29,740 --> 00:03:34,900 S1: and security. Specifically that when startups do security, it's usually really, 62 00:03:34,900 --> 00:03:37,240 S1: really bad. But one, they don't have the expertise, they 63 00:03:37,240 --> 00:03:39,880 S1: don't have the resources, they don't have the time. And 64 00:03:39,880 --> 00:03:44,650 S1: they're already facing existential crises like every day, and security 65 00:03:44,650 --> 00:03:47,350 S1: usually isn't one of them. Another way to say that 66 00:03:47,350 --> 00:03:51,670 S1: is startups generally run with scissors, and AI startups run 67 00:03:51,670 --> 00:03:54,970 S1: extra fast with extra scissors. Like I've been saying, if 68 00:03:54,970 --> 00:03:57,610 S1: you think this is bad, wait until it's actually days 69 00:03:57,610 --> 00:04:00,910 S1: that are getting compromised where people have uploaded like their 70 00:04:00,910 --> 00:04:05,260 S1: traumas and their journals and their personal conversations and just everything, 71 00:04:05,260 --> 00:04:08,280 S1: their most intimate details. And when those start ups start 72 00:04:08,280 --> 00:04:11,220 S1: getting breached, it's going to be way worse. And this 73 00:04:11,220 --> 00:04:13,950 S1: is the attack surface map that I put together a 74 00:04:13,950 --> 00:04:16,830 S1: while back, and I think it's still pretty useful. Russian 75 00:04:16,830 --> 00:04:20,430 S1: hacking group Apt29, also known as Cozy Bear, breached team 76 00:04:20,430 --> 00:04:25,830 S1: viewers corporate IT environment. And this is another. So I've 77 00:04:25,830 --> 00:04:28,680 S1: migrated over the years to a very simple stance on 78 00:04:28,680 --> 00:04:32,550 S1: security tooling or really any core tooling. It is use 79 00:04:32,550 --> 00:04:35,820 S1: the official offerings from big companies whenever possible, and that's 80 00:04:35,820 --> 00:04:39,150 S1: because they have giant security teams, they have giant security budgets, 81 00:04:39,150 --> 00:04:41,549 S1: and they have a lot to lose in terms of 82 00:04:41,550 --> 00:04:44,580 S1: PR and market share. So basically, I only want to 83 00:04:44,580 --> 00:04:47,250 S1: trust my data to companies that have both the incentive 84 00:04:47,250 --> 00:04:50,789 S1: and the resources to protect that data. And those tend 85 00:04:50,790 --> 00:04:54,090 S1: to be the big players like Microsoft, Google, Apple, whatever. 86 00:04:54,089 --> 00:04:57,979 S1: Chinese hackers are using ransomware as a cover for cyber espionage. 87 00:04:57,980 --> 00:05:00,890 S1: Perplexity AI is under fire by a lot of people. 88 00:05:00,890 --> 00:05:03,380 S1: A lot of people really upset with them because they're 89 00:05:03,380 --> 00:05:08,150 S1: essentially scraping and crawling and, you know, basically feeding their 90 00:05:08,180 --> 00:05:13,250 S1: their AI with tactics that nobody really likes. And it's 91 00:05:13,250 --> 00:05:16,490 S1: kind of turning people off to it. Metaculus is launching 92 00:05:16,490 --> 00:05:20,120 S1: a series of quarterly tournaments to benchmark AI forecasting against 93 00:05:20,120 --> 00:05:24,770 S1: human forecasting on real world questions. So I am really 94 00:05:24,770 --> 00:05:29,539 S1: obsessed with this rigorous predictions, basically. So there are groups. 95 00:05:29,540 --> 00:05:33,050 S1: Metaculus is one where people make specific predictions. And I 96 00:05:33,050 --> 00:05:37,610 S1: learned about this from this book. Superforecasting. And this is 97 00:05:37,610 --> 00:05:42,140 S1: now very similar to Superforecasting, except for its AI players 98 00:05:42,140 --> 00:05:45,440 S1: playing addition to the human players and specifically their competing. 