WEBVTT - UL NO. 439: Humans vs. AI in Prediction Markets 0:00:00.050 --> 0:00:03.440 Whether you're starting or scaling your company's security program, demonstrating 0:00:03.470 --> 0:00:06.980 top notch security practices and establishing trust is more important 0:00:06.980 --> 0:00:12.110 than ever. Vanta automates compliance for Soc2, ISO 27,001 and more, 0:00:12.110 --> 0:00:16.070 saving you time and money while helping you build customer trust. Plus, 0:00:16.070 --> 0:00:20.329 you can streamline security reviews by automating questionnaires and demonstrating 0:00:20.329 --> 0:00:24.050 your security posture with a customer facing trust center, all 0:00:24.050 --> 0:00:28.610 powered by advanced AI. Over 7000 global companies like Atlassian, 0:00:28.610 --> 0:00:31.770 Flow Health and Quora use Vanta to manage risk and 0:00:31.770 --> 0:00:35.730 prove security in real time. Get $1,000 off Vanta when 0:00:35.729 --> 0:00:42.150 you go to Vanta comm slash unsupervised. That's vanta.com/supervised for 0:00:42.150 --> 0:00:47.370 $1,000 off. Welcome to Unsupervised Learning, a security, AI, and 0:00:47.370 --> 0:00:50.099 meaning focused podcast that looks at how best to thrive 0:00:50.100 --> 0:00:54.390 as humans in a post AI world. It combines original ideas, 0:00:54.390 --> 0:00:57.900 analysis and mental models to bring not just the news, 0:00:57.900 --> 0:01:05.429 but why it matters and how to respond. All right, 0:01:05.430 --> 0:01:10.800 welcome to unsupervised learning. This is Daniel Miessler. Okay. Pretty 0:01:10.800 --> 0:01:13.890 much heads down on doing talks and courses right now. Uh, 0:01:13.890 --> 0:01:16.020 a bunch of essays, a bunch of video content. It 0:01:16.020 --> 0:01:18.310 feels like I've got a lot of ideas but feels 0:01:18.310 --> 0:01:21.759 bad to be behind. Matt Williams put out a quality 0:01:21.760 --> 0:01:24.760 introduction to fabric on his YouTube channel, so that was cool. 0:01:24.760 --> 0:01:28.840 Really well done. Video got the augmented course is updated. Uh, 0:01:28.840 --> 0:01:33.190 so essentially we're expanding it to four plus hours where 0:01:33.190 --> 0:01:36.070 it was only three hours before. So it's a lot 0:01:36.069 --> 0:01:39.340 more content. Got a whole section on augmenting AI with 0:01:39.340 --> 0:01:43.180 personal context, building your own work life workflows, which is 0:01:43.180 --> 0:01:45.339 going to be super cool. We're actually going to do 0:01:45.340 --> 0:01:48.580 it live with 1 or 2 people in the class 0:01:48.580 --> 0:01:51.820 as well, so it's going to be kind of hands on. 0:01:51.820 --> 0:01:55.960 We got a full section on obsidian as well, and uh, yeah, 0:01:55.960 --> 0:01:58.540 just a whole bunch of fabric uses lots and lots 0:01:58.540 --> 0:02:02.080 of examples. And, uh, I really like the cohesive way. 0:02:02.080 --> 0:02:06.350 It's kind of pulling together the philosophical versus the technical. 0:02:06.350 --> 0:02:09.889 So it's very practical in like, okay, you actually do this, 0:02:09.889 --> 0:02:13.460 but it's kind of framed in this philosophical way, or 0:02:13.460 --> 0:02:15.530 at least that's what I'm trying to pull off. Okay, 0:02:15.530 --> 0:02:18.980 so I think I cracked Trump's popularity. I think unless 0:02:18.980 --> 0:02:21.260 the DNC figures this out, it actually doesn't matter who 0:02:21.290 --> 0:02:23.870 they run, they have to figure this out. So I 0:02:23.870 --> 0:02:25.430 got a post on that. Not going to go into 0:02:25.430 --> 0:02:28.820 that because that's politics stories. So there's a new zero 0:02:28.820 --> 0:02:32.460 day in OpenSSH that allows remote code execution. It's a 0:02:32.460 --> 0:02:35.610 little bit convoluted to sort of attack, and I think 0:02:35.610 --> 0:02:38.370 people are figuring that out. One