WEBVTT - UL NO. 467 | Why You Should Care About AGI (And a Definition) 0:00:00.920 --> 0:00:05.040 Unsupervised Learning is a podcast about trends and ideas in cybersecurity, 0:00:05.160 --> 0:00:10.000 national security, AI, technology and society, and how best to 0:00:10.039 --> 0:00:17.840 upgrade ourselves to be ready for what's coming. All right, 0:00:17.840 --> 0:00:21.320 welcome to unsupervised learning. This is Daniel Miessler. All right. 0:00:21.320 --> 0:00:25.120 So absolutely must see conversation. I'm going to open this 0:00:25.120 --> 0:00:25.800 up real quick. 0:00:26.640 --> 0:00:27.400 One of the top. 0:00:28.200 --> 0:00:34.360 Yeah. This these guys here. Unbelievable conversation about AI specifically. 0:00:34.840 --> 0:00:39.840 It is extraordinarily good on the aspect of what just 0:00:39.840 --> 0:00:44.120 happened with deep seek and actually competition around chips and 0:00:44.120 --> 0:00:48.800 AI and Nvidia and all that stuff. And basically how 0:00:48.800 --> 0:00:53.320 competition is happening between all the pinnacle makers, open versus 0:00:53.320 --> 0:00:57.040 closed source, basically everything. They're just doing so well in 0:00:57.080 --> 0:01:01.900 that conversation. So highly, highly recommend that we made the 0:01:01.900 --> 0:01:06.179 list of san's top security newsletters and podcasts. That was cool. 0:01:06.220 --> 0:01:08.820 Got a quick thing on how to use O1 and 0:01:08.819 --> 0:01:13.340 O3 on fabric? It's just a PP is basically the 0:01:13.340 --> 0:01:17.740 flag you want to use. You can't use s um, 0:01:17.780 --> 0:01:20.100 you have to use R, which is raw, because you 0:01:20.100 --> 0:01:24.259 cannot send a temperature to O1 or O3 where you 0:01:24.300 --> 0:01:27.900 can with most of the other models. And uh, after 0:01:27.900 --> 0:01:30.500 listening to that conversation I just told you about, I 0:01:30.500 --> 0:01:33.620 ended up tightening up my definition of AGI, which we're 0:01:33.620 --> 0:01:37.780 going to talk about in a second. But it's essentially, uh, 0:01:37.780 --> 0:01:39.580 we're going to go into more depth here in a second, 0:01:39.580 --> 0:01:42.300 but it's the ability for an AI, and that could 0:01:42.300 --> 0:01:45.060 be a model, a product or a system to perform 0:01:45.060 --> 0:01:51.300 the work of an average US based knowledge worker from 2022. And, uh, yeah, 0:01:51.300 --> 0:01:53.500 I talk about how this is the most debated term, 0:01:53.700 --> 0:01:55.660 and I want to spend a lot of time on it. 0:01:55.780 --> 0:02:00.330 This is in, um, raid, uh, real world AI definitions. 0:02:00.930 --> 0:02:03.450 And I've got two criteria for whether or not something 0:02:03.450 --> 0:02:08.290 is a good definition of AGI. Regular people should understand 0:02:08.290 --> 0:02:11.649 if we reached it or not. And attaining it should 0:02:11.650 --> 0:02:15.770 be significant to society. So who cares if nobody cares 0:02:15.770 --> 0:02:18.610 about it? And also, who cares if you can't figure 0:02:18.610 --> 0:02:22.050 out like what somebody is talking about? You know, they 0:02:22.050 --> 0:02:25.050 start using big words or something. I feel like that's 0:02:25.050 --> 0:02:28.210 not going to stick. It's not a useful definition. All right. Security, 0:02:28.250 --> 0:02:32.410 deep sky exposed customer data and unprotected database. So basically 0:02:32.410 --> 0:02:36.890 as soon as they launched and all this press starts happening, 0:02:36.889 --> 0:02:40.850 Nvidia's stock starts falling. Some people a lot of people 0:02:40.850 --> 0:02:45.050 and definitely some people, researchers at Wiz found an open 0:02:45.050 --> 0:02:50.010 database full of nasty stuff chat logs, API keys. It's 0:02:50.010 --> 0:02:52.730 in Chinese, so I'm not sure if there's any. This 0:02:52.730 --> 0:02:55.730 is kind of what I was wondering. Does it reveal 0:02:55.770 --> 0:02:59.430 any evidence that they used OpenAI for training? because that 0:02:59.430 --> 0:03:02.389 has been one of the things that people have been alleging, 0:03:02.630 --> 0:03:04.550 that the only reason they got so good is because 0:03:04.550 --> 0:03:08.430 they were using OpenAI to actually train on. And I'm 0:03:08.430 --> 0:03:13.230 curious if that's talked about in the logs or, um, or, 0:03:13.270 --> 0:03:16.470 you know, if there's any evidence of that happening inside 0:03:16.470 --> 0:03:19.550 of the thing inside of the leak. Cisa found that 0:03:19.550 --> 0:03:24.510 contact patient monitors have been secretly been sending patient data 0:03:24.510 --> 0:03:29.550 to China, and you can download and execute files remotely. Evidently, 0:03:29.550 --> 0:03:35.710 there's some sort of, uh, yeah, RCI type deal there. Yeah. Evidently, 0:03:36.510 --> 0:03:39.350 when you actually fixed it, it still didn't fix it. 0:03:39.350 --> 0:03:42.790 It just disabled the network interface, which the backdoor re-enables. 