WEBVTT - UL NO. 451: Altman Says ASI in "Thousands of Days" 0:00:01.129 --> 0:00:04.460 Welcome to Unsupervised Learning, a security, AI and meaning focused 0:00:04.460 --> 0:00:07.310 podcast that looks at how best to thrive as humans 0:00:07.310 --> 0:00:11.540 in a post AI world. It combines original ideas, analysis, 0:00:11.570 --> 0:00:14.780 and mental models to bring not just the news, but 0:00:14.780 --> 0:00:22.400 why it matters and how to respond. All right, welcome 0:00:22.400 --> 0:00:25.130 to unsupervised learning. This is Daniel Miessler. Okay. First big 0:00:25.130 --> 0:00:30.860 news that just happened is there was a release for Gemini. 0:00:30.890 --> 0:00:34.700 So there was rumor last night or, I don't know, 0:00:34.729 --> 0:00:36.980 last few days that there was going to be a 0:00:36.979 --> 0:00:39.830 big model that got released. And it turns out it 0:00:39.830 --> 0:00:43.460 was Google as opposed to anthropic, although maybe anthropic is 0:00:43.490 --> 0:00:45.200 going to do something as well. I was hoping for 0:00:45.229 --> 0:00:47.839 like an opus four or something, or at least an 0:00:47.840 --> 0:00:51.229 opus 3.5. So it turns out to have been Google 0:00:51.229 --> 0:00:55.790 and it looks like new Gemini 1.5 Pro and Flash models. 0:00:55.790 --> 0:00:59.990 So right off the bat, I mean, these seem very minor. 0:01:00.020 --> 0:01:04.339 They're minor updates to 1.5 Pro, but whatever. I'm looking 0:01:04.340 --> 0:01:10.069 for reasons to be excited. 52% reduction in output token pricing, 0:01:10.100 --> 0:01:16.760 two times higher rate limits for Gemini 1.5 flash and 0:01:16.760 --> 0:01:20.720 three times higher rate limits for Gemini 1.5 Pro. So 0:01:20.720 --> 0:01:24.050 these are really updates to 1.5 Pro. And if I 0:01:24.050 --> 0:01:27.500 go to let's see here, if I go to Kitty 0:01:27.500 --> 0:01:32.120 and I do this yeah I'm not seeing 002. Oh 0:01:32.120 --> 0:01:35.300 there we go. Yeah 20. Okay. So if I put 0:01:35.330 --> 0:01:40.370 20in here into fabric I'm now using Gemini 1.5 Pro 0:01:40.400 --> 0:01:43.100 version two, which is the brand new one that just 0:01:43.100 --> 0:01:46.880 came out. So I guess we could do like a classic. Um, 0:01:46.910 --> 0:01:51.740 so this is the rivets video pipe to fabric SP 0:01:52.670 --> 0:01:58.340 extract wisdom DM all right, so streaming works. Seems pretty fast. 0:01:58.350 --> 0:02:01.020 See how these insights are? So first of all, it 0:02:01.050 --> 0:02:04.500 extracted a whole lot of insights. That's or no that 0:02:04.500 --> 0:02:08.400 was ideas. Now insights should be fewer. And these look 0:02:08.400 --> 0:02:10.770 really really good. So here's one thing that I love 0:02:10.770 --> 0:02:13.980 about AI right now. So if I give this same 0:02:13.980 --> 0:02:19.110 challenge to Gemini and GPT four or any of the 0:02:19.110 --> 0:02:23.520 modern ones from OpenAI or anthropic, they all come up 0:02:23.520 --> 0:02:26.370 with this one as the top quote. You can't necessarily 0:02:26.370 --> 0:02:28.620 think yourself into the answers. You have to create space 0:02:28.620 --> 0:02:30.840 for the answers to come to you. And I think 0:02:30.840 --> 0:02:33.450 that is really, really cool. And all these other quotes 0:02:33.450 --> 0:02:35.609 are just fantastic. This is like one of the videos 0:02:35.610 --> 0:02:37.350 I know the best. It's like one of my favorite 0:02:37.350 --> 0:02:42.120 videos ever. So anyway, this is the results of the 0:02:42.120 --> 0:02:47.280 capture from 1.500002, which is the brand new one. And 0:02:47.280 --> 0:02:49.830 let's look at the habits. First of all, it followed 0:02:49.830 --> 0:02:52.769 instructions really well. And by the way, the DM version 0:02:52.770 --> 0:02:55.860 of Extract Wisdom has a whole bunch of extra instructions. 0:02:55.889 --> 0:02:59.460 It's like way more chain of thought and way more 0:02:59.460 --> 0:03:02.460 prescriptive for like what I'm looking for. So that should 0:03:02.460 --> 0:03:07.079 give a pretty good advantage. And it looks like Google 0:03:07.080 --> 0:03:10.020 did really well with it. I like the way that 0:03:10.020 --> 0:03:12.180 they're saying this. Oh, you know what I want to 0:03:12.180 --> 0:03:15.060 do my other new favourite one. So let's take that 0:03:15.060 --> 0:03:20.460 one and let's do create story explanation I love this 0:03:20.460 --> 0:03:25.530 create story explanation one. Absolutely fantastic. So it basically just 0:03:25.530 --> 0:03:29.100 walks you through like you're in a conversation about whatever 0:03:29.100 --> 0:03:32.579 the input is. So they lament the decline of creativity 0:03:32.580 --> 0:03:35.430 and deep thought. In the West. They discuss the impact 0:03:35.430 --> 0:03:39.180 of technology, social media and constant barrage of information. One 0:03:39.180 --> 0:03:41.760 speaker shares her experience of deep diving into the history 0:03:41.760 --> 0:03:45.450 of the New Testament manuscripts. They discuss the importance of 0:03:45.450 --> 0:03:51.089 deep reading, embracing diverse interests. So yeah, the conversation concludes 0:03:51.090 --> 0:03:53.790 with a powerful image of the decline of the creative spirit. 