1 00:00:01,129 --> 00:00:04,460 S1: Welcome to Unsupervised Learning, a security, AI and meaning focused 2 00:00:04,460 --> 00:00:07,310 S1: podcast that looks at how best to thrive as humans 3 00:00:07,310 --> 00:00:11,540 S1: in a post AI world. It combines original ideas, analysis, 4 00:00:11,570 --> 00:00:14,780 S1: and mental models to bring not just the news, but 5 00:00:14,780 --> 00:00:22,400 S1: why it matters and how to respond. All right, welcome 6 00:00:22,400 --> 00:00:25,130 S1: to unsupervised learning. This is Daniel Miessler. Okay. First big 7 00:00:25,130 --> 00:00:30,860 S1: news that just happened is there was a release for Gemini. 8 00:00:30,890 --> 00:00:34,700 S1: So there was rumor last night or, I don't know, 9 00:00:34,729 --> 00:00:36,980 S1: last few days that there was going to be a 10 00:00:36,979 --> 00:00:39,830 S1: big model that got released. And it turns out it 11 00:00:39,830 --> 00:00:43,460 S1: was Google as opposed to anthropic, although maybe anthropic is 12 00:00:43,490 --> 00:00:45,200 S1: going to do something as well. I was hoping for 13 00:00:45,229 --> 00:00:47,839 S1: like an opus four or something, or at least an 14 00:00:47,840 --> 00:00:51,229 S1: opus 3.5. So it turns out to have been Google 15 00:00:51,229 --> 00:00:55,790 S1: and it looks like new Gemini 1.5 Pro and Flash models. 16 00:00:55,790 --> 00:00:59,990 S1: So right off the bat, I mean, these seem very minor. 17 00:01:00,020 --> 00:01:04,339 S1: They're minor updates to 1.5 Pro, but whatever. I'm looking 18 00:01:04,340 --> 00:01:10,069 S1: for reasons to be excited. 52% reduction in output token pricing, 19 00:01:10,100 --> 00:01:16,760 S1: two times higher rate limits for Gemini 1.5 flash and 20 00:01:16,760 --> 00:01:20,720 S1: three times higher rate limits for Gemini 1.5 Pro. So 21 00:01:20,720 --> 00:01:24,050 S1: these are really updates to 1.5 Pro. And if I 22 00:01:24,050 --> 00:01:27,500 S1: go to let's see here, if I go to Kitty 23 00:01:27,500 --> 00:01:32,120 S1: and I do this yeah I'm not seeing 002. Oh 24 00:01:32,120 --> 00:01:35,300 S1: there we go. Yeah 20. Okay. So if I put 25 00:01:35,330 --> 00:01:40,370 S1: 20in here into fabric I'm now using Gemini 1.5 Pro 26 00:01:40,400 --> 00:01:43,100 S1: version two, which is the brand new one that just 27 00:01:43,100 --> 00:01:46,880 S1: came out. So I guess we could do like a classic. Um, 28 00:01:46,910 --> 00:01:51,740 S1: so this is the rivets video pipe to fabric SP 29 00:01:52,670 --> 00:01:58,340 S1: extract wisdom DM all right, so streaming works. Seems pretty fast. 30 00:01:58,350 --> 00:02:01,020 S1: See how these insights are? So first of all, it 31 00:02:01,050 --> 00:02:04,500 S1: extracted a whole lot of insights. That's or no that 32 00:02:04,500 --> 00:02:08,400 S1: was ideas. Now insights should be fewer. And these look 33 00:02:08,400 --> 00:02:10,770 S1: really really good. So here's one thing that I love 34 00:02:10,770 --> 00:02:13,980 S1: about AI right now. So if I give this same 35 00:02:13,980 --> 00:02:19,110 S1: challenge to Gemini and GPT four or any of the 36 00:02:19,110 --> 00:02:23,520 S1: modern ones from OpenAI or anthropic, they all come up 37 00:02:23,520 --> 00:02:26,370 S1: with this one as the top quote. You can't necessarily 38 00:02:26,370 --> 00:02:28,620 S1: think yourself into the answers. You have to create space 39 00:02:28,620 --> 00:02:30,840 S1: for the answers to come to you. And I think 40 00:02:30,840 --> 00:02:33,450 S1: that is really, really cool. And all these other quotes 41 00:02:33,450 --> 00:02:35,609 S1: are just fantastic. This is like one of the videos 42 00:02:35,610 --> 00:02:37,350 S1: I know the best. It's like one of my favorite 43 00:02:37,350 --> 00:02:42,120 S1: videos ever. So anyway, this is the results of the 44 00:02:42,120 --> 00:02:47,280 S1: capture from 1.500002, which is the brand new one. And 45 00:02:47,280 --> 00:02:49,830 S1: let's look at the habits. First of all, it followed 46 00:02:49,830 --> 00:02:52,769 S1: instructions really well. And by the way, the DM version 47 00:02:52,770 --> 00:02:55,860 S1: of Extract Wisdom has a whole bunch of extra instructions. 