0:03:43.590 --> 0:03:46.390 So that was kind of a mess. Sonicwall says there's 0:03:46.390 --> 0:03:53.070 a massive exploit campaign against authentication bypass. And, uh, got 0:03:53.070 --> 0:03:56.510 some details here. It's 84. 43 is the port about 0:03:56.510 --> 0:04:00.540 2000 of these things vulnerable on showdown right now. Major 0:04:00.540 --> 0:04:06.100 hacking forums seized in international operations. Law enforcement took down 0:04:06.100 --> 0:04:09.660 some of the biggest hacking forums included Cracked and Nulled, 0:04:09.940 --> 0:04:13.220 which had over 10 million users combined. A new startup 0:04:13.220 --> 0:04:16.780 called backline raised $9 million to use AI agents that 0:04:16.779 --> 0:04:21.140 automatically fix vulnerabilities. This this is interesting, right? Everyone's talking 0:04:21.140 --> 0:04:25.580 about using AI to find vulnerabilities. What about fixing the vulnerabilities? 0:04:26.260 --> 0:04:31.860 So I'm really looking forward to vulnerability management remediation services 0:04:31.860 --> 0:04:34.860 that use AI. I personally think that this is going 0:04:34.860 --> 0:04:37.860 to require a whole lot of understanding of the company. 0:04:37.860 --> 0:04:41.420 It's going to require a lot of asset management. You're 0:04:41.420 --> 0:04:43.100 going to need to know people. You're going to need 0:04:43.100 --> 0:04:46.060 to know organizations. You're going to need to know how 0:04:46.060 --> 0:04:49.500 different engineering groups actually push code. So there's a lot 0:04:49.500 --> 0:04:51.620 of context you're going to have to get from something 0:04:51.620 --> 0:04:54.420 like an asset management system to be able to do this. 0:04:54.460 --> 0:04:57.520 Tulsi Gabbard is facing a lot of pushback back in 0:04:57.520 --> 0:05:01.960 her thing. I'm not sure if she's already been passed through. Uh, 0:05:02.000 --> 0:05:05.080 I'm trying not to pay attention to the news right now. 0:05:05.200 --> 0:05:09.360 It's a bit depressing, honestly. But, um. Yeah. Want to 0:05:09.360 --> 0:05:11.440 make a little bit of a statement here? I think 0:05:11.440 --> 0:05:14.120 if you dump top secret documents to the internet, this 0:05:14.120 --> 0:05:15.720 is one of the things she was being asked about 0:05:15.760 --> 0:05:19.560 was Snowden. Or if you break into the Capitol building 0:05:19.640 --> 0:05:24.840 on verification day, on results verification day, because you want 0:05:24.839 --> 0:05:29.279 to hopefully change the results, you are an actual criminal. 0:05:29.320 --> 0:05:31.440 That's that's what I think. At one point I saw 0:05:31.440 --> 0:05:35.359 Snowden as a whistleblower, too. I think he actually technically 0:05:35.360 --> 0:05:38.000 is one, and I think he might have been trying 0:05:38.000 --> 0:05:40.799 to do some good at some point, but I think 0:05:40.839 --> 0:05:42.880 it kind of got all messed up. And I haven't 0:05:42.880 --> 0:05:45.960 really been on that boat for many years now. Bedfordshire 0:05:46.000 --> 0:05:49.600 Police just became the first UK force to deploy Palantir's 0:05:49.640 --> 0:05:53.520 AI system, and they got some good results. 123 at 0:05:53.560 --> 0:05:57.470 risk kids were found in just one day argument that 0:05:57.470 --> 0:05:59.950 we're getting passkeys all wrong, and that they should be 0:05:59.950 --> 0:06:04.510 used alongside magic links, not as a complete replacement for 0:06:04.510 --> 0:06:08.110 other auth methods. I've been thinking about this a lot. 0:06:08.470 --> 0:06:12.589 I really do love Passkeys, but I do worry that 0:06:12.589 --> 0:06:15.390 it's I don't know, it feels like we might be 0:06:15.390 --> 0:06:17.910 being a bit sloppy with them. All right. This is 0:06:17.910 --> 0:06:21.029 why I think you should care about us reaching AGI. 0:06:21.070 --> 0:06:23.390 Wanted to say more about the AGI thing. I think 0:06:23.390 --> 0:06:27.190 it's the most important topic in AI. Actually, tons of 0:06:27.190 --> 0:06:32.590 very smart people don't know why they should care about it. Like, 0:06:32.589 --> 0:06:37.150 who cares if AGI hits this or that benchmark. And 0:06:37.150 --> 0:06:41.190 I think there's only one good reason why you should care. 0:06:41.190 --> 0:06:45.710 And here's how I talk through that. So think I 0:06:45.750 --> 0:06:50.030 coworkers right. Like actual coworkers that you have in person. 