0:03:53.820 --> 0:03:56.500 I mean, this is just a cool way of describing 0:03:56.500 --> 0:03:58.930 a piece of content. So I think I could do 0:03:58.960 --> 0:04:01.960 like another one. Neri Oxman so this is one of 0:04:01.960 --> 0:04:04.570 my other favorite YouTube videos ever. Yeah, look at this. 0:04:04.600 --> 0:04:08.559 Oxman believes nature possesses a wisdom beyond human intelligence. She's 0:04:08.560 --> 0:04:13.360 building a company called Oxman to explore the wisdom and 0:04:13.360 --> 0:04:16.690 augment nature with technology. Her goal is to grow products, 0:04:16.690 --> 0:04:19.539 blurring the line between made and grown. See? Check this 0:04:19.540 --> 0:04:23.230 out when I go to summarize why I love this 0:04:23.230 --> 0:04:25.510 video so much, it's hard to capture it. It's hard 0:04:25.510 --> 0:04:29.110 to remember what I love about it so much, and 0:04:29.110 --> 0:04:32.620 this does really well at that. The tech does really 0:04:32.620 --> 0:04:35.860 well at capturing what makes her so awesome and it's 0:04:35.860 --> 0:04:39.070 got advice. It's just like great. In fact, I might 0:04:39.070 --> 0:04:42.520 even get rid of this pre and post sentence. I 0:04:42.520 --> 0:04:45.609 think I might just keep that content right there. But 0:04:45.610 --> 0:04:49.240 I'm not going to do that live. But anyway. Yeah interesting. 0:04:49.270 --> 0:04:51.010 Looks like I have a little bit of a key showing. 0:04:51.010 --> 0:04:53.650 But that is not a real key. That is a 0:04:53.650 --> 0:04:58.900 HTTP Cookie header. Always like to watch my OpSec here. 0:04:58.930 --> 0:05:00.760 Going to leave a token at some point. I'm going 0:05:00.790 --> 0:05:02.979 to leave a token to see if I'm going to 0:05:02.980 --> 0:05:06.789 leave an API key out there and and monitor to 0:05:06.820 --> 0:05:10.930 see if anyone comes in on it. Exercise for the reader. Okay. 0:05:10.930 --> 0:05:14.650 So cool. So yeah, we don't have the video of 0:05:14.650 --> 0:05:17.080 Parul and Sam Altman yet. It looks like he has 0:05:17.080 --> 0:05:20.229 not dropped it yet, but that's fine. Just go back 0:05:20.230 --> 0:05:24.760 to the show. Basically some model upgrades from Gemini and 0:05:24.760 --> 0:05:27.609 they look to work quite nicely. And I just changed 0:05:27.610 --> 0:05:30.910 my default model to it just now. So I'll see 0:05:30.910 --> 0:05:37.390 how that does performance wise. Okay. Um, Thomas Rocca created 0:05:37.390 --> 0:05:40.299 a web GUI for fabric called fabric UI. Check this 0:05:40.300 --> 0:05:44.770 thing out. Look at that. And Jonathan, who's on the 0:05:44.770 --> 0:05:47.650 fabric team, is actually working on something like this as well. 0:05:47.680 --> 0:05:49.420 But look at this. You can select the model. You 0:05:49.420 --> 0:05:51.730 want to use a pretty good selection there. You put 0:05:51.730 --> 0:05:53.870 it in your API key. The guy's a hard core 0:05:53.870 --> 0:05:55.790 security guy, so I don't think he's going to be 0:05:55.790 --> 0:05:59.030 stealing keys anytime soon. Um, and then, look, you pick 0:05:59.029 --> 0:06:02.570 your pattern here, and. Yeah, you got a guy to 0:06:02.600 --> 0:06:07.340 run fabric with. And again, Jonathan on the fabric team 0:06:07.339 --> 0:06:10.220 is actually building something similar. He's building like an API 0:06:10.250 --> 0:06:14.270 server and also guy. So that's going to be really cool. 0:06:14.420 --> 0:06:19.820 But this guy, uh, Thomas Rocha. Super cool guy. I've 0:06:19.820 --> 0:06:22.400 known him for years and years. Uh, he's part of 0:06:22.400 --> 0:06:25.940 the UL community and, uh, evidently just built a UI 0:06:25.970 --> 0:06:28.400 for fabric. So that was very cool. And I just 0:06:28.400 --> 0:06:30.560 spent a whole bunch of time, like, deep cleaning on 0:06:30.560 --> 0:06:34.729 my phone, which is the new iPhone 16 Pro and deleted, 0:06:34.730 --> 0:06:40.489 probably 40 applications, phone screen cleanup, widgets, Refactors watch faces. 0:06:40.520 --> 0:06:43.849 I redid all my focus modes, which are these things 0:06:43.850 --> 0:06:47.330 up here and uh, like right here. So you got 0:06:47.360 --> 0:06:50.150 like work and personal and Do Not Disturb and all that. 0:06:50.150 --> 0:06:52.880 And all of those have individual settings which map over 0:06:52.880 --> 0:06:55.789 to the phone as well. So it's like it's basically 0:06:55.790 --> 0:07:00.919 a giant ecosystem upgrade, like computing ecosystem upgrade within the 0:07:00.920 --> 0:07:04.159 Apple ecosystem. And I do that every year, but sometimes 0:07:04.160 --> 0:07:06.200 I only do a very little. And like the last 0:07:06.200 --> 0:07:08.120 couple years, I've only done a little bit of it, 0:07:08.120 --> 0:07:11.120 and I probably spent 5 or 6 hours this year 0:07:11.120 --> 0:07:13.610 doing a massive cleanup. And the reason I mention it 0:07:13.640 --> 0:07:16.880 is because it feels really good to do that. You've 0:07:16.880 --> 0:07:21.470 probably been feeling weight weighing on you of like, oh man, 0:07:21.470 --> 0:07:23.990 these apps are nasty. Why do I have these? Do 0:07:23.990 --> 0:07:27.260 I have subscriptions that are still turned on? So all 0:07:27.290 --> 0:07:29.960 that to say, I encourage you to go do a cleanup. 0:07:29.960 --> 0:07:33.110 It will feel good. Okay, so I did an analysis 0:07:33.110 --> 0:07:38.600 of OpenAI's long form conversation with the zero one strawberry team. 0:07:38.780 --> 0:07:40.550 So this is the conversation. 0:07:40.580 --> 0:07:43.730 And I think, you know, for go players and chess 0:07:43.760 --> 0:07:45.530 players that would have come if you go to. 0:07:45.530 --> 0:07:47.750 The beginning, you see I lead the research team here 0:07:47.750 --> 0:07:50.240 at OpenAI. Look, if you see here zero one mini, 0:07:50.270 --> 0:07:52.910 as I click through, it's like, look at that. 0:07:53.450 --> 0:07:56.390 Shot of before there was the first, there was a 0:07:56.390 --> 0:07:58.369 whole bunch of we started talking to the model and 0:07:58.370 --> 0:08:00.740 people were like, wow, this is going to be doing 0:08:00.740 --> 0:08:03.740 something like that. And I think there was a certain 0:08:03.740 --> 0:08:05.270 moment in our in our training. 