thing to keep in mind, though, 0:02:38.370 --> 0:02:40.770 is it always gets easier to attack these things. It 0:02:40.770 --> 0:02:43.440 never gets harder to attack these things unless people patch, 0:02:43.470 --> 0:02:46.320 of course. But even if it's a little convoluted now, 0:02:46.320 --> 0:02:48.570 I mean, you still really need to take a look 0:02:48.570 --> 0:02:51.210 at what you have exposed and whether or not your 0:02:51.210 --> 0:02:55.230 particular SSH stack is vulnerable. There's a full 10.0 critical 0:02:55.230 --> 0:02:59.370 vulnerability in juniper network routers that basically allows you to 0:02:59.370 --> 0:03:05.760 bypass authentic full control. Cvss 10.0 snowflake had a breach, 0:03:05.760 --> 0:03:09.750 as everyone knows, and it's expanding with over now 165 victims, 0:03:09.750 --> 0:03:14.010 including Ticketek and Advanced Auto Parts. And some folks from 0:03:14.010 --> 0:03:18.160 the Shiny Hunters are saying that they access snowflake via 0:03:18.160 --> 0:03:21.639 third party contractors. And as part of that snowflake incident, 0:03:21.639 --> 0:03:25.930 Santander's US branch is notifying over 12,000 people that their 0:03:25.930 --> 0:03:28.180 personal info is stolen. So this is one of those 0:03:28.180 --> 0:03:31.300 things where it's just, like, contagious because the third party 0:03:31.300 --> 0:03:34.030 nature of it, it just keeps spreading. And yeah, like 0:03:34.030 --> 0:03:38.140 we said above 165 victims currently read Juliet, a Chinese 0:03:38.140 --> 0:03:41.560 state sponsored group, has been exploiting network Edge devices to 0:03:41.560 --> 0:03:46.720 target Taiwanese government, academic, technology and diplomatic organizations thanks to 0:03:46.720 --> 0:03:51.820 tines for sponsoring. And if everyone remembers that orange R1 0:03:51.850 --> 0:03:56.170 I device, well, basically you can you can extract or 0:03:56.170 --> 0:03:59.380 it was possible to extract all responses that ever came 0:03:59.380 --> 0:04:03.550 back from them and uh, yeah, it basically nightmare fuel 0:04:03.550 --> 0:04:06.830 and anything you got back from your personal AI device 0:04:06.830 --> 0:04:10.430 visible to whoever. I think this is exactly what most 0:04:10.430 --> 0:04:14.750 security experts predicted with regard to AI and security. Specifically 0:04:14.750 --> 0:04:19.160 that when startups do security, it's usually really, really bad. 0:04:19.160 --> 0:04:21.020 But one, they don't have the expertise, they don't have 0:04:21.020 --> 0:04:24.200 the resources, they don't have the time. And they're already 0:04:24.200 --> 0:04:28.880 facing existential crises like every day. And security usually isn't 0:04:28.880 --> 0:04:31.680 one of them. Another way to say that is startups 0:04:31.680 --> 0:04:35.940 generally run with scissors, and AI startups run extra fast 0:04:35.940 --> 0:04:38.789 with extra scissors. Like I've been saying, if you think 0:04:38.790 --> 0:04:41.220 this is bad, wait. Wait until it's actually days that 0:04:41.220 --> 0:04:44.940 are getting compromised where people have uploaded like their traumas 0:04:44.940 --> 0:04:48.750 and their journals and their personal conversations and just everything, 0:04:48.750 --> 0:04:52.680 their most intimate details. And when those startups start getting breached, 0:04:52.680 --> 0:04:55.200 it's going to be way worse. And this is the 0:04:55.200 --> 0:04:58.110 attack surface map that I put together a while back, 0:04:58.110 --> 0:05:02.040 and I think it's still pretty useful. Russian hacking group Apt29, 0:05:02.040 --> 0:05:06.180 also known as Cozy Bear, breached team viewers corporate IT environment. 