0:06:50.029 --> 0:06:52.349 But imagine your team at work. You've got like five 0:06:52.350 --> 0:06:56.409 coworkers or 20 or 35 whatever, however big your team is. 0:06:56.570 --> 0:07:01.409 Now imagine it's like 10,000 co-workers instead. Like overnight. So 0:07:01.410 --> 0:07:04.170 it was seven people. You show up and now it's 0:07:04.170 --> 0:07:08.490 a zoom call with 150 people or 10,000 people instead 0:07:08.490 --> 0:07:13.290 of seven. And you're watching them work. You're seeing them work. 0:07:13.890 --> 0:07:17.130 A couple of them, you see physically work you see 0:07:17.170 --> 0:07:20.930 them the results of their code, you see it come in. 0:07:20.930 --> 0:07:23.170 You get other signals telling you how good of a 0:07:23.170 --> 0:07:26.770 job they're doing or whatever, but you find out they're 0:07:26.770 --> 0:07:30.650 not perfect. They make mistakes just like anyone else. Someone 0:07:30.650 --> 0:07:33.210 still has to review their work. They still get lost. 0:07:33.210 --> 0:07:36.130 Sometimes they're like, what are we doing? Like, I'm not sure. 0:07:36.170 --> 0:07:37.770 You know, I'm not sure if I should be working 0:07:37.770 --> 0:07:40.610 on this project. Does this look correct? Or they mess 0:07:40.610 --> 0:07:44.050 up the code base or whatever. In some ways, they're way, 0:07:44.090 --> 0:07:48.210 way better than your human counterparts, and in some ways 0:07:48.210 --> 0:07:51.930 they're just way more stupid, right? But what they do 0:07:52.370 --> 0:07:56.350 is they make steady progress. They show up for video calls. 0:07:56.510 --> 0:08:00.390 They can read docs. They can read code. They're talking 0:08:00.430 --> 0:08:03.830 to you in slack. They're asking questions. When you ask 0:08:03.830 --> 0:08:06.270 them a question, they give you a response about the 0:08:06.270 --> 0:08:09.230 work that they're doing, or they help you find a document, whatever. 0:08:09.430 --> 0:08:13.630 And they can also readjust based on a coworker or 0:08:13.630 --> 0:08:15.710 their manager telling them, hey, that's not the way you 0:08:15.710 --> 0:08:17.710 should be doing this, or we got a new directive 0:08:17.710 --> 0:08:21.150 from above. You know, the goal is slightly shifted or whatever. 0:08:21.230 --> 0:08:23.390 They can change what they're doing. But the key here 0:08:23.390 --> 0:08:28.310 is there's actually 10,000 of these people, right? There's 10,000 0:08:28.350 --> 0:08:32.230 of these agents instead of ten, or there's 100,000 of 0:08:32.230 --> 0:08:36.110 these instead of your team of 100, and they work 0:08:36.110 --> 0:08:38.830 24 over seven. They put in the work, they put 0:08:38.830 --> 0:08:42.110 in the effort, and they do not get tired. And 0:08:42.110 --> 0:08:45.069 they also constantly improve. So if they start off with 0:08:45.070 --> 0:08:49.030 like 105 IQ, which is, you know, decently smart or 0:08:49.030 --> 0:08:52.830 two years of knowledge in security or programming or intern 0:08:52.830 --> 0:08:55.740 level or whatever. After a year or two, they've had 0:08:55.780 --> 0:08:59.700 like 20 updates, which were silent updates or whatever. And 0:08:59.700 --> 0:09:03.220 assuming none of those updates went horribly wrong, uh, now 0:09:03.220 --> 0:09:07.100 they got 110 IQ. Now they got 120 IQ. Right. 0:09:07.140 --> 0:09:12.580 Or whatever. The AGI alternative to uh, or analog to 0:09:12.620 --> 0:09:15.460 IQ is. It's not quite IQ, but call it raw 0:09:15.460 --> 0:09:19.540 fluid intelligence. Right. So they're getting those constant updates, but 0:09:19.540 --> 0:09:22.220 they're also learning to be better coders, not just by 0:09:22.220 --> 0:09:24.820 the work that they're doing, but actually by just better 0:09:24.820 --> 0:09:28.980 and better models. So now instead of 105 IQ with 0:09:28.980 --> 0:09:34.939 two years work experience, now it's 120 IQ with 20 0:09:34.980 --> 0:09:38.780 years of experience. So it's like a staff engineer, but 0:09:38.780 --> 0:09:42.620 you still have 100,000 of them, right? This is why, 0:09:43.340 --> 0:09:45.860 in my opinion, AGI is a big