0:08:05.270 --> 0:08:07.550 Use it. Why? It's better than other things. 0:08:07.550 --> 0:08:10.670 Put more computers in our area than before of GPT 0:08:10.700 --> 0:08:14.810 four trained, first of all generating a long form conversation. 0:08:14.930 --> 0:08:16.250 And wow, this looks. 0:08:16.250 --> 0:08:16.910 Like fairly long. 0:08:16.940 --> 0:08:19.429 Very different than before 32nd. For me, this is the 0:08:19.430 --> 0:08:20.420 moment for. 0:08:20.420 --> 0:08:25.040 So if we, um, take the video URL and we 0:08:25.070 --> 0:08:28.640 come back here, uh, by the way, you don't have 0:08:28.640 --> 0:08:31.790 to use it anymore. You see what I'm doing here? 0:08:31.790 --> 0:08:36.079 This video is being parsed by fabric itself with the 0:08:36.080 --> 0:08:40.370 Y parameter for YouTube. Okay. So if I do that 0:08:40.400 --> 0:08:45.110 I'm going to get back the transcript like that. Um, 0:08:45.110 --> 0:08:48.560 so if I do yeah. Let's do let's do create 0:08:48.559 --> 0:08:51.560 story explanation again. And again. This is using the Pro 0:08:51.590 --> 0:08:54.950 model from Gemini. It feels a little strange summarizing an 0:08:54.950 --> 0:08:59.150 OpenAI video with Gemini, but whatever. It's not that deep. Okay, yeah, 0:08:59.150 --> 0:09:01.069 they've been working on this for a long time. It's 0:09:01.070 --> 0:09:05.360 a reasoning model, meaning it takes time to think before answering. 0:09:05.390 --> 0:09:08.900 The team had several aha moments, like, I feel like 0:09:08.929 --> 0:09:13.250 you could just read these ten bullets and have a conversational, 0:09:13.250 --> 0:09:18.110 explanatory explanation of what's in this video. And that's why 0:09:18.110 --> 0:09:21.890 I created this pattern called Create Story Explanation. It's just 0:09:21.890 --> 0:09:25.130 a way of explaining anything in a super approachable way, 0:09:25.130 --> 0:09:29.329 including like the most crazy, crazy. Okay, I'm going to 0:09:29.330 --> 0:09:33.560 do particle physics, so let's do string theory. Look at this. 0:09:33.590 --> 0:09:36.469 Imagine everything in the universe not made of little balls, 0:09:36.470 --> 0:09:40.190 but tiny vibrating strings. They vibrate at different frequencies, and 0:09:40.190 --> 0:09:43.819 those frequencies determine what kind of particle it appears to be. 0:09:43.970 --> 0:09:47.330 Look at this. That's clean. That is so clean. And 0:09:47.330 --> 0:09:50.910 it's different than then extract wisdom, which I don't even 0:09:50.910 --> 0:09:54.840 know what that would do. Yeah, that. I'm curious. Take 0:09:54.840 --> 0:09:58.560 all of string theory, basically that the model knows about 0:09:58.559 --> 0:10:02.220 and send that to extract wisdom. I very curious what 0:10:02.220 --> 0:10:05.220 this is going to do. Ideas? Holy crap. It is 0:10:05.220 --> 0:10:08.939 extracting all the ideas out of string theory insights. Okay, 0:10:08.970 --> 0:10:15.360 it's doing actual insights. Yeah. Quotes. Nothing. Facts. Nothing. Habits. Nothing. Recommendations. Nice. Okay, 0:10:15.390 --> 0:10:17.699 I'm pretty happy with this, actually. See what we get 0:10:17.700 --> 0:10:21.960 from the insights potential unified framework for understanding all fundamental 0:10:21.960 --> 0:10:26.400 forces and particles. Well, yeah, that's that's why string theory 0:10:26.400 --> 0:10:29.940 was exciting and or is still exciting. Not sure if 0:10:29.940 --> 0:10:32.910 it actually still is. Extra dimensions. A key element of 0:10:32.910 --> 0:10:36.420 string theory could reshape our understanding of the universe. See, 0:10:36.450 --> 0:10:41.880 this is really cool, right? Look at this. Look at ideas. Insights. 0:10:41.880 --> 0:10:45.840 This is really cool and powerful, but it's not as 0:10:45.840 --> 0:10:48.160 cool as this. If you're trying to explain it to 0:10:48.190 --> 0:10:52.480 just a regular person or or heck to me on 0:10:52.480 --> 0:10:55.630 any given day, I would like to hear it this way. 0:10:55.630 --> 0:10:57.820 And then if I really want to go deep on it, 0:10:57.820 --> 0:11:01.270 like then maybe extract wisdom to pull this stuff out 0:11:01.270 --> 0:11:05.380 and definitely extract wisdom if I want to extract, you know, 0:11:05.410 --> 0:11:08.350 purely the ideas or whatever, which I actually have separate 0:11:08.350 --> 0:11:11.830 patterns for those. But anyway, bottom line, the reason I 0:11:11.830 --> 0:11:15.370 got on to this is that I did an extraction 0:11:15.400 --> 0:11:19.510 of the techniques that were that they said they were 0:11:19.510 --> 0:11:23.050 using the zero one model from in this interview, they 0:11:23.050 --> 0:11:25.929 said we here's what we love doing. And they went 0:11:25.929 --> 0:11:27.910 around the room to all those different people and it 0:11:27.910 --> 0:11:31.449 was like, hey, what? What do you do? What what 0:11:31.480 --> 0:11:33.820 have you found that oh, one can do that GPT 0:11:33.850 --> 0:11:37.180 four can't. And so they came up with this list. 0:11:37.179 --> 0:11:42.819 Generate code part pass error messages for debugging. Learn complex 0:11:42.820 --> 0:11:51.420 technical subjects asking it to explain concepts