99 00:05:45,440 --> 00:05:48,920 S1: So I can't wait to watch this. It's really, really 100 00:05:48,920 --> 00:05:51,409 S1: exciting to me. I'm actually going to build myself a 101 00:05:51,410 --> 00:05:58,760 S1: little intelligence, daily, daily intelligence, brief product using substrate and 102 00:05:58,910 --> 00:06:01,669 S1: a bunch of the AI stuff that I'm building. So 103 00:06:01,670 --> 00:06:05,510 S1: I'm essentially going to be capturing a whole bunch of, 104 00:06:05,510 --> 00:06:08,830 S1: of these superforecasters combined with a whole bunch of point 105 00:06:08,830 --> 00:06:14,409 S1: sources like Osint people, national security people, financial information people 106 00:06:14,410 --> 00:06:17,799 S1: capturing all their point sources, capturing all the experts and 107 00:06:17,800 --> 00:06:22,450 S1: what they're predicting. And then having my AI basically collect 108 00:06:22,450 --> 00:06:25,660 S1: all that together, turn it into stories and narratives, and 109 00:06:25,660 --> 00:06:28,989 S1: most importantly, put like based on these experts, what are 110 00:06:28,990 --> 00:06:32,740 S1: the most likely outcomes over the next six months, 18 months, whatever, 111 00:06:32,740 --> 00:06:35,920 S1: three years. And it won't be perfect, obviously. I mean, 112 00:06:35,920 --> 00:06:39,130 S1: first of all, AI is good at, you know, building 113 00:06:39,130 --> 00:06:41,919 S1: narratives when they don't exist or whatever. So there's all 114 00:06:41,920 --> 00:06:44,590 S1: sorts of things that need to be careful of, but 115 00:06:44,589 --> 00:06:48,670 S1: with lots of really, really good input from these different sources, 116 00:06:48,670 --> 00:06:51,790 S1: I think there's a lot of potential here. Uh, first 117 00:06:51,790 --> 00:06:54,340 S1: of all, I can have it. I'll have the history 118 00:06:54,339 --> 00:06:56,969 S1: of these things. So as the AI gets better, or 119 00:06:56,970 --> 00:07:00,270 S1: as I write a better prompt for the eye or 120 00:07:00,270 --> 00:07:03,270 S1: set of prompts for the eye pipeline, all those results 121 00:07:03,270 --> 00:07:05,400 S1: will get better. In either way. I'm going to have 122 00:07:05,400 --> 00:07:09,330 S1: the results of the the point predictions and the expert predictions. 123 00:07:09,330 --> 00:07:12,300 S1: They'll all be stored, so I can always go retroactively 124 00:07:12,300 --> 00:07:14,910 S1: and build a better product. And people are talking about 125 00:07:14,910 --> 00:07:18,990 S1: how to run billion parameter scale llms on 13W of power, 126 00:07:18,990 --> 00:07:21,280 S1: which is 50 times more efficient. And this is what 127 00:07:21,280 --> 00:07:24,250 S1: I call slack in the rope, which is what Leopold 128 00:07:24,250 --> 00:07:26,890 S1: Aschenbrenner calls hobbling. And this is why I think we're 129 00:07:26,890 --> 00:07:30,430 S1: at like 1% of like where we are going. And 130 00:07:30,430 --> 00:07:32,830 S1: that might be way too large, actually. It might be 131 00:07:32,830 --> 00:07:36,130 S1: like 0.001%, who knows? But to me the game is 132 00:07:36,130 --> 00:07:40,960 S1: scale times algorithms, times tricks. So improve scale, improve the 133 00:07:40,960 --> 00:07:46,610 S1: algorithms and find tricks that magnify both of these. So 134 00:07:46,610 --> 00:07:49,040 S1: tricks are finding slack in the rope, which can potentially 135 00:07:49,040 --> 00:07:53,060 S1: massively improve the algorithms or advantages from scale. So these 136 00:07:53,060 --> 00:07:56,600 S1: two get magnified by this one. And Leopold is basically 137 00:07:56,600 --> 00:07:59,870 S1: calling this one removing hobbling. Yeah. And by the way 138 00:07:59,870 --> 00:08:04,670 S1: situational awareness by Leopold. It is like the best discussion 139 00:08:04,670 --> 00:08:08,120 S1: of this particular topic of like why you should believe 140 00:08:08,120 --> 00:08:10,770 