deal. And I 0:09:45.860 --> 0:09:48.740 think we're getting really close to this. I've got another 0:09:48.900 --> 0:09:50.940 thing to read about. This is what I just posted 0:09:50.940 --> 0:09:54.280 earlier on one of the Socials, but I think we're 0:09:54.280 --> 0:09:57.960 getting close to this. It's not one component, right? This 0:09:57.960 --> 0:10:00.000 is not like one model that's going to release and 0:10:00.000 --> 0:10:02.800 do this. It will be a system. It will behave 0:10:02.800 --> 0:10:05.480 like one person or one thing or whatever you want 0:10:05.480 --> 0:10:08.800 to call it, but it'll be this composite that lets 0:10:08.800 --> 0:10:10.960 it behave in a cohesive way, which is actually the 0:10:10.960 --> 0:10:14.400 way our brains work as well. My guess was in 0:10:14.400 --> 0:10:17.600 2023 that it was going to be 25 to 28. 0:10:17.720 --> 0:10:19.360 That was my thing. And I put that in like 0:10:19.360 --> 0:10:23.280 two years ago. Okay. Like March of 23, I think 0:10:23.280 --> 0:10:26.959 I made this prediction. 25 to 28 is when we're 0:10:26.960 --> 0:10:30.199 going to have this. And my definition for this is 0:10:30.240 --> 0:10:35.400 basically somebody who can replace a human knowledge worker, a 0:10:35.400 --> 0:10:40.320 decent human knowledge worker working at some remote job. Right. 0:10:40.360 --> 0:10:42.160 So I don't think they're going to be able to like, 0:10:42.200 --> 0:10:45.560 bring you a coffee because that's robotics. But working a 0:10:45.559 --> 0:10:53.750 remote job, pushing code, doing organization, project management. Um, editing documents, 0:10:54.590 --> 0:10:58.230 that kind of stuff. 25 to 28. I think that's 0:10:58.270 --> 0:11:02.550 what is when we're going to actually hit this. So 0:11:02.550 --> 0:11:06.630 my current estimate for this is late 25 or maybe 0:11:06.630 --> 0:11:10.550 sometime in 26, and it could actually go into 27. 0:11:10.550 --> 0:11:13.750 Because this type of progress that we're seeing in AI, 0:11:13.750 --> 0:11:17.350 it's big jumps. But then it kind of plateaus for 0:11:17.550 --> 0:11:20.750 weeks or months. But that plateau could actually go for 0:11:20.750 --> 0:11:23.030 a year. I don't think it will. I think it's 0:11:23.030 --> 0:11:26.390 going to continue to jump kind of violently, but it 0:11:26.390 --> 0:11:31.869 could plateau out and then maybe it turns into more 2728. 0:11:32.630 --> 0:11:35.990 I think the chances that it happens after 29 are 0:11:35.990 --> 0:11:40.590 really low, maybe 5%, which is a lot. It's a lot, right? 0:11:40.630 --> 0:11:46.870 So my guarantee for this right now, my high, it's like, uh, 0:11:46.950 --> 0:11:52.210 what's the CIA, CIA range? I think the CIA range is, Um, 0:11:52.250 --> 0:11:56.929 almost guaranteed or something like that. It's, um, extremely probable. 0:11:56.929 --> 0:12:00.330 It's one of those. But it's only like 95%. And 0:12:00.330 --> 0:12:03.170 that's where I'm at because I'm not really 100% on anything. 0:12:03.290 --> 0:12:04.969 And when this does happen, I think it'll be the 0:12:04.970 --> 0:12:09.010 single biggest impact on humanity from tech, by far bigger 0:12:09.010 --> 0:12:13.770 than the internet and both in the negative and positive directions. 0:12:13.770 --> 0:12:16.250 So that's why I think you should care about AGI. 0:12:16.410 --> 0:12:20.490 And in perfect timing, MuleSoft reports that almost all IT 0:12:20.530 --> 0:12:24.690 leaders are planning to use autonomous AI agents within two years. 0:12:24.690 --> 0:12:28.930 That's 93% and about half are already doing so. Sam 0:12:28.929 --> 0:12:31.450 Altman basically says he feels like he was on the 0:12:31.450 --> 0:12:34.730 wrong side of history with open source. He didn't say 0:12:34.730 --> 0:12:37.650 he's going to do anything about it or what he's 0:12:37.650 --> 0:12:39.650 going to do about it. It seems like kind of 0:12:39.690 --> 0:12:42.650 obvious that he will do something. My guess is they 0:12:42.650 --> 0:12:47.130 will release some nerfed versions of some of their models 0:12:47.130 --> 0:12:49.450 just to be like, look, we are playing nice with 0:12:49.450 --> 0:12:52.429 the open source model, but I don't think they're going 0:12:52.470 --> 0:12:55.750 to completely pivot to like what meta is doing, for example. 