without hallucinations. Brainstorm solutions. Okay. Um, okay. 0:11:51.450 --> 0:11:55.080 So I think this might be my Twitter post about it. Yeah. Yeah. 0:11:55.080 --> 0:11:57.329 So I did a thread here. I'll just walk through 0:11:57.330 --> 0:12:00.089 these real quick. So some thoughts. A lot of these 0:12:00.090 --> 0:12:04.200 takeaways are similar to GPT four sonnet that can already do. 0:12:04.200 --> 0:12:07.830 But the main theme is simply asking more of the 0:12:07.830 --> 0:12:11.309 model and expecting more out of it. And another key 0:12:11.309 --> 0:12:15.479 thing seems to be idea generation brainstorming creativity. So it's 0:12:15.480 --> 0:12:18.450 more like a creative partner. Oh, one is rather than 0:12:18.480 --> 0:12:21.660 previous models. So that was one thing they were really 0:12:21.660 --> 0:12:24.720 excited about. One of the ones was I loved the 0:12:24.720 --> 0:12:28.080 bit about asking to connect ideas. So if we go 0:12:28.080 --> 0:12:31.559 back to this list, look at this, um, ask oh 0:12:31.559 --> 0:12:37.199 one deep questions about abstract concepts to receive thoughtful, detailed answers. Um, 0:12:37.200 --> 0:12:40.559 implement projects with little prior knowledge. Oh, here we go. 0:12:40.590 --> 0:12:44.970 Organize unstructured thoughts by asking oh, one to connect ideas 0:12:44.970 --> 0:12:49.660 and identify missing elements. That is cool. I'll just open 0:12:49.660 --> 0:12:55.150 this up, make text more creative by requesting multiple alternative ideas. 0:12:55.150 --> 0:12:58.120 So again, all of these are very thinking based and 0:12:58.120 --> 0:13:01.810 very creativity based. I generate ideas for structure and content 0:13:01.809 --> 0:13:06.429 of writing. Yeah, really really cool stuff. So that's what 0:13:06.429 --> 0:13:09.400 that was about. Um, yeah. And it's in the newsletter, 0:13:09.400 --> 0:13:11.440 so you can go check it out. All right. Wrote 0:13:11.440 --> 0:13:16.120 a piece for AT&T business about AI transforming real time monitoring. 0:13:16.120 --> 0:13:20.890 So this is basically a security post about many AIS. And, uh, yeah, 0:13:20.920 --> 0:13:23.319 I think you'll like it. It's, uh, a piece I've 0:13:23.320 --> 0:13:24.910 been thinking about, so I was happy to do it 0:13:24.910 --> 0:13:29.410 for AT&T business. Who, uh, who sponsored the piece. So 0:13:29.410 --> 0:13:32.949 I'm excited to be a keynote speaker at Swiss Cyberstorm, 0:13:32.950 --> 0:13:36.250 which is a really, really cool, uh, security conference. I've 0:13:36.250 --> 0:13:39.400 been reading more about it. They're not about, like, all 0:13:39.400 --> 0:13:42.160 the latest hacks and like, oh, we we hacked this thing. 0:13:42.190 --> 0:13:43.839 Let me show you how we can break into this 0:13:43.840 --> 0:13:46.390 thing or whatever, which again, I've done a million of 0:13:46.390 --> 0:13:49.089 those talks. I'm all for them. At this stage of 0:13:49.090 --> 0:13:51.520 my career, though, it's like I'm not finding that as 0:13:51.520 --> 0:13:55.240 interesting just because I feel like it's too easy to 0:13:55.270 --> 0:13:58.030 hack things. There are lots of kids standing at the 0:13:58.030 --> 0:14:00.400 top of a stairway, and it's like, easy to walk 0:14:00.400 --> 0:14:02.620 by and push them. Why would you want to walk 0:14:02.620 --> 0:14:06.850 by and push them? The whole world is super fragile. 0:14:06.850 --> 0:14:10.809 It's a surprise that any of this works like I just. 0:14:10.809 --> 0:14:13.060 I'm not getting the same enjoyment that I used to 0:14:13.059 --> 0:14:18.070 get from breaking into things or knocking something over, you 0:14:18.070 --> 0:14:20.110 know what I mean? It's like it doesn't give me 0:14:20.110 --> 0:14:22.420 a good feeling. So the thing that's giving me a 0:14:22.420 --> 0:14:25.510 good feeling now is building and trying to figure out 0:14:25.510 --> 0:14:29.800 what the most difficult problems are in the world. Um, anyway, 0:14:30.070 --> 0:14:33.880 bottom line for this thing, this conference here keep going 0:14:33.910 --> 0:14:37.840 on tangents. Um, this conference focuses on that. It focuses 0:14:37.840 --> 0:14:41.080 on the ideas around security as opposed to like the 0:14:41.080 --> 0:14:45.290 latest hacking thing. And I like a mix of both, honestly. 0:14:45.290 --> 0:14:48.500 But I prefer more the the thoughts and the trends 0:14:48.500 --> 0:14:51.800 and stuff like that of like, where is this going? 0:14:51.800 --> 0:14:54.530 Are we moving to a place where there's no trust? 0:14:54.560 --> 0:14:58.700 How does all security change when there's no trust? What 0:14:58.730 --> 0:15:01.340 sorts of infrastructure are we going to need in that world? 0:15:01.340 --> 0:15:03.740 That kind of thing. And you could use code unsupervised 0:15:03.740 --> 0:15:09.110 learning for 15% off for your registration. And the theme 0:15:09.110 --> 0:15:14.600 for this one is AI Revolution. So yeah. Cyberstorm Swiss Cyberstorm. 0:15:14.630 --> 0:15:17.930 All right. Security Israel launched an extraordinary attack on Hezbollah 0:15:17.960 --> 0:15:21.500 using a combination of supply chain and remote triggering techniques. 