S1: that we're going to scale in a certain pace. Businesses 141 00:08:10,770 --> 00:08:14,400 S1: are desperate for AI guidance, and big consulting firms are 142 00:08:14,400 --> 00:08:19,470 S1: stepping in to help. McKinsey says generative AI will be 40% 143 00:08:19,470 --> 00:08:24,720 S1: of its business this year. In 2024, 40% of McKinsey's 144 00:08:24,720 --> 00:08:29,130 S1: business is AI, and this basically started like three days 145 00:08:29,130 --> 00:08:31,470 S1: ago or 18 months ago. I mean, it's a blink 146 00:08:31,470 --> 00:08:34,600 S1: of an eye and it's almost half of their business. 147 00:08:34,600 --> 00:08:36,819 S1: And so here's a question how much of their business 148 00:08:36,820 --> 00:08:39,640 S1: is crypto related? Okay. If you're trying to compare, you're 149 00:08:39,640 --> 00:08:43,390 S1: trying to be like, oh, they're both hype. Uh, huge difference. 150 00:08:43,390 --> 00:08:48,370 S1: Alibaba's coin models take the top three spots on hugging face. 151 00:08:48,370 --> 00:08:52,240 S1: And a lot of us competitors are lagging behind. And 152 00:08:52,240 --> 00:08:56,679 S1: this new leaderboard is testing models on tasks like solving 153 00:08:56,679 --> 00:09:00,459 S1: 100 word murder mysteries and high school math equations. I 154 00:09:00,460 --> 00:09:05,230 S1: love the more practical and real and not trackable or 155 00:09:05,230 --> 00:09:09,579 S1: hackable or cheat able. These benchmarks are, and I don't 156 00:09:09,580 --> 00:09:12,940 S1: like the fact that these Chinese models are doing so well. 157 00:09:12,940 --> 00:09:15,910 S1: I think it's disturbing. AI and drone tech are two 158 00:09:15,910 --> 00:09:19,690 S1: places we absolutely need to be beat. China people in 159 00:09:19,690 --> 00:09:24,690 S1: high income democracies are increasingly satisfied with how democracy, dissatisfied 160 00:09:24,690 --> 00:09:29,309 S1: with how democracy is working. Since 2021, satisfaction has dropped 161 00:09:29,309 --> 00:09:32,970 S1: significantly in countries like Canada, Germany, Greece, South Korea, the UK, 162 00:09:33,000 --> 00:09:36,690 S1: the US and this is fine. A study showed that 163 00:09:36,690 --> 00:09:40,559 S1: loneliness in midlife is linked to believing in conspiracy theories. 164 00:09:40,559 --> 00:09:42,900 S1: And if I design an education curriculum, one of the 165 00:09:42,900 --> 00:09:45,600 S1: main themes will be hard work is leads you to 166 00:09:45,600 --> 00:09:48,600 S1: an easy life. Laziness leads you to a hard life 167 00:09:48,600 --> 00:09:51,839 S1: and the concept of resilience. And honestly, I would focus 168 00:09:51,840 --> 00:09:54,270 S1: a lot on the Stoics, but let me just pull 169 00:09:54,270 --> 00:09:57,930 S1: it up. Yeah, I love this graphic. Absolutely love this graphic. 170 00:09:57,929 --> 00:10:00,179 S1: It's just great. I'm gonna zoom in, look at this, 171 00:10:00,179 --> 00:10:04,470 S1: make hard decisions. Really, this kind of means discipline, right? 172 00:10:04,470 --> 00:10:08,400 S1: If you climb this mountain, you're you're doing self-discipline. And 173 00:10:08,400 --> 00:10:11,270 S1: you get to an easy life, easy decisions like, okay, 174 00:10:11,270 --> 00:10:15,020 S1: we're watching Netflix, we're doing cannabis, and boom, you slide 175 00:10:15,020 --> 00:10:18,050 S1: down here and now you're so far away from an 176 00:10:18,050 --> 00:10:20,780 S1: easy life. Now you have a hard life. Really powerful. Okay, 177 00:10:20,780 --> 00:10:24,830 S1: discovery project Nap time, Google's new AI framework for vulnerability 178 00:10:24,830 --> 00:10:29,150 S1: research lets humans take regular naps while it mimics human 179 00:10:29,150 --> 00:10:31,910 S1: security researchers. So it's just going to go off and 180 00:10:31,910 --> 00:10:34,640 S1: do its thing. The human could go away and it's 181 00:10:34,640 --> 00:10:37,849 S1: working on its whole thing. These frameworks just keep getting better. 