0:05:06.180 --> 0:05:10.800 And this is another. So I've migrated over the years 0:05:10.800 --> 0:05:13.650 to a very simple stance on security tooling or really 0:05:13.650 --> 0:05:17.370 any core tooling. It is use the official offerings from 0:05:17.370 --> 0:05:20.349 big companies whenever possible, and that's because they have giant 0:05:20.350 --> 0:05:23.260 security teams, they have giant security budgets, and they have 0:05:23.260 --> 0:05:26.410 a lot to lose in terms of PR and market share. 0:05:26.410 --> 0:05:28.960 So basically, I only want to trust my data to 0:05:28.960 --> 0:05:32.620 companies that have both the incentive and the resources to 0:05:32.620 --> 0:05:34.930 protect that data. And those tend to be the big 0:05:34.930 --> 0:05:39.010 players like Microsoft, Google, Apple, whatever. Chinese hackers are using 0:05:39.010 --> 0:05:42.850 ransomware as a cover for cyber espionage. Perplexity AI is 0:05:42.850 --> 0:05:44.620 under fire by a lot of people. A lot of 0:05:44.620 --> 0:05:48.100 people are really upset with them because they're essentially scraping 0:05:48.100 --> 0:05:52.510 and crawling and, you know, basically feeding their their AI 0:05:52.510 --> 0:05:57.039 with tactics that nobody really likes. And it's kind of 0:05:57.040 --> 0:06:00.430 turning people off to it. Metaculus is launching a series 0:06:00.430 --> 0:06:04.690 of quarterly tournaments to benchmark AI forecasting against human forecasting 0:06:04.690 --> 0:06:08.779 on real world questions. So I am really obsessed with 0:06:08.779 --> 0:06:13.909 this rigorous predictions basically. So there are groups metaculus is 0:06:13.910 --> 0:06:16.940 one where people make specific predictions. And I learned about 0:06:16.940 --> 0:06:22.460 this from this book. Superforecasting. And this is now very 0:06:22.460 --> 0:06:26.510 similar to Superforecasting, except for it's AI players playing addition 0:06:26.510 --> 0:06:29.599 to the human players. And specifically they're competing. So I 0:06:29.600 --> 0:06:33.420 can't wait to watch this. It's really, really exciting to me. 0:06:33.420 --> 0:06:37.770 I'm actually going to build myself a little intelligence, uh, daily, 0:06:37.770 --> 0:06:42.419 daily intelligence, brief product, uh, using substrate and, um, a 0:06:42.540 --> 0:06:45.270 bunch of the AI stuff that I'm building. So I'm 0:06:45.270 --> 0:06:49.320 essentially going to be capturing a whole bunch of, of 0:06:49.320 --> 0:06:52.830 these superforecasters combined with a whole bunch of point sources 0:06:52.830 --> 0:06:58.440 like Osint people, national security people, financial information, people capturing 0:06:58.440 --> 0:07:01.500 all their point sources, capturing all the experts and what 0:07:01.500 --> 0:07:06.089 they're predicting, and then having my AI basically collect all 0:07:06.089 --> 0:07:10.020 that together, turn it into stories and narratives, and most importantly, 0:07:10.020 --> 0:07:12.960 put like based on these experts, what are the most 0:07:12.960 --> 0:07:16.260 likely outcomes over the next six months, 18 months, whatever, 0:07:16.260 --> 0:07:19.420 three years. And it won't be perfect, obviously. I mean, 0:07:19.420 --> 0:07:22.630 first of all, I is good at, you know, building 0:07:22.630 --> 0:07:25.420 narratives when they don't exist or whatever. So there's all 0:07:25.420 --> 0:07:28.090 sorts of things I need to be careful of, but 0:07:28.090 --> 0:07:32.200 with lots of really, really good input from these different sources, 0:07:32.200 --> 0:07:35.290 I think there's a lot of potential here. Uh, first 0:07:35.290 --> 0:07:37.840 of all, I can have it. I'll have the history 0:07:37.840 --> 0:07:40.480 of these things. So as the AI gets better, or 0:07:40.480 --> 0:07:43.750 as I write a better prompt for the eye or 0:07:43.750 --> 0:07:46.750 set of prompts for the eye pipeline, all those results 0:07:46.750 --> 0:07:48.910 will get better. And either way, I'm going to have 0:07:48.910 --> 0:07:52.840 the results of the the point predictions and the expert predictions. 