0:15:21.500 --> 0:15:23.900 And I've got a whole bunch here. You know, I 0:15:23.900 --> 0:15:26.180 don't think I'm going to go into this. It's it's 0:15:26.180 --> 0:15:30.080 political and it's a bit long and we're already like 0:15:30.110 --> 0:15:32.300 pretty far into this podcast. So I think I'm going 0:15:32.330 --> 0:15:34.760 to let you go and read that one security researcher 0:15:34.760 --> 0:15:39.890 named X, y, z three VA found a catastrophic flaw 0:15:39.890 --> 0:15:43.850 in the Arc browser That lets attackers inject arbitrary code 0:15:43.850 --> 0:15:48.290 into user sessions using just a user ID. That is, 0:15:48.530 --> 0:15:53.210 that is nasty. So presumably I haven't looked deeply at 0:15:53.210 --> 0:15:56.960 this one, but it's like presumably each user has some 0:15:56.960 --> 0:16:00.859 sort of guide. And then there's infrastructure that controls like 0:16:00.860 --> 0:16:03.680 the guides. And then if you attack that infrastructure, you're 0:16:03.680 --> 0:16:07.370 now attacking the user who's using the browser if they're 0:16:07.370 --> 0:16:11.990 in as that guide. That is just so gross. So gross. 0:16:12.020 --> 0:16:19.190 All right. Hacker named Adkar 72 424 has leaked a 0:16:19.190 --> 0:16:23.600 massive database of 3.3 billion unique email addresses or claiming 0:16:23.600 --> 0:16:26.720 they're unique. And he said, yeah, it's not dupes. It's 0:16:26.720 --> 0:16:32.510 not old stuff. It's a new captcha. And evidently it's 21.8GB. 0:16:32.540 --> 0:16:36.590 We're thinking about putting this into Suckless, but that's pretty huge. 0:16:36.590 --> 0:16:41.000 Suckless is already massive as it is. Chinese scientists have 0:16:41.000 --> 0:16:43.880 figured out how to use Starlink satellites to detect stealth 0:16:43.880 --> 0:16:49.520 aircraft and drones, which are designed to dodge radar. Presumably, 0:16:49.520 --> 0:16:54.380 they're basically it's background radiation coming from the satellites. And 0:16:54.380 --> 0:16:58.220 if you look up and you detect the satellite radiation, 0:16:58.370 --> 0:17:01.760 the signals coming from the satellites, and you see things missing, 0:17:01.760 --> 0:17:05.330 you see little cutouts. Well, that's where the thing is. 0:17:05.330 --> 0:17:08.480 That's my guess. Either that or there's bounce off the 0:17:08.480 --> 0:17:12.740 thing that produces like interference, and they're detecting that interference. 0:17:12.740 --> 0:17:15.320 I'm betting it's one of those two things. Thanks to 0:17:15.350 --> 0:17:21.680 Nudge Security for sponsoring nuclei templates. Version ten is out, 0:17:21.710 --> 0:17:25.340 including new Azure Config stuff. Google is making it easier 0:17:25.340 --> 0:17:29.390 to use Passkeys by syncing them automatically via Google Password 0:17:29.390 --> 0:17:34.160 and across Chrome on windows, Mac and Linux and Android 0:17:34.160 --> 0:17:37.460 and iOS. Support is coming soon. Greynoise has been tracking 0:17:37.460 --> 0:17:43.109 mysterious noise Storms of spoofed internet traffic since January of 2020, 0:17:43.109 --> 0:17:46.590 but still remains unknown. They're still trying to figure it out, 0:17:46.619 --> 0:17:51.540 but it includes this love Ascii string in ICMP, and 0:17:51.540 --> 0:17:55.290 they might be like covert communications or DDoS coordination. Kind 0:17:55.320 --> 0:17:57.930 of interesting. That's a man that reminds me of like 0:17:57.960 --> 0:18:01.260 old school, like ham radio stuff or CB radio where 0:18:01.290 --> 0:18:04.560 like you, you hear this signal and it sounds like 0:18:04.590 --> 0:18:07.020 a pulse or something, and then you research and you 0:18:07.020 --> 0:18:10.440 find out it's like some Russian submarine somewhere, or some 0:18:10.470 --> 0:18:14.340 like Siberian radio station. But there's all this digging and 0:18:14.340 --> 0:18:16.830 this is like pre-internet. So it's like really hard to 0:18:16.859 --> 0:18:20.520 find information. Then you find out, oh, Jim. Oh, you 0:18:20.550 --> 0:18:22.770 got to talk to Jim. Jim knows about this. He's 0:18:22.770 --> 0:18:26.040 in whatever Saskatchewan, Idaho. And you talk to Jim and 0:18:26.040 --> 0:18:28.409 he's like, oh yeah, well, there's this one time. I 0:18:28.410 --> 0:18:32.850 love that idea of like digging up archaic knowledge. And 0:18:32.850 --> 0:18:36.180 I find this really interesting where it's like, Greynoise has 0:18:36.180 --> 0:18:39.179 a bunch of sensors and they're seeing this love Ascii 0:18:39.210 --> 0:18:42.630 thing and they're like, what is this? We've been tracking 0:18:42.630 --> 0:18:46.649 it since, you know, January 2020. We don't know what 0:18:46.650 --> 0:18:48.000 it is. We're trying to figure it out. I think 0:18:48.000 --> 0:18:51.090 that's cool. And thanks to Threatlocker for sponsoring the podcast. 0:18:51.119 --> 0:18:53.910 All right, AI tech. So Sam Altman just dropped an 0:18:53.910 --> 0:18:56.220 essay called The Intelligence Age. I'm going to pull this 0:18:56.220 --> 0:18:58.260 thing up. We're not going to read it. But first 0:18:58.260 --> 0:19:02.220 of all, interesting website. I mean, it's just like super 0:19:02.250 --> 0:19:05.280 got this piece of art, which is no doubt AI art, 0:19:05.280 --> 0:19:07.859 I'm guessing probably a doll image. Who knows, maybe he 0:19:07.859 --> 0:19:10.890 made it, but it's like this super basic website. It's 0:19:10.890 --> 0:19:15.780 got a date. It's IoT. Sam Altman. Com and essentially 0:19:15.780 --> 0:19:18.119 what he does. Well, actually, here's what we're going to do. 0:19:18.150 --> 0:19:21.660 Watch this. See, I just did, uh, command a and 0:19:21.660 --> 0:19:23.970 we're going to go back here and we're going to 0:19:23.970 --> 0:19:27.780 do uh, we're going to paste that in. So I 0:19:27.780 --> 0:19:32.220 did a create story explanation of the Sam Altman piece. 