182 00:10:37,850 --> 00:10:41,300 S1: So remember when we saw Will Smith eating the spaghetti 183 00:10:41,300 --> 00:10:43,219 S1: and it was like rah rah, rah. His mouth was 184 00:10:43,220 --> 00:10:45,590 S1: like giant. It was like totally messed up. The same 185 00:10:45,590 --> 00:10:48,710 S1: thing is going to happen with hacking frameworks, but there's 186 00:10:48,710 --> 00:10:51,140 S1: going to be a place at the top. Top 5%, 187 00:10:51,140 --> 00:10:54,530 S1: top 10%, top 1%. Depends where you cut it. But 188 00:10:54,530 --> 00:10:57,980 S1: either way, it's a small percentage, but still quite a 189 00:10:57,980 --> 00:11:02,260 S1: bit of room that basically only the really, really advanced 190 00:11:02,260 --> 00:11:06,490 S1: human testers can do right. If you find the top 1% 191 00:11:06,490 --> 00:11:10,679 S1: of pen testers bug bounty people or say, the 1% 192 00:11:10,679 --> 00:11:12,790 S1: or 1%, right? Which is still a lot of people, 193 00:11:12,790 --> 00:11:15,820 S1: keep in mind, 1% of 1% is still a lot 194 00:11:15,820 --> 00:11:19,030 S1: of people in in a very large space. Okay. And 195 00:11:19,030 --> 00:11:21,819 S1: bug bounty is pretty small, but pen testers are much, 196 00:11:21,820 --> 00:11:26,200 S1: much larger. But either way, let's just call it manual testers. 1% 197 00:11:26,200 --> 00:11:29,920 S1: of 1% of manual testers. They are doing things that 198 00:11:29,920 --> 00:11:33,459 S1: automation can't really do, and most manual testers can't really do, 199 00:11:33,460 --> 00:11:35,740 S1: and it's going to take a very long time. I 200 00:11:35,740 --> 00:11:38,770 S1: don't know how long maybe it's going to take full 201 00:11:38,770 --> 00:11:43,090 S1: AGI and possibly ASI and a whole lot more tricks 202 00:11:43,090 --> 00:11:46,479 S1: or anti hobbling in terms of the tool sets to 203 00:11:46,480 --> 00:11:49,390 S1: be able to replicate what they do. But the other 204 00:11:49,390 --> 00:11:53,410 S1: 90 to 90 5% or 99% or whatever it is 205 00:11:53,410 --> 00:11:57,910 S1: that is work that an average manual tester is doing. 206 00:11:57,910 --> 00:12:01,809 S1: Those frameworks will these frameworks will be able to copy 207 00:12:01,809 --> 00:12:04,420 S1: that very soon, I would say, in the next couple 208 00:12:04,420 --> 00:12:08,469 S1: of years, even now and even next year. And, you know, 209 00:12:08,470 --> 00:12:12,390 S1: it's just kind of spinning up. So imagine manual testing 210 00:12:12,390 --> 00:12:15,270 S1: massively being attacked. But does that mean it could do 211 00:12:15,270 --> 00:12:18,959 S1: everything that a really advanced attacker can do? No. And 212 00:12:18,960 --> 00:12:21,660 S1: that won't happen for quite some time. And the final 213 00:12:21,660 --> 00:12:24,030 S1: thing I say on this is these frameworks will be 214 00:12:24,030 --> 00:12:27,780 S1: used by all attackers and defenders, because you'll have to. 215 00:12:27,780 --> 00:12:31,290 S1: And the window between new vulnerabilities and either exploitation or 216 00:12:31,290 --> 00:12:36,160 S1: mitigation will shorten dramatically. So basically when everyone's running these 217 00:12:36,160 --> 00:12:39,790 S1: tools and they're constantly going and the moment you have 218 00:12:39,790 --> 00:12:42,910 S1: a new name published, it's instantly going to go find 219 00:12:42,910 --> 00:12:45,069 S1: all the subdomains, it's instantly going to go find all 220 00:12:45,070 --> 00:12:46,840 S1: the hosts. It's going to look at the hosts, it's 221 00:12:46,840 --> 00:12:49,060 S1: going to fingerprint them. It's going to do that. And 222 00:12:49,059 --> 00:12:51,610 S1: the defender needs to be doing that because the attacker 223 00:12:51,610 --> 00:12:53,979 S1: is going to be doing that right. And if there's 224 00:12:53,980 --> 00:12:57,550 S1: something vulnerable, oh, it's an open Postgres or whatever it 225 00:12:57,550 --> 00:13:00,740 S1: is and there's data in there. Well that is just 226 00:13:00,740 --> 00:13:03,020 S1: going to kick off an agent framework. It's going to 227 00:13:03,020 --> 00:13:05,059 S1: go download the stuff, it's going to parse the stuff. 