0:07:52.840 --> 0:07:55.780 They'll all be stored, so I can always go retroactively 0:07:55.780 --> 0:07:58.420 and build a better product. And people are talking about 0:07:58.420 --> 0:08:02.500 how to run billion parameter scale llms on 13W of power, 0:08:02.500 --> 0:08:04.780 which is 50 times more efficient. And this is what 0:08:04.780 --> 0:08:07.760 I call slack in the rope, which is what Leopold 0:08:07.760 --> 0:08:10.370 Aschenbrenner calls hobbling. And this is why I think we're 0:08:10.370 --> 0:08:13.910 at like 1% of like where we are going. And 0:08:13.910 --> 0:08:16.310 that might be way too large, actually. It might be 0:08:16.310 --> 0:08:19.640 like 0.001%, who knows? But to me the game is 0:08:19.640 --> 0:08:24.440 scale times algorithms, times tricks. So improve scale, improve the 0:08:24.440 --> 0:08:30.110 algorithms and find tricks that magnify both of these. So 0:08:30.110 --> 0:08:32.550 tricks are finding slack on the rope, which can potentially 0:08:32.550 --> 0:08:36.570 massively improve the algorithms or advantages from scale. So these 0:08:36.570 --> 0:08:40.080 two get magnified by this one. And Leopold is basically 0:08:40.080 --> 0:08:43.380 calling this one removing hobbling. Yeah. And by the way 0:08:43.380 --> 0:08:48.179 situational awareness by Leopold. It is like the best discussion 0:08:48.179 --> 0:08:51.630 of this particular topic of like why you should believe 0:08:51.630 --> 0:08:54.210 that we're going to scale in a certain pace. Businesses 0:08:54.210 --> 0:08:57.870 are desperate for AI guidance, and big consulting firms are 0:08:57.870 --> 0:09:02.910 stepping in to help. McKinsey says generative AI will be 40% 0:09:02.910 --> 0:09:08.160 of its business this year. In 2024, 40% of McKinsey's 0:09:08.160 --> 0:09:12.570 business is AI, and this basically started like three days 0:09:12.570 --> 0:09:14.910 ago or 18 months ago. I mean, it's a blink 0:09:14.910 --> 0:09:18.120 of an eye and it's almost half of their business. 0:09:18.120 --> 0:09:20.319 And so here's a question how much of their business 0:09:20.320 --> 0:09:23.140 is crypto related? Okay. If you're trying to compare, you're 0:09:23.140 --> 0:09:26.860 trying to be like, oh, they're both hype. Uh, huge difference. 0:09:26.860 --> 0:09:31.870 Alibaba's coin models take the top three spots on hugging face. 0:09:31.870 --> 0:09:35.740 And a lot of us competitors are lagging behind. And 0:09:35.740 --> 0:09:40.179 this new leaderboard is testing models on tasks like solving 0:09:40.179 --> 0:09:43.959 100 word murder mysteries and high school math equations. I 0:09:43.960 --> 0:09:48.730 love the more practical and real and not trackable or 0:09:48.730 --> 0:09:53.079 hackable or cheat able. These benchmarks are, and I don't 0:09:53.080 --> 0:09:56.440 like the fact that these Chinese models are doing so well. 0:09:56.440 --> 0:09:59.410 I think it's disturbing. AI and drone tech are two 0:09:59.410 --> 0:10:03.190 places we absolutely need to be beat. China. People in 0:10:03.190 --> 0:10:08.180 high income democracies are increasingly satisfied with how democracy dissatisfied 0:10:08.179 --> 0:10:12.800 with how democracy is working. Since 2021, satisfaction has dropped 0:10:12.800 --> 0:10:16.460 significantly in countries like Canada, Germany, Greece, South Korea, the UK, 0:10:16.490 --> 0:10:20.180 the US and this is fine. A study showed that 0:10:20.179 --> 0:10:24.050 loneliness in midlife is linked to believing in conspiracy theories. 