0:19:32.220 --> 0:19:36.389 So here we go. Society's collective intelligence has always brought 0:19:36.390 --> 0:19:40.150 us forward and I supercharges that. AI tools will help 0:19:40.150 --> 0:19:44.469 us solve hard problems. Imagine personal AI teams of virtual 0:19:44.470 --> 0:19:48.760 experts helping us create anything shared prosperity where everyone's life 0:19:48.760 --> 0:19:52.090 is better. Deep learning is the key. AI will become 0:19:52.090 --> 0:19:56.260 increasingly capable. AI assistants will handle complex tasks like coordinating 0:19:56.260 --> 0:19:58.780 medical care for us. It will help us create better 0:19:58.780 --> 0:20:02.920 AI and accelerate scientific discovery. It's going to need a 0:20:02.950 --> 0:20:06.790 lot of computing power and energy. There will be challenges. 0:20:06.790 --> 0:20:11.619 The potential benefits are immense, similar to solving climate change 0:20:11.619 --> 0:20:15.760 and even with job market shifts. AI will amplify our 0:20:15.760 --> 0:20:19.510 abilities and create new opportunities. So it's basically a rah 0:20:19.510 --> 0:20:22.660 rah for AI, which is not surprising coming from somebody 0:20:22.660 --> 0:20:25.510 running an AI company. But it was it was interesting. 0:20:25.510 --> 0:20:27.850 It was a good read. And, uh, yeah, a good 0:20:27.850 --> 0:20:33.580 summary there from Google. Gemini 1.5, Pro Dash 002 and 0:20:33.580 --> 0:20:38.440 the create Story explanation pattern in fabric. South Korea's research 0:20:38.440 --> 0:20:43.540 institute has unveiled Deja Vu AI system that analyzes CCTV 0:20:43.540 --> 0:20:48.130 footage to predict and potentially prevent crimes. Yep, basically every 0:20:48.130 --> 0:20:51.430 sci fi ever preventing crimes. Like. I'm not saying this 0:20:51.430 --> 0:20:54.700 is not possible, but I think just because it's possible 0:20:54.700 --> 0:20:57.130 doesn't mean you should run full speed, you know, carrying 0:20:57.130 --> 0:21:00.790 scissors towards that possibility. Look, a lot of good can 0:21:00.790 --> 0:21:03.640 come from anything that's even starts out a little bad. 0:21:03.640 --> 0:21:06.699 And the opposite is absolutely true as well. Like if 0:21:06.700 --> 0:21:11.590 somebody is an obvious, obvious criminal, they're walking like a criminal. 0:21:11.740 --> 0:21:13.510 What does that mean to walk like a criminal? I 0:21:13.510 --> 0:21:16.840 don't know, maybe it means like having a hoodie on 0:21:16.840 --> 0:21:20.379 and glancing suspiciously from side to side at three in 0:21:20.410 --> 0:21:24.670 the morning while carrying a crowbar in a neighborhood full 0:21:24.670 --> 0:21:28.660 of cars that get broken into at roughly 3 a.m. 0:21:28.660 --> 0:21:31.600 with crowbars. Right? That would be an example of like 0:21:31.630 --> 0:21:35.650 a high correlation where, like, you have little much less 0:21:35.650 --> 0:21:39.820 chance of bias, much less chance of jumping to conclusions. 0:21:39.940 --> 0:21:43.449 It's like yeah. Oh, and also they're limping on the 0:21:43.450 --> 0:21:46.300 right side in this particular group. Like puts all the 0:21:46.300 --> 0:21:48.310 loot in their right side or whatever. You know what 0:21:48.310 --> 0:21:50.020 I mean? It's like the more you add up the 0:21:50.020 --> 0:21:55.060 evidence or whatever from like actual data and not random 0:21:55.060 --> 0:21:58.510 things of like matching the clothing or matching the skin 0:21:58.510 --> 0:22:03.159 color or matching whatever, you know, human bias based thing, 0:22:03.190 --> 0:22:06.160 wherever you can have as much data as possible and 0:22:06.160 --> 0:22:11.290 really good analysis and retroactively and currently look at it 0:22:11.320 --> 0:22:14.800 and evaluate it and make sure you're not being like, 0:22:14.830 --> 0:22:17.800 biased in a, in a in a bad way. Sure. 0:22:17.830 --> 0:22:21.399 The problem is that's not where it will stop, okay. 0:22:21.430 --> 0:22:24.340 They will people will deploy that and it will be 0:22:24.340 --> 0:22:28.540 cool and it will be useful to so-called predict crimes. 0:22:28.540 --> 0:22:31.000 But the problem is they'll be like okay, cool. So 0:22:31.000 --> 0:22:34.430 let's open up that faucet. And now suddenly we're detecting 0:22:34.430 --> 0:22:37.430 all people who look different in the neighborhood because they're 0:22:37.430 --> 0:22:40.370 in the wrong neighborhood. Like, well, people who look like 0:22:40.369 --> 0:22:42.980 that don't ever come into this place. And then unless 0:22:42.980 --> 0:22:45.980 they're up to no good and some I could actually learn, 0:22:45.980 --> 0:22:48.410 that doesn't mean the good ones will, but there's going 0:22:48.410 --> 0:22:51.740 to be lots of bad and halfway bad eyes in 0:22:51.740 --> 0:22:54.859 addition to the really, really good ones. So I think 0:22:54.859 --> 0:22:58.670 any time you start predicting crimes and then affecting people's 0:22:58.670 --> 0:23:01.880 lives as a result of those predictions, like that's sci 0:23:01.880 --> 0:23:04.910 fi shit and it's not a good thing. Johnny IV 0:23:04.940 --> 0:23:08.450 has confirmed he's building a hardware AI device with OpenAI. 0:23:08.480 --> 0:23:10.910 This thing actually might crush the others. This is what 0:23:10.910 --> 0:23:13.790 I was talking about, because the synergy between hardware, software 0:23:13.790 --> 0:23:17.419 and aesthetics is really important in AI gadgets. Like something 0:23:17.420 --> 0:23:19.910 you're going to have on you, something you're wearing, whatever. 