228 00:13:05,059 --> 00:13:07,610 S1: It's going to turn it into a ransomware email. It's 229 00:13:07,610 --> 00:13:09,740 S1: going to find the people it should send that ransomware 230 00:13:09,740 --> 00:13:12,079 S1: email to. And like all this is just going to 231 00:13:12,080 --> 00:13:14,960 S1: be automated with AI. And so the defenders have to 232 00:13:14,960 --> 00:13:17,720 S1: be doing the exact same thing so they can block 233 00:13:17,720 --> 00:13:20,870 S1: it and do it beforehand. And importantly, when a new 234 00:13:20,870 --> 00:13:23,960 S1: vuln pops up or a new attack surface pops up, 235 00:13:23,960 --> 00:13:28,610 S1: the time between it coming available and either defense moving 236 00:13:28,610 --> 00:13:31,939 S1: on it, or attacker moving on, it is going to become, 237 00:13:31,940 --> 00:13:35,960 S1: you know, minutes or seconds instead of hours or days 238 00:13:35,960 --> 00:13:39,739 S1: or weeks or years. Extending Burpsuite for fun and profit 239 00:13:39,770 --> 00:13:44,780 S1: a guide by Federico Dota 11 labs Text, audio. They've 240 00:13:44,780 --> 00:13:48,250 S1: launched a new iOS app that sounds really good. I mean, 241 00:13:48,250 --> 00:13:51,969 S1: it sounds exactly like real people. I can't tell the difference. 242 00:13:51,970 --> 00:13:55,780 S1: Claud projects new feature in Claude. That's Anthropic's answer to 243 00:13:55,780 --> 00:14:00,250 S1: OpenAI assistance DApp, your new platform where publishers set a 244 00:14:00,250 --> 00:14:03,400 S1: price for using their content in model training. Kind of 245 00:14:03,400 --> 00:14:07,000 S1: like selling your medical data or something. A Better Paradise 246 00:14:07,000 --> 00:14:12,130 S1: Absurd ventures new podcast looks to elevate a fictional episodic 247 00:14:12,130 --> 00:14:16,090 S1: series with a billionaire leading the world towards a digital dystopia. 248 00:14:16,120 --> 00:14:18,850 S1: I actually want to go listen to this and recommendation 249 00:14:18,850 --> 00:14:20,230 S1: of the week. As soon as you get a chance, 250 00:14:20,230 --> 00:14:23,140 S1: go for a ride in a Waymo in San Francisco. 251 00:14:23,140 --> 00:14:26,800 S1: It is. It's open to everyone. Now you basically just 252 00:14:26,800 --> 00:14:28,540 S1: go get the app and you pay for it or whatever. 253 00:14:28,540 --> 00:14:31,750 S1: But it used to be a closed like alpha or beta. 254 00:14:31,750 --> 00:14:35,800 S1: But it is a remarkable experience. And what I want 255 00:14:35,800 --> 00:14:37,350 S1: you to do when you're in there is watch the 256 00:14:37,350 --> 00:14:41,310 S1: screen in the vehicle and look at all the dozens 257 00:14:41,310 --> 00:14:45,330 S1: or hundreds of things that it is tracking. So you 258 00:14:45,330 --> 00:14:47,640 S1: will see the dog across the street, you will see 259 00:14:47,640 --> 00:14:51,870 S1: the bicyclist. You will see the bicyclists, multiple bicyclists moving 260 00:14:51,870 --> 00:14:54,239 S1: in different directions. You will see people on the side 261 00:14:54,240 --> 00:14:55,950 S1: of the road. You will see when they cross over 262 00:14:55,950 --> 00:14:58,740 S1: into the street. And what you realize is like, that's 263 00:14:58,740 --> 00:15:01,120 S1: a lot of stuff to be tracking all at once. 264 00:15:01,120 --> 00:15:05,080 S1: And then you realize how distractible you are as a human. 