0:10:24.050 --> 0:10:26.420 And if I design an education curriculum, one of the 0:10:26.420 --> 0:10:29.090 main themes will be hard work is leads you to 0:10:29.090 --> 0:10:32.100 an easy life. Laziness leads you to a hard life 0:10:32.100 --> 0:10:35.339 and the concept of resilience. And honestly, I would focus 0:10:35.340 --> 0:10:37.770 a lot on the Stoics, but let me just pull 0:10:37.770 --> 0:10:41.430 it up. Yeah, I love this graphic. Absolutely love this graphic. 0:10:41.429 --> 0:10:43.679 It's just great. I'm gonna zoom in, look at this, 0:10:43.679 --> 0:10:47.970 make hard decisions. Really, this kind of means discipline, right? 0:10:47.970 --> 0:10:51.900 If you climb this mountain, you're you're doing self-discipline. And 0:10:51.900 --> 0:10:54.780 you get to an easy life, easy decisions like, okay, 0:10:54.780 --> 0:10:58.500 we're watching Netflix, we're doing cannabis, and boom, you slide 0:10:58.500 --> 0:11:01.530 down here and now you're so far away from an 0:11:01.530 --> 0:11:04.290 easy life. Now you have a hard life. Really powerful. Okay, 0:11:04.290 --> 0:11:08.340 discovery project Nap time, Google's new AI framework for vulnerability 0:11:08.340 --> 0:11:12.630 research lets humans take regular naps while it mimics human 0:11:12.630 --> 0:11:15.420 security researchers. So it's just going to go off and 0:11:15.420 --> 0:11:18.150 do its thing. The human could go away and it's 0:11:18.150 --> 0:11:21.370 working on its whole thing. These frameworks just keep getting better. 0:11:21.370 --> 0:11:24.790 So remember when we saw Will Smith eating the spaghetti 0:11:24.790 --> 0:11:26.709 and it was like rah rah, rah. His mouth was 0:11:26.710 --> 0:11:29.080 like giant. It was like totally messed up. The same 0:11:29.080 --> 0:11:32.199 thing is going to happen with hacking frameworks, but there's 0:11:32.200 --> 0:11:34.660 going to be a place at the top. Top 5%, 0:11:34.660 --> 0:11:38.020 top 10%, top 1%. Depends where you cut it. But 0:11:38.020 --> 0:11:41.469 either way, it's a small percentage, but still quite a 0:11:41.470 --> 0:11:45.760 bit of room that basically only the really, really advanced 0:11:45.760 --> 0:11:49.990 human testers can do right. If you find the top 1% 0:11:49.990 --> 0:11:54.160 of pen testers bug bounty people or say, the 1% 0:11:54.160 --> 0:11:56.290 or 1%, right? Which is still a lot of people, 0:11:56.290 --> 0:11:59.320 keep in mind, 1% of 1% is still a lot 0:11:59.320 --> 0:12:02.530 of people in in a very large space. Okay. And 0:12:02.530 --> 0:12:05.319 bug bounty is pretty small, but pen testers are much, 0:12:05.320 --> 0:12:09.690 much larger. But either way, let's just call it manual testers. 1% 0:12:09.690 --> 0:12:13.440 of 1% of manual testers. They are doing things that 0:12:13.440 --> 0:12:16.980 automation can't really do, and most manual testers can't really do, 0:12:16.980 --> 0:12:19.260 and it's going to take a very long time. I 0:12:19.260 --> 0:12:22.290 don't know how long maybe it's going to take full 0:12:22.290 --> 0:12:26.610 AGI and possibly ASI and a whole lot more tricks 0:12:26.610 --> 0:12:30.000 or anti hobbling in terms of the tool sets to 0:12:30.000 --> 0:12:32.880 be able to replicate what they do. but the other 0:12:32.880 --> 0:12:36.929 90 to 90 5% or 99% or whatever it is 0:12:36.929 --> 0:12:41.400 that is work that an average manual tester is doing. 0:12:41.400 --> 0:12:45.300 Those frameworks will these frameworks will be able to copy 0:12:45.300 --> 0:12:47.910 that very soon, I would say, in the next couple 0:12:47.910 --> 0:12:51.959 of years, even now and even next year. And, you know, 0:12:51.960 --> 0:12:55.890 it's just kind of spinning up. So imagine manual testing 0:12:55.890 --> 0:12:58.770 massively being attacked. But does that mean it could do 0:12:58.770 --> 0:13:02.459 everything that a really advanced attacker can do? No. And 0:13:02.460 --> 0:13:05.160 that won't happen for quite some time. And the final 