0:23:19.940 --> 0:23:23.510 Who gets that better than the, you know, British Apple guy? 0:23:23.540 --> 0:23:26.960 Not many people. Blackrock and Microsoft are teaming up to 0:23:26.990 --> 0:23:32.730 raise $30 billion for AI infrastructure, putting it into $100 0:23:32.730 --> 0:23:36.380 billion in investments, hopefully. And they're building data centers and 0:23:36.380 --> 0:23:39.229 energy projects primarily in the US. But this is a 0:23:39.230 --> 0:23:44.270 partnership between Blackrock, Microsoft and the UAE to build data 0:23:44.300 --> 0:23:48.980 centers and energy products projects in the US. Cool Canadian 0:23:48.980 --> 0:23:52.310 study found an AI tool can reduce unexpected deaths in 0:23:52.310 --> 0:23:56.990 hospitals by 26%. LinkedIn has quietly opted users into using 0:23:56.990 --> 0:23:59.869 their data to trade generative AI models. I think they 0:23:59.869 --> 0:24:02.330 might already be rolling this back because there was so 0:24:02.330 --> 0:24:05.120 much backlash about it, and I think in the UK 0:24:05.150 --> 0:24:09.830 they definitely said no. So I'm not sure how much 0:24:09.830 --> 0:24:13.310 that's still continuing. But they tried. They tried and they 0:24:13.310 --> 0:24:15.650 did the right thing by telling people and they freaked out. 0:24:15.680 --> 0:24:21.650 Recent study by ring rover. No ring over found that 76.5%. 0:24:21.650 --> 0:24:26.840 So 77% of recruiters preferred AI generated headshots over real ones. 0:24:26.840 --> 0:24:29.480 Pardon the drinking there or edit it out. Okay, a 0:24:29.480 --> 0:24:31.939 lot of Amazon employees are upset about the requirement to 0:24:31.970 --> 0:24:34.400 go back to five days in the office. Basically, people 0:24:34.400 --> 0:24:38.000 are freaking out. They're like, this is super lame. Whatever. 0:24:38.000 --> 0:24:40.640 I said my piece on that a million times. Sure, 0:24:40.640 --> 0:24:43.700 I'll say it again. Apple's iOS update has RCS support, 0:24:43.700 --> 0:24:46.399 which means you're going to get to do more iPhone 0:24:46.400 --> 0:24:53.840 like things with Android users with green bubbles. So typing indications, emojis, 0:24:53.840 --> 0:24:55.640 stuff like that. It's all going to be a lot 0:24:55.640 --> 0:25:00.290 more smooth now that you're on iOS 18. Apple's A16 0:25:00.290 --> 0:25:03.710 mobile processors are being produced in the US at TSMC's 0:25:03.740 --> 0:25:09.950 Arizona facility. Love this self-reliance American manufacturing push. Absolutely love it. 0:25:09.980 --> 0:25:14.870 Apple's iPhone 16 now supports wireless firmware restoration. I'm guessing 0:25:14.869 --> 0:25:18.470 Android has had this since like 2002. The Apple's Watch 0:25:18.500 --> 0:25:21.410 Apple Watch's Remote app now lets you adjust volume with 0:25:21.410 --> 0:25:25.220 Digital Crown, so it's basically like a better remote control 0:25:25.220 --> 0:25:28.460 now for your Apple TV or whatever. And somebody shares 0:25:28.460 --> 0:25:33.030 his journey of turning blog content into an e-book. Facundo Solano. 0:25:33.060 --> 0:25:37.890 Turning blog content into an e-book. Humans Rick Beato argues 0:25:37.890 --> 0:25:40.979 that music is getting worse because technology has made it 0:25:40.980 --> 0:25:44.220 too easy to produce and consume. There's a new study 0:25:44.220 --> 0:25:47.610 showing that omega three fatty fatty acids can help reduce 0:25:47.609 --> 0:25:51.119 system symptoms of anxiety and depression in mice. So if 0:25:51.119 --> 0:25:54.450 you're a mouse, huge, huge news for you. US Department 0:25:54.450 --> 0:25:57.450 of Energy is rolling out over $3 billion to fund 0:25:57.450 --> 0:26:01.710 more than two dozen battery projects across 14 states. Love it, 0:26:01.740 --> 0:26:04.740 love it, love it. Astronomers have discovered the largest black 0:26:04.740 --> 0:26:08.669 hole jets ever observed. So basically, the black hole on 0:26:08.670 --> 0:26:12.270 its axis is shooting out stuff. These are the largest 0:26:12.270 --> 0:26:18.150 jets they've ever seen, stretching 23 million light years, which 0:26:18.150 --> 0:26:24.120 is lining up 140 Milky Way galaxies. That is ridiculous. 0:26:24.150 --> 0:26:27.930 Voyager one, my favorite like friend. Like, I don't know, 0:26:27.960 --> 0:26:30.570 I'll talk to this guy like, hey, we're still watching you. 0:26:30.600 --> 0:26:33.240 You're doing good. You're doing good, buddy. You're doing good. 0:26:33.270 --> 0:26:38.430 47 year old spacecraft cruising through space since the late 70s. 0:26:38.430 --> 0:26:41.310 Fired up thrusters that he hasn't used in decades. I 0:26:41.340 --> 0:26:43.740 right here. Meanwhile, the asphalt on our roads has to 0:26:43.740 --> 0:26:46.409 be replaced, like, every 45 minutes. They don't make them 0:26:46.410 --> 0:26:48.510 like they used to. Interesting piece in the Wall Street 0:26:48.510 --> 0:26:53.010 Journal about how pediatricians might have sparked the peanut allergy epidemic. 0:26:53.010 --> 0:26:55.500 So basically they they told parents, make sure they don't 0:26:55.530 --> 0:26:58.860 eat peanuts because they're make sure you don't eat peanuts 0:26:58.859 --> 0:27:02.490 because peanuts will give you allergies. Not eating the peanuts 0:27:02.520 --> 0:27:05.070 is what gives you the peanut allergy. I think that's 0:27:05.070 --> 0:27:08.100 what it's saying. I'm not a peanut doctor, but I'm 0:27:08.100 --> 0:27:10.170 pretty sure that's what it's saying. And I've heard this 0:27:10.170 --> 0:27:14.580 from a million places. Ridiculous. Is that my new favorite word? Probably. 