48 00:02:55,889 --> 00:02:59,460 S1: It's like way more chain of thought and way more 49 00:02:59,460 --> 00:03:02,460 S1: prescriptive for like what I'm looking for. So that should 50 00:03:02,460 --> 00:03:07,079 S1: give a pretty good advantage. And it looks like Google 51 00:03:07,080 --> 00:03:10,020 S1: did really well with it. I like the way that 52 00:03:10,020 --> 00:03:12,180 S1: they're saying this. Oh, you know what I want to 53 00:03:12,180 --> 00:03:15,060 S1: do my other new favourite one. So let's take that 54 00:03:15,060 --> 00:03:20,460 S1: one and let's do create story explanation I love this 55 00:03:20,460 --> 00:03:25,530 S1: create story explanation one. Absolutely fantastic. So it basically just 56 00:03:25,530 --> 00:03:29,100 S1: walks you through like you're in a conversation about whatever 57 00:03:29,100 --> 00:03:32,579 S1: the input is. So they lament the decline of creativity 58 00:03:32,580 --> 00:03:35,430 S1: and deep thought. In the West. They discuss the impact 59 00:03:35,430 --> 00:03:39,180 S1: of technology, social media and constant barrage of information. One 60 00:03:39,180 --> 00:03:41,760 S1: speaker shares her experience of deep diving into the history 61 00:03:41,760 --> 00:03:45,450 S1: of the New Testament manuscripts. They discuss the importance of 62 00:03:45,450 --> 00:03:51,089 S1: deep reading, embracing diverse interests. So yeah, the conversation concludes 63 00:03:51,090 --> 00:03:53,790 S1: with a powerful image of the decline of the creative spirit. 64 00:03:53,820 --> 00:03:56,500 S1: I mean, this is just a cool way of describing 65 00:03:56,500 --> 00:03:58,930 S1: a piece of content. So I think I could do 66 00:03:58,960 --> 00:04:01,960 S1: like another one. Neri Oxman so this is one of 67 00:04:01,960 --> 00:04:04,570 S1: my other favorite YouTube videos ever. Yeah, look at this. 68 00:04:04,600 --> 00:04:08,559 S1: Oxman believes nature possesses a wisdom beyond human intelligence. She's 69 00:04:08,560 --> 00:04:13,360 S1: building a company called Oxman to explore the wisdom and 70 00:04:13,360 --> 00:04:16,690 S1: augment nature with technology. Her goal is to grow products, 71 00:04:16,690 --> 00:04:19,539 S1: blurring the line between made and grown. See? Check this 72 00:04:19,540 --> 00:04:23,230 S1: out when I go to summarize why I love this 73 00:04:23,230 --> 00:04:25,510 S1: video so much, it's hard to capture it. It's hard 74 00:04:25,510 --> 00:04:29,110 S1: to remember what I love about it so much, and 75 00:04:29,110 --> 00:04:32,620 S1: this does really well at that. The tech does really 76 00:04:32,620 --> 00:04:35,860 S1: well at capturing what makes her so awesome and it's 77 00:04:35,860 --> 00:04:39,070 S1: got advice. It's just like great. In fact, I might 78 00:04:39,070 --> 00:04:42,520 S1: even get rid of this pre and post sentence. I 79 00:04:42,520 --> 00:04:45,609 S1: think I might just keep that content right there. But 80 00:04:45,610 --> 00:04:49,240 S1: I'm not going to do that live. But anyway. Yeah interesting. 81 00:04:49,270 --> 00:04:51,010 S1: Looks like I have a little bit of a key showing. 82 00:04:51,010 --> 00:04:53,650 S1: But that is not a real key. That is a 83 00:04:53,650 --> 00:04:58,900 S1: HTTP Cookie header. Always like to watch my OpSec here. 84 00:04:58,930 --> 00:05:00,760 S1: Going to leave a token at some point. I'm going 85 00:05:00,790 --> 00:05:02,979 S1: to leave a token to see if I'm going to 86 00:05:02,980 --> 00:05:06,789 S1: leave an API key out there and and monitor to 87 00:05:06,820 --> 00:05:10,930 S1: see if anyone comes in on it. Exercise for the reader. Okay. 88 00:05:10,930 --> 00:05:14,650 S1: So cool. So yeah, we don't have the video of 89 00:05:14,650 --> 00:05:17,080 S1: Parul and Sam Altman yet. It looks like he has 90 00:05:17,080 --> 00:05:20,229 S1: not dropped it yet, but that's fine. Just go back 91 00:05:20,230 --> 00:05:24,760 S1: to the show. Basically some model upgrades from Gemini and 92 00:05:24,760 --> 00:05:27,609 S1: they look to work quite nicely. And I just changed 93 00:05:27,610 --> 00:05:30,910 S1: my default model to it just now. So I'll see 94 00:05:30,910 --> 00:05:37,390 S1: how that does performance wise. Okay. Um, Thomas Rocca created 95 00:05:37,390 --> 00:05:40,299 S1: a web GUI for fabric called fabric UI. Check this 96 00:05:40,300 --> 00:05:44,770 S1: thing out. Look at that. And Jonathan, who's on the 97 00:05:44,770 --> 00:05:47,650 S1: fabric team, is actually working on something like this as well. 98 00:05:47,680 --> 00:05:49,420 S1: But look at this. You can select the model. You 99 00:05:49,420 --> 00:05:51,730 S1: want to use a pretty good selection there. You put 100 00:05:51,730 --> 00:05:53,870 S1: it in your API key. The guy's a hard core 101 00:05:53,870 --> 00:05:55,790 S1: security guy, so I don't think he's going to be 102 00:05:55,790 --> 00:05:59,030 S1: stealing keys anytime soon. Um, and then, look, you pick 103 00:05:59,029 --> 00:06:02,570 S1: your pattern here, and. Yeah, you got a guy to 104 00:06:02,600 --> 00:06:07,340 S1: run fabric with. And again, Jonathan on the fabric team 105 00:06:07,339 --> 00:06:10,220 S1: is actually building something similar. He's building like an API 106 00:06:10,250 --> 00:06:14,270 S1: server and also guy. So that's going to be really cool. 107 00:06:14,420 --> 00:06:19,820 S1: But this guy, uh, Thomas Rocha. Super cool guy. I've 108 00:06:19,820 --> 00:06:22,400 S1: known him for years and years. Uh, he's part of 109 00:06:22,400 --> 00:06:25,940 S1: the UL community and, uh, evidently just built a UI 110 00:06:25,970 --> 00:06:28,400 S1: for fabric. So that was very cool. And I just 111 00:06:28,400 --> 00:06:30,560 S1: spent a whole bunch of time, like, deep cleaning on 112 00:06:30,560 --> 00:06:34,729 S1: my phone, which is the new iPhone 16 Pro and deleted, 113 00:06:34,730 --> 00:06:40,489 S1: probably 40 applications, phone screen cleanup, widgets, Refactors watch faces. 114 00:06:40,520 --> 00:06:43,849 S1: I redid all my focus modes, which are these things 115 00:06:43,850 --> 00:06:47,330 S1: up here and uh, like right here. So you got 116 00:06:47,360 --> 00:06:50,150 S1: like work and personal and Do Not Disturb and all that. 117 00:06:50,150 --> 00:06:52,880 S1: And all of those have individual settings which map over 118 00:06:52,880 --> 00:06:55,789 S1: to the phone as well. So it's like it's basically 119 00:06:55,790 --> 00:07:00,919 S1: a giant ecosystem upgrade, like computing ecosystem upgrade within the 120 00:07:00,920 --> 00:07:04,159 S1: Apple ecosystem. And I do that every year, but sometimes 121 00:07:04,160 --> 00:07:06,200 S1: I only do a very little. And like the last 122 00:07:06,200 --> 00:07:08,120 S1: couple years, I've only done a little bit of it, 123 00:07:08,120 --> 00:07:11,120 S1: and I probably spent 5 or 6 hours this year 124 00:07:11,120 --> 00:07:13,610 S1: doing a massive cleanup. And the reason I mention it 125 00:07:13,640 --> 00:07:16,880 S1: is because it feels really good to do that. You've 126 00:07:16,880 --> 00:07:21,470 S1: probably been feeling weight weighing on you of like, oh man, 127 00:07:21,470 --> 00:07:23,990 S1: these apps are nasty. Why do I have these? Do 128 00:07:23,990 --> 00:07:27,260 S1: I have subscriptions that are still turned on? So all 129 00:07:27,290 --> 00:07:29,960 S1: that to say, I encourage you to go do a cleanup. 130 00:07:29,960 --> 00:07:33,110 S1: It will feel good. Okay, so I did an analysis 131 00:07:33,110 --> 00:07:38,600 S1: of OpenAI's long form conversation with the zero one strawberry team. 132 00:07:38,780 --> 00:07:40,550 S1: So this is the conversation. 