0:12:55.910 --> 0:12:59.710 And like I said, yeah. OpenAI claims Chinese rival deep 0:12:59.750 --> 0:13:03.190 tech stole training data. Yeah, I don't actually think it's 0:13:03.190 --> 0:13:06.110 kind of funny to be like, oh, OpenAI is worried 0:13:06.110 --> 0:13:08.830 about their stuff being stolen when, you know, they stole 0:13:08.830 --> 0:13:12.670 the whole internet to train on. I think. I mean, 0:13:12.830 --> 0:13:14.630 anything you put on the internet is just going to 0:13:14.630 --> 0:13:17.390 be on the internet, and that's free game. I think, 0:13:17.550 --> 0:13:20.829 I think unless you have specific like licenses or something, 0:13:21.030 --> 0:13:23.589 and obviously I don't want people to steal like paid 0:13:23.590 --> 0:13:28.150 content that I'm producing. Right. But but I don't release 0:13:28.270 --> 0:13:33.190 paid content live to the public, on social media or 0:13:33.190 --> 0:13:36.110 on a blog post, and then get mad when someone 0:13:36.110 --> 0:13:39.110 scrapes it, right? I consider all the stuff that I 0:13:39.110 --> 0:13:42.069 put out since like 1999 or whatever, how long I've 0:13:42.070 --> 0:13:44.230 been on the internet. You gave it to the world 0:13:44.230 --> 0:13:46.790 when you posted it online. That's the way I see it. 0:13:46.790 --> 0:13:51.100 I always have R0 or R1 zero Shows AI reasoning 0:13:51.100 --> 0:13:54.380 without human training. This is really cool because like I 0:13:54.380 --> 0:13:57.940 talk about here, it's like what happened in chess. So 0:13:57.980 --> 0:14:01.660 at first they used they built a chess computer. Um, 0:14:01.660 --> 0:14:04.340 I think this was Google that was doing this, and 0:14:04.340 --> 0:14:07.179 it just got really good at learning the rules, and 0:14:07.179 --> 0:14:10.260 it watched human players and it improved on that. And 0:14:10.300 --> 0:14:14.020 of course, it could beat previous chess computers like Deep 0:14:14.020 --> 0:14:16.940 Blue is nothing compared to these computers. They just crush it. 0:14:17.220 --> 0:14:21.500 What the big breakthrough was with AlphaZero and also with 0:14:21.500 --> 0:14:27.540 R1 zero is it's learning more actually, or at least 0:14:28.140 --> 0:14:34.340 significantly more. It's learning significantly more from the reinforcement learning 0:14:35.140 --> 0:14:38.380 from the world. It's learning from the world as opposed 0:14:38.380 --> 0:14:41.980 to from the rules or watching other people interact with 0:14:41.980 --> 0:14:45.660 the world. So it's just an accelerator because it could 0:14:45.660 --> 0:14:48.620 do that by itself, right? Without a lot of effort 0:14:48.620 --> 0:14:52.320 from other people and that's really, really powerful. A new 0:14:52.320 --> 0:14:55.640 study shows deepfakes AI model refuses to answer the vast 0:14:55.640 --> 0:15:00.240 majority of sensitive questions about China. Shocking, shocking. This is 0:15:00.240 --> 0:15:03.240 why it matters where you get your models right. Because 0:15:03.440 --> 0:15:06.080 according to them, Tiananmen Square didn't happen. It's all a 0:15:06.080 --> 0:15:10.320 breakdown of how to properly evaluate ragtag systems and llms 0:15:10.320 --> 0:15:16.560 in Practice by Salman Khan and Andrej Karpathy talks about 0:15:16.680 --> 0:15:19.480 vibe coding, where you basically get in a flow state 0:15:19.640 --> 0:15:22.200 and code like you're playing an instrument. It says the 0:15:22.200 --> 0:15:25.600 key is to stop overthinking and just think and respond 0:15:25.600 --> 0:15:28.000 and let the AI do most of the work. I 0:15:28.000 --> 0:15:30.840 think this is spot on and a lot of people 0:15:30.840 --> 0:15:33.080 are really mad. They're like, oh, I can't believe you know, 0:15:33.080 --> 0:15:35.000 you're a hard core coder and you're just going to 0:15:35.000 --> 0:15:38.440 let AI code for you. And again, he doesn't care. 0:15:38.520 --> 0:15:41.800 I don't care. People like us, we don't care because 0:15:42.160 --> 0:15:45.360 we could see what's actually coming, which is pretty soon. 0:15:45.360 --> 0:15:47.960 You don't fire up an IDE at all. What you 0:15:47.960 --> 0:15:51.390 do is you fire up a voice conversation and the 0:15:51.390 --> 0:15:55.870 thing starts showing you what the product looks like on 0:15:55.870 --> 0:15:59.230 one side, and it's showing you roughly like the code 0:15:59.230 --> 0:16:01.910 or the diagram or whatever it is on the left, 0:16:02.550 --> 0:16:05.630 and you're having a verbal back and forth like that 0:16:05.630 --> 0:16:09.390 is coding within, you know, 1 to 3 years. So 0:16:09.390 --> 0:16:11.630 this whole thing of