0:13:05.160 --> 0:13:07.530 thing I say on this is these frameworks will be 0:13:07.530 --> 0:13:11.280 used by all attackers and defenders, because you'll have to. 0:13:11.280 --> 0:13:14.790 And the window between new vulnerabilities and either exploitation or 0:13:14.790 --> 0:13:19.650 mitigation will shorten dramatically. So basically when everyone's running these 0:13:19.650 --> 0:13:23.290 tools and they're constantly going, and the moment you have 0:13:23.290 --> 0:13:26.410 a new name published, it's instantly going to go find 0:13:26.410 --> 0:13:28.540 all the subdomains. It's instantly going to go find all 0:13:28.540 --> 0:13:30.339 the hosts. It's going to look at the hosts, it's 0:13:30.340 --> 0:13:32.560 going to fingerprint them. It's going to do that. And 0:13:32.559 --> 0:13:35.110 the defender needs to be doing that because the attacker 0:13:35.110 --> 0:13:37.479 is going to be doing that right. And if there's 0:13:37.480 --> 0:13:41.050 something vulnerable, oh, it's an open Postgres or whatever it 0:13:41.050 --> 0:13:44.230 is and there's data in there. Well that is just 0:13:44.230 --> 0:13:46.510 going to kick off an agent framework. It's going to 0:13:46.510 --> 0:13:48.550 go download the stuff, it's going to parse the stuff. 0:13:48.550 --> 0:13:51.130 It's going to turn it into a ransomware email. It's 0:13:51.130 --> 0:13:53.260 going to find the people it should send that ransomware 0:13:53.260 --> 0:13:55.569 email to. And like all this is just going to 0:13:55.570 --> 0:13:58.480 be automated with AI. And so the defenders have to 0:13:58.480 --> 0:14:01.240 be doing the exact same thing so they can block 0:14:01.240 --> 0:14:04.359 it and do it beforehand. And importantly, when a new 0:14:04.360 --> 0:14:07.480 vuln pops up or a new attack surface pops up, 0:14:07.480 --> 0:14:12.110 the time between it coming available and either defense moving 0:14:12.110 --> 0:14:15.439 on it, or attacker moving on, it is going to become, 0:14:15.440 --> 0:14:19.460 you know, minutes or seconds instead of hours or days 0:14:19.460 --> 0:14:22.430 or weeks or years. Extending Burp Suite for fun and 0:14:22.430 --> 0:14:27.860 Profit a guide by Federico Dota 11 labs text, audio. 0:14:27.890 --> 0:14:31.130 They've launched a new iOS app that sounds really good. 0:14:31.130 --> 0:14:34.650 I mean, it sounds exactly like real people. I can't 0:14:34.650 --> 0:14:37.680 tell the difference. Claud projects new feature in Claude. That's 0:14:37.680 --> 0:14:42.780 Anthropic's answer to OpenAI assistance DApp, your new platform where 0:14:42.780 --> 0:14:46.620 publishers set a price for using their content in model training. 0:14:46.620 --> 0:14:49.440 Kind of like selling your medical data or something. A 0:14:49.440 --> 0:14:53.880 Better Paradise Absurd ventures new podcast looks to elevate a 0:14:53.880 --> 0:14:58.380 fictional episodic series with a billionaire leading the world towards 0:14:58.380 --> 0:15:01.050 a digital dystopia. I actually want to go listen to 0:15:01.050 --> 0:15:03.150 this and recommendation of the week. As soon as you 0:15:03.150 --> 0:15:05.190 get a chance, go for a ride in a Waymo 0:15:05.190 --> 0:15:09.450 in San Francisco. It is. It's open to everyone. Now 0:15:09.450 --> 0:15:11.250 you basically just go get the app and you pay 0:15:11.250 --> 0:15:13.350 for it or whatever. But it used to be a 0:15:13.350 --> 0:15:17.790 closed like alpha or beta, but it is a remarkable experience. 