265 00:15:05,080 --> 00:15:09,100 S1: You realize how distractible most drivers are as humans. You 266 00:15:09,100 --> 00:15:13,300 S1: realize the statistics of how many bicyclists get hit constantly, 267 00:15:13,300 --> 00:15:17,500 S1: every single year and, you know, either injured or sometimes killed. 268 00:15:17,500 --> 00:15:21,340 S1: And the reason isn't like evil drivers. The reason is 269 00:15:21,340 --> 00:15:23,830 S1: humans are bad drivers. I mean, there's going to come 270 00:15:23,830 --> 00:15:26,450 S1: a point at some point in the future where it's like, 271 00:15:26,450 --> 00:15:29,210 S1: you mean you really just had people and they were 272 00:15:29,210 --> 00:15:34,460 S1: manually controlling these cars right next to pedestrians and right, 273 00:15:34,460 --> 00:15:38,060 S1: right next to bicyclists? Like, how were they watching everything? Well, well, 274 00:15:38,060 --> 00:15:41,930 S1: the idea is the human driver would look forward and 275 00:15:41,930 --> 00:15:44,630 S1: they would just watch everything. It's like, yeah, yeah, but 276 00:15:44,630 --> 00:15:47,660 S1: but they can't see behind them. Well, you just you 277 00:15:47,660 --> 00:15:49,970 S1: just look behind you. That's all you do. You just 278 00:15:49,970 --> 00:15:53,270 S1: look behind you. Well, yeah. But then you're not looking forward. Well, 279 00:15:53,270 --> 00:15:55,340 S1: well yeah. But but when you need to look forward, 280 00:15:55,340 --> 00:15:57,380 S1: you just turn around again. And then you look forward 281 00:15:57,380 --> 00:16:00,470 S1: and you can look side to side. It worked, it worked. 282 00:16:00,470 --> 00:16:03,110 S1: It worked for a while. It's like explaining that to 283 00:16:03,110 --> 00:16:08,480 S1: somebody who has who's been driven around in automated vehicles 284 00:16:08,480 --> 00:16:12,380 S1: that watch everything all the time and never blink, never 285 00:16:12,380 --> 00:16:16,390 S1: get tired, never get sleepy. Never check text messages. They 286 00:16:16,390 --> 00:16:19,900 S1: just watch everything all the time and can instantly, like 287 00:16:19,900 --> 00:16:23,380 S1: swerve the car, stop the car, do whatever. If someone 288 00:16:23,380 --> 00:16:26,020 S1: does something stupid on a bike, they hit a pothole. 289 00:16:26,020 --> 00:16:28,359 S1: They fall in the road in front of you. Like 290 00:16:28,360 --> 00:16:30,040 S1: what are the chances? You're just going to miss that 291 00:16:30,040 --> 00:16:33,430 S1: because it's dark, or because you're tired, or because you've 292 00:16:33,430 --> 00:16:36,250 S1: been working three jobs and you're falling asleep or whatever 293 00:16:36,250 --> 00:16:38,680 S1: the reason. So think about that when you're looking at 294 00:16:38,680 --> 00:16:41,530 S1: the screen in a Waymo and the aphorism of the week, 295 00:16:41,530 --> 00:16:44,200 S1: every event has two handles, one by which it can 296 00:16:44,200 --> 00:16:47,050 S1: be carried and one by which it can't. Every event 297 00:16:47,050 --> 00:16:49,480 S1: has two handles, one by which it can be carried 298 00:16:49,480 --> 00:16:54,490 S1: and one by which it can't. Epictetus. Unsupervised learning is 299 00:16:54,490 --> 00:16:57,640 S1: produced and edited by Daniel Miessler on a Neumann U87 300 00:16:57,640 --> 00:17:01,720 S1: AI microphone using Hindenburg. Intro and outro music is by 301 00:17:01,720 --> 00:17:04,929 S1: Zomby with the Y, and to get the text and 302 00:17:04,930 --> 00:17:07,300 S1: links from this episode, sign up for the newsletter version 303 00:17:07,300 --> 00:17:12,970 S1: of the show at Daniel missler.com/newsletter. We'll see you next time.