like, oh, you're not doing the 0:16:11.630 --> 0:16:14.510 real thing unless you're doing the real coding. Once again, 0:16:14.510 --> 0:16:17.350 you have to take that all the way back. Um, 0:16:17.350 --> 0:16:20.870 are you writing assembler by hand? No you're not. Okay, well, 0:16:21.550 --> 0:16:24.470 you're cheating by using Python or whatever, right? So it 0:16:24.470 --> 0:16:27.590 just keeps going farther and farther away, and eventually it's 0:16:27.590 --> 0:16:29.190 going to be like, I wish I had a product 0:16:29.190 --> 0:16:32.830 that looked like this, blah, blah, blah, and boom, it 0:16:32.830 --> 0:16:37.030 just goes into the world. And what did you really contribute? 0:16:37.030 --> 0:16:41.270 You contributed the raw base idea. But at some point, 0:16:41.270 --> 0:16:43.790 the A's are good enough to contribute that as well. 0:16:43.790 --> 0:16:47.070 And now you're competing with them on that. And that's 0:16:47.070 --> 0:16:50.290 when it turns a little bit sad. Apple quietly added 0:16:50.290 --> 0:16:54.530 Starlink satellite support to iPhones through a software update. And 0:16:54.530 --> 0:16:58.410 this is a partnership with SpaceX and T-Mobile. So yeah, 0:16:58.450 --> 0:17:01.130 SpaceX is how your iPhone is able to talk to 0:17:01.130 --> 0:17:05.570 satellites when you have no signal. Humans massive new survey 0:17:06.010 --> 0:17:11.649 revealed that 87% of astrobiologists think that extraterrestrial life exists 0:17:11.650 --> 0:17:15.970 somewhere in the universe. 87%. I am one of those people, 0:17:15.970 --> 0:17:18.090 not one of the astrobiologists, but one of the people 0:17:18.090 --> 0:17:20.689 who agrees with that study. A Montana shows that drones 0:17:20.690 --> 0:17:24.330 are way better at keeping grizzlies away from humans, so 0:17:24.330 --> 0:17:27.250 they basically fly up to them and chase them and 0:17:27.250 --> 0:17:32.169 scare them. Could it be cool to take drones with 0:17:32.170 --> 0:17:35.210 you when you go into the forest? So you're worried 0:17:35.210 --> 0:17:38.369 about bears? You release like three drones or whatever. That's 0:17:38.369 --> 0:17:41.250 part of your panic response. You hit a button. It 0:17:41.250 --> 0:17:44.490 makes a really loud sound, like some sort of horn 0:17:44.490 --> 0:17:47.280 or something, and you could point the horn at the thing, 0:17:47.680 --> 0:17:50.960 but it also launches the drones and the drones go 0:17:50.960 --> 0:17:55.119 over and maybe they make noises. Maybe they have sounds, whatever, 0:17:55.240 --> 0:17:59.359 but they just start buzzing around the the bear. And maybe, like, 0:17:59.400 --> 0:18:02.359 you could sell that as a package is like bear defense. 0:18:02.440 --> 0:18:05.680 Maybe they shoot, uh, bear spray as well. That would 0:18:05.680 --> 0:18:09.200 be cool. All right. Swerving Broncos I was thinking about well, 0:18:09.200 --> 0:18:12.120 I was basically driving on the 101 at like 11:00 0:18:12.119 --> 0:18:15.520 at night or something just a couple days ago. And 0:18:16.359 --> 0:18:18.600 I am watching the cars on the, on the road 0:18:18.600 --> 0:18:22.320 in front of me. And they're just like, drifting, like 0:18:22.800 --> 0:18:25.600 right over the line. The second wheel goes over the line. 0:18:25.600 --> 0:18:29.600 They're just like drifting. Obvious that they're on their phones 0:18:29.600 --> 0:18:32.600 or drunk or high. And then like, I get away 0:18:32.600 --> 0:18:35.520 from that thing or it turns off the highway or whatever. 0:18:35.520 --> 0:18:38.240 And like, here comes another giant truck, a giant like 0:18:38.240 --> 0:18:40.840 Cadillac or a Bronco or something like that comes next 0:18:40.840 --> 0:18:43.560 to me, and it just starts drifting into my lane. 0:18:43.560 --> 0:18:46.300 And I'm like, what are you doing? And I And 0:18:46.300 --> 0:18:48.220 I look at them, they're like looking down. They're not 0:18:48.220 --> 0:18:51.620 even paying attention. And I see a third and this 0:18:51.619 --> 0:18:54.659 is within like 30 miles or something. I see a 0:18:54.660 --> 0:18:57.700 third one. They're just like drifting over. There's no cops around. 0:18:57.700 --> 0:19:01.260 I hardly ever see cops anywhere. Like on the on 0:19:01.260 --> 0:19:05.180 the highways anymore. There's no cops. There's no one stopping them. 0:19:05.180 --> 0:19:08.180 They're just kind of like driving like crazy people. And 0:19:08.180 --> 0:19:12.820 I'm thinking, you know what won't do? This is Waymo. 