0:15:17.790 --> 0:15:20.040 And what I want you to do when you're in 0:15:20.040 --> 0:15:22.660 there is watch the screen in the vehicle and look 0:15:22.660 --> 0:15:27.100 at all the dozens or hundreds of things that it 0:15:27.100 --> 0:15:30.640 is tracking. So you will see the dog across the street, 0:15:30.640 --> 0:15:33.610 you will see the bicyclist. You will see the bicyclists, 0:15:33.610 --> 0:15:37.359 multiple bicyclists moving in different directions. You will see people 0:15:37.360 --> 0:15:38.770 on the side of the road. You will see when 0:15:38.770 --> 0:15:41.380 they cross over into the street. And what you realize 0:15:41.380 --> 0:15:43.780 is like, that's a lot of stuff to be tracking 0:15:43.780 --> 0:15:47.620 all at once. And then you realize how distractible you 0:15:47.620 --> 0:15:51.280 are as a human. You realize how distractible most drivers 0:15:51.280 --> 0:15:54.580 are as humans. You realize the statistics of how many 0:15:54.580 --> 0:15:58.720 bicyclists get hit constantly, every single year and, you know, 0:15:58.720 --> 0:16:02.980 either injured or sometimes killed. And the reason isn't like 0:16:02.980 --> 0:16:06.580 evil drivers. The reason is humans are bad drivers. I mean, 0:16:06.580 --> 0:16:08.770 there's going to come a point at some point in 0:16:08.780 --> 0:16:11.030 the future where it's like, you mean you really just 0:16:11.030 --> 0:16:15.500 had people and they were manually controlling these cars right 0:16:15.500 --> 0:16:19.700 next to pedestrians and right next to bicyclists? Like, how 0:16:19.700 --> 0:16:23.510 were they watching everything? Well, well, the idea is the 0:16:23.510 --> 0:16:26.810 human driver would look forward and they would just watch everything. 0:16:26.810 --> 0:16:30.260 It's like, yeah, yeah, but but they can't see behind them. Well, 0:16:30.260 --> 0:16:33.090 you just you just look behind you. That's all you do. 0:16:33.120 --> 0:16:35.310 You just look behind you. Well, yeah, but then you're 0:16:35.310 --> 0:16:38.190 not looking forward. Well, well yeah. But but when you 0:16:38.190 --> 0:16:40.110 need to look forward, you just turn around again. And 0:16:40.110 --> 0:16:42.270 then you look forward and you can look side to side. 0:16:42.270 --> 0:16:45.540 It worked, it worked. It worked for a while. It's 0:16:45.540 --> 0:16:49.890 like explaining that to somebody who has who's been driven 0:16:49.890 --> 0:16:54.060 around in automated vehicles that watch everything all the time 0:16:54.060 --> 0:16:58.170 and never blink, never get tired, never get sleepy. Never 0:16:58.170 --> 0:17:01.590 check text messages. They just watch everything all the time 0:17:01.590 --> 0:17:05.550 and can instantly, like swerve the car, stop the car, 0:17:05.550 --> 0:17:08.430 do whatever. If someone does something stupid on a bike, 0:17:08.430 --> 0:17:10.770 they hit a pothole. They fall in the road in 0:17:10.770 --> 0:17:12.840 front of you. Like what are the chances? You're just 0:17:12.840 --> 0:17:16.080 going to miss that because it's dark, or because you're tired, 0:17:16.080 --> 0:17:18.630 or because you've been working three jobs and you're falling 0:17:18.630 --> 0:17:21.580 asleep or whatever the reason. So think about that when 0:17:21.580 --> 0:17:23.709 you're looking at the screen in a Waymo and the 0:17:23.710 --> 0:17:27.070 aphorism of the week, every event has two handles, one 0:17:27.070 --> 0:17:29.140 by which it can be carried and one by which 0:17:29.170 --> 0:17:32.140 it can't. Every event has two handles, one by which 0:17:32.170 --> 0:17:35.530 it can be carried and one by which it can't. Epictetus. 0:17:36.850 --> 0:17:39.970 Unsupervised learning is produced and edited by Daniel Miessler on 0:17:39.970 --> 0:17:44.590 a Neumann U87 AI microphone using Hindenburg. Intro and outro 0:17:44.590 --> 0:17:47.920 music is by Zomby with the Y, and to get 0:17:47.920 --> 0:17:49.990 the text and links from this episode, sign up for 0:17:49.990 --> 0:17:55.630 the newsletter version of the show at Daniel miessler.com/newsletter. We'll 0:17:55.630 --> 0:17:56.470 see you next time.