0:27:14.609 --> 0:27:19.050 Ohio is directly funding private schools with taxpayer money. I'm 0:27:19.050 --> 0:27:21.479 all for more structure in schools, but I think we 0:27:21.480 --> 0:27:23.520 need to be careful about how we get there, like 0:27:23.520 --> 0:27:27.640 not with a theocracy. Hopefully, Motus is revolutionizing wildlife tracking, 0:27:27.640 --> 0:27:32.050 using lightweight radio transmitters to monitor the movements of small 0:27:32.050 --> 0:27:35.199 flying animals like birds, bats and insects. Putting a radio 0:27:35.200 --> 0:27:39.970 transmitter on an insect that's a big insect. 50,000 animals 0:27:39.970 --> 0:27:45.490 across 400 species since 2014. All right. Discovery reCAPTCHA fish. 0:27:45.520 --> 0:27:48.370 My buddy John Hammond created a fishing tool that mimics 0:27:48.369 --> 0:27:52.240 a reCAPTCHA form and steals your stuff. RGA rip grep 0:27:52.270 --> 0:27:56.919 on growth hormone basically lets you search through not only 0:27:56.950 --> 0:27:59.230 the normal stuff you could do with rip grep, because 0:27:59.230 --> 0:28:01.270 this is a wrapper for rip grep, but lets you 0:28:01.270 --> 0:28:05.380 search through PDFs, ebooks, office documents, and even compressed files. 0:28:05.380 --> 0:28:07.690 So it's got a whole bunch of extras built in 0:28:07.690 --> 0:28:09.879 so that rip grep is just way better. In fact, 0:28:09.880 --> 0:28:11.919 I need to install that like right now. Anyway, I'll 0:28:11.920 --> 0:28:14.890 do that later. All right. Nuclei templates. We talked about 0:28:14.890 --> 0:28:17.020 that one Dune shell a new take on the command 0:28:17.020 --> 0:28:22.090 line experiment experience. Aim to bring a cozy customizable feel 0:28:22.090 --> 0:28:27.250 that bash lacks dam vulnerable drone drone hacking simulator sci 0:28:27.250 --> 0:28:30.700 fi ideas. Someone compiled a massive CSV file containing every 0:28:30.730 --> 0:28:37.540 sci fi idea imaginable. Eli Burzynski Bendersky Eli Bendersky talks 0:28:37.540 --> 0:28:41.890 about building LLM powered applications in go fabric. Recently switched 0:28:41.890 --> 0:28:45.550 to go. We got a sick go developer now helping 0:28:45.550 --> 0:28:49.330 Jonathan and I out and yeah, it's going really well. 0:28:49.660 --> 0:28:52.060 That's definitely the next language I'm going to be doing. 0:28:52.060 --> 0:28:54.550 Everything in is all go. Asset node talks about their 0:28:54.550 --> 0:28:57.040 approach to recon, and Paul Graham's one pager on how 0:28:57.040 --> 0:28:59.650 to start a startup. Really like the format of this thing. 0:28:59.680 --> 0:29:04.030 Look at this thing. Pretty clean idea people. What customers want. Yeah, 0:29:04.060 --> 0:29:07.390 quite nice ideas. Don't call them LMS. Probably the biggest 0:29:07.390 --> 0:29:10.330 idea that's exploded in my mind lately is Karpathy's point 0:29:10.330 --> 0:29:14.290 about LMS being poorly named. So he basically says Transformers 0:29:14.290 --> 0:29:18.490 are general purpose computer systems and that LMS are actually 0:29:18.490 --> 0:29:22.270 sequence predictors. And crucially, it doesn't matter what the stream is. 0:29:22.300 --> 0:29:25.000 We just happen to be sending language right now. But 0:29:25.000 --> 0:29:28.420 you can actually send anything. And it actually that takes 0:29:28.420 --> 0:29:31.600 us right into the recommendation of the week. Start reframing 0:29:31.600 --> 0:29:35.020 your thinking about AI and specifically about llms away from 0:29:35.020 --> 0:29:38.590 just the next token of text. Instead, think of it 0:29:38.590 --> 0:29:42.580 as sequence prediction and then one level after that, think 0:29:42.580 --> 0:29:46.480 of it as answer prediction. So as Karpathy talks about 0:29:46.510 --> 0:29:52.120 transformer architecture works on sequences of anything languages was, you know, 0:29:52.150 --> 0:29:55.630 language was just a natural start. It works on whatever 0:29:55.630 --> 0:29:58.660 you feed it. It's just a sequence thing. So the 0:29:58.660 --> 0:30:01.540 recommendation of the week is to update your model of 0:30:01.540 --> 0:30:06.880 AI from specific text predictor to generalized answer predictor. And 0:30:06.880 --> 0:30:10.120 in fact, that's my new favorite acronym for AI. I'm 0:30:10.120 --> 0:30:12.940 going to try not seriously to try to convince people 0:30:12.940 --> 0:30:18.280 to use this gap gap generalized answer predictor. I think 0:30:18.280 --> 0:30:21.310 it's way better than LM. It doesn't mean it's going 0:30:21.340 --> 0:30:25.010 to win it. You know, the best acronym doesn't win. 0:30:25.010 --> 0:30:28.100 It's the one that people just decide to use. And 0:30:28.100 --> 0:30:30.980 the aphorism for the week people are strange. They are 0:30:30.980 --> 0:30:34.070 constantly angered by trivial things. But on a major matter 0:30:34.070 --> 0:30:37.670 like totally wasting their lives, they hardly seem to notice. 0:30:37.700 --> 0:30:41.780 People are strange. They're constantly angered by trivial things. But 0:30:41.780 --> 0:30:45.680 on a major matter like totally wasting their lives, they 0:30:45.710 --> 0:30:51.740 hardly seem to notice. Charles Bukowski. Unsupervised learning is produced 0:30:51.740 --> 0:30:54.800 and edited by Daniel Meisler on a Neumann U87 AI 0:30:54.800 --> 0:30:59.000 microphone using Hindenburg. Intro and outro music is by zombie 0:30:59.030 --> 0:31:02.030 with a Y. And to get the text and links 0:31:02.030 --> 0:31:04.280 from this episode, sign up for the newsletter version of 0:31:04.280 --> 0:31:09.920 the show at Daniel meisler.com/newsletter. We'll see you next time.