0:19:12.859 --> 0:19:17.060 Waymo won't do this either. Will autonomous Tesla or an 0:19:17.060 --> 0:19:21.340 autonomous like BYD device from China or whatever. Say what 0:19:21.340 --> 0:19:26.220 you want about AI driving. They bicyclists love them because 0:19:26.220 --> 0:19:29.500 they never crash into them. I mean, I think we 0:19:29.500 --> 0:19:33.460 just have to realize how bad human drivers are before 0:19:33.580 --> 0:19:37.980 we start criticizing how bad I drivers are. AI's big jump. 0:19:38.180 --> 0:19:43.020 This one is very simple. I think AI from Apple 0:19:43.020 --> 0:19:46.040 is like it's been the worst for a while and 0:19:46.040 --> 0:19:48.280 it's about to be the best. That's the bottom line. 0:19:48.480 --> 0:19:54.200 They are putting out stuff in 18.4 that basically finally 0:19:54.200 --> 0:19:58.840 connects Siri to the context inside the phone. So your phone, 0:19:58.840 --> 0:20:01.720 you have your exercise, you have your heartbeat, you have 0:20:01.720 --> 0:20:06.040 your appointments and your calendar and your email. And it 0:20:06.040 --> 0:20:09.680 could see all the different stuff. It could see your life. Right. 0:20:09.680 --> 0:20:13.480 But it's never unified that stuff. Well, Apple just got really, 0:20:13.480 --> 0:20:16.760 really serious about Siri or they're in the process of 0:20:16.760 --> 0:20:19.600 doing that. They got a new leader over it. Like 0:20:19.600 --> 0:20:23.399 it's just become super high priority. They're doing that at 0:20:23.400 --> 0:20:26.560 the same time that they're doing the cloud secure enclave. 0:20:26.800 --> 0:20:29.440 At the same time, they're doing the partnership with ChatGPT 0:20:30.000 --> 0:20:33.320 at the same time that they're now giving Siri access 0:20:33.320 --> 0:20:36.560 to all that context that they have, and access to 0:20:36.960 --> 0:20:41.240 call apps and pass data between the different apps. So 0:20:41.240 --> 0:20:44.439 they've been setting up this infrastructure for AI for the 0:20:44.440 --> 0:20:46.870 last couple of years. They've built all this stuff. They've 0:20:46.869 --> 0:20:50.430 spent all this R&D. And again, I used to work there. 0:20:50.470 --> 0:20:52.510 This has nothing to do with anything. I've been gone 0:20:52.510 --> 0:20:56.190 for a while. And also, I wouldn't be disclosing anything 0:20:56.190 --> 0:20:58.990 if I actually worked on a project like that. But 0:20:58.990 --> 0:21:03.189 I've been watching them since, whatever, 2007. And I'm telling 0:21:03.190 --> 0:21:06.229 you that they are about to switch this thing on. 0:21:06.830 --> 0:21:09.389 And I really do think that they're about to have 0:21:09.390 --> 0:21:15.430 the best personal ecosystem integration AI type play. And it's 0:21:15.430 --> 0:21:18.310 going to go from being like a are they even 0:21:18.310 --> 0:21:21.430 doing anything to Holy crap, this is what they've been 0:21:21.430 --> 0:21:25.149 working on. That is my prediction around that. And I 0:21:25.190 --> 0:21:27.790 novels are coming. I actually have a YouTube video about this. 0:21:27.910 --> 0:21:30.550 But yeah, I'll wait till that comes out. It might 0:21:30.590 --> 0:21:33.109 actually be out soon, but yeah. Anyway, I'm going to 0:21:33.109 --> 0:21:35.629 wait to talk about this one. Basically, it's going to 0:21:35.630 --> 0:21:39.990 be possible to make a complete decent novel using AI. 0:21:40.030 --> 0:21:43.790 I think within months or years. I've already got a 0:21:43.790 --> 0:21:45.689 got a couple of emails actually, where they're like, what 0:21:45.690 --> 0:21:47.330 are you talking about? I've been doing that for a 0:21:47.330 --> 0:21:51.410 year and a half. So I think it was possible 0:21:51.410 --> 0:21:53.330 to do a year and a half ago, and I've 0:21:53.330 --> 0:21:58.410 messed around with some, uh, longer, uh, like, fiction stories, 0:21:58.530 --> 0:22:01.890 and it was halfway decent. And that was like a year, 0:22:01.930 --> 0:22:05.050 year and a half ago. But it was difficult. What 0:22:05.050 --> 0:22:07.050 I'm trying to say is it's going to get easy 0:22:07.250 --> 0:22:09.850 and lots of people are going to do it. Um, 0:22:09.890 --> 0:22:12.889 ChatGPT new tasks feature. A lot of people are talking 0:22:12.890 --> 0:22:17.609 about this. It's basically scheduled and automated, um, agents to 0:22:17.650 --> 0:22:20.129 go and do things for you. I have not messed 0:22:20.130 --> 0:22:23.409 with this one, but it looks really, really exciting. Really 0:22:23.410 --> 0:22:27.330 clever way to use UV which is a fast Python 0:22:27.330 --> 0:22:30.690 package manager. Look at this. You put this at the 0:22:30.690 --> 0:22:33.290 top of oh man, I haven't messed with this yet. 0:22:33.290 --> 0:22:35.490 I have to write this down. I ran it once, 0:22:35.490 --> 0:22:38.609 but I haven't put it in like all my main scripts. 0:22:39.369 --> 0:22:43.600 This allows you to run anything that has multiple dependencies 0:22:43.600 --> 0:22:49.800 and normally breaks, and it will automatically instantiate you a 0:22:49.800 --> 0:22:53.800 working environment to make your script run. Isn't that just beautiful? 0:22:53.840 --> 0:22:57.719 It's absolutely beautiful. I mean, this is so much better 0:22:57.720 --> 0:23:01.679 than regular Pip, because if you have multiple environments like 0:23:02.040 --> 0:23:04.950 that's my problem. My regular Python has like 3 or 0:23:04.950 --> 0:23:10.439 4 different environments, different virtual environments, like nested spread all 0:23:10.440 --> 0:23:14.040 over the file system. It's like a nightmare. Um, so 0:23:14.040 --> 0:23:18.080 that's why I use UV now. And this is just. Yeah, 0:23:18.119 --> 0:23:20.640 this is really great. I haven't seen a downside yet, 0:23:20.640 --> 0:23:22.840 but I haven't messed with it that much. Uh, cool. 0:23:22.840 --> 0:23:28.120 Developer created Napthine's malicious software that traps aggressive AI web 0:23:28.119 --> 0:23:31.119 crawlers in infinite loops and feeds them garbage data to 0:23:31.160 --> 0:23:35.560 poison their models. Smiling, not smiling OpenAI dropped new ChatGPT 0:23:35.560 --> 0:23:39.240 capability called deep Research. Supposedly you only get like 100 0:23:39.280 --> 0:23:42.320 a month because they're really expensive and it takes like 0:23:42.320 --> 0:23:46.260 takes like ten minutes to finish, but I've heard it's 0:23:46.260 --> 0:23:51.300 created some serious, serious research. Really good papers. Haven't checked 0:23:51.300 --> 0:23:53.820 if I have access to that API yet. All right. 0:23:53.859 --> 0:23:58.979 Justin McGurk explores how William Gibson novels perfectly capture our 0:23:58.980 --> 0:24:03.020 obsession with commodifying everything unique and authentic until it loses 0:24:03.020 --> 0:24:06.739 all meaning. This is why I'm doing human 3.0. I 0:24:06.740 --> 0:24:10.660 do not want to be replaced by AI and technology. 0:24:10.780 --> 0:24:15.380 No thank you. I want it to enhance humanity. Found 0:24:15.380 --> 0:24:19.380 a really clean python script that lets you download YouTube videos. 0:24:19.420 --> 0:24:22.900 This thing works really well, I just worried. I'm just 0:24:22.900 --> 0:24:25.900 worried it's going to get blocked soon. Recommendation of the week. 0:24:25.900 --> 0:24:28.620 If you ever get overwhelmed by what all this AI 0:24:28.660 --> 0:24:31.620 stuff even means, or you want to explain to someone else, 0:24:31.740 --> 0:24:34.020 I got a little quote here. So within the next 0:24:34.020 --> 0:24:37.580 few years we might have something called AGI where I 0:24:37.619 --> 0:24:40.620 can work as a full knowledge worker, like joining the 0:24:40.619 --> 0:24:46.490 onboarding cohorts, reading documentation, participating on slack, submitting code, adjusting 0:24:46.490 --> 0:24:49.130 their work based on the work of others. But instead 0:24:49.130 --> 0:24:52.810 of 2 or 5 of these, imagine hundreds of them 0:24:52.810 --> 0:24:56.090 for the cost of one human employee. I think that's 0:24:56.090 --> 0:24:59.730 pretty cool way to kind of explain the possible impact 0:24:59.850 --> 0:25:02.850 in a way that, like regular people who don't pay 0:25:02.850 --> 0:25:06.050 attention to this all the time could understand. And the 0:25:06.050 --> 0:25:09.330 aphorism of the week, some books leave us free and 0:25:09.330 --> 0:25:12.810 some books make us free, some books leave us free, 0:25:13.250 --> 0:25:18.449 and some books make us free. Ralph Waldo Emerson. Unsupervised 0:25:18.450 --> 0:25:21.770 learning is produced on Hindenburg Pro using an SM seven 0:25:21.810 --> 0:25:25.450 B microphone. A video version of the podcast is available 0:25:25.450 --> 0:25:29.090 on the Unsupervised Learning YouTube channel, and the text version 0:25:29.090 --> 0:25:34.210 with full links and notes is available at Daniel Miessler newsletter. 0:25:34.850 --> 0:25:35.810 We'll see you next time.