133 00:07:40,580 --> 00:07:43,730 S2: And I think, you know, for go players and chess 134 00:07:43,760 --> 00:07:45,530 S2: players that would have come if you go to. 135 00:07:45,530 --> 00:07:47,750 S1: The beginning, you see I lead the research team here 136 00:07:47,750 --> 00:07:50,240 S1: at OpenAI. Look, if you see here zero one mini, 137 00:07:50,270 --> 00:07:52,910 S1: as I click through, it's like, look at that. 138 00:07:53,450 --> 00:07:56,390 S2: Shot of before there was the first, there was a 139 00:07:56,390 --> 00:07:58,369 S2: whole bunch of we started talking to the model and 140 00:07:58,370 --> 00:08:00,740 S2: people were like, wow, this is going to be doing 141 00:08:00,740 --> 00:08:03,740 S2: something like that. And I think there was a certain 142 00:08:03,740 --> 00:08:05,270 S2: moment in our in our training. 143 00:08:05,270 --> 00:08:07,550 S1: Use it. Why? It's better than other things. 144 00:08:07,550 --> 00:08:10,670 S2: Put more computers in our area than before of GPT 145 00:08:10,700 --> 00:08:14,810 S2: four trained, first of all generating a long form conversation. 146 00:08:14,930 --> 00:08:16,250 S2: And wow, this looks. 147 00:08:16,250 --> 00:08:16,910 S1: Like fairly long. 148 00:08:16,940 --> 00:08:19,429 S2: Very different than before 32nd. For me, this is the 149 00:08:19,430 --> 00:08:20,420 S2: moment for. 150 00:08:20,420 --> 00:08:25,040 S1: So if we, um, take the video URL and we 151 00:08:25,070 --> 00:08:28,640 S1: come back here, uh, by the way, you don't have 152 00:08:28,640 --> 00:08:31,790 S1: to use it anymore. You see what I'm doing here? 153 00:08:31,790 --> 00:08:36,079 S1: This video is being parsed by fabric itself with the 154 00:08:36,080 --> 00:08:40,370 S1: Y parameter for YouTube. Okay. So if I do that 155 00:08:40,400 --> 00:08:45,110 S1: I'm going to get back the transcript like that. Um, 156 00:08:45,110 --> 00:08:48,560 S1: so if I do yeah. Let's do let's do create 157 00:08:48,559 --> 00:08:51,560 S1: story explanation again. And again. This is using the Pro 158 00:08:51,590 --> 00:08:54,950 S1: model from Gemini. It feels a little strange summarizing an 159 00:08:54,950 --> 00:08:59,150 S1: OpenAI video with Gemini, but whatever. It's not that deep. Okay, yeah, 160 00:08:59,150 --> 00:09:01,069 S1: they've been working on this for a long time. It's 161 00:09:01,070 --> 00:09:05,360 S1: a reasoning model, meaning it takes time to think before answering. 162 00:09:05,390 --> 00:09:08,900 S1: The team had several aha moments, like, I feel like 163 00:09:08,929 --> 00:09:13,250 S1: you could just read these ten bullets and have a conversational, 164 00:09:13,250 --> 00:09:18,110 S1: explanatory explanation of what's in this video. And that's why 165 00:09:18,110 --> 00:09:21,890 S1: I created this pattern called Create Story Explanation. It's just 166 00:09:21,890 --> 00:09:25,130 S1: a way of explaining anything in a super approachable way, 167 00:09:25,130 --> 00:09:29,329 S1: including like the most crazy, crazy. Okay, I'm going to 168 00:09:29,330 --> 00:09:33,560 S1: do particle physics, so let's do string theory. Look at this. 169 00:09:33,590 --> 00:09:36,469 S1: Imagine everything in the universe not made of little balls, 170 00:09:36,470 --> 00:09:40,190 S1: but tiny vibrating strings. They vibrate at different frequencies, and 171 00:09:40,190 --> 00:09:43,819 S1: those frequencies determine what kind of particle it appears to be. 172 00:09:43,970 --> 00:09:47,330 S1: Look at this. That's clean. That is so clean. And 173 00:09:47,330 --> 00:09:50,910 S1: it's different than then extract wisdom, which I don't even 174 00:09:50,910 --> 00:09:54,840 S1: know what that would do. Yeah, that. I'm curious. Take 175 00:09:54,840 --> 00:09:58,560 S1: all of string theory, basically that the model knows about 176 00:09:58,559 --> 00:10:02,220 S1: and send that to extract wisdom. I very curious what 177 00:10:02,220 --> 00:10:05,220 S1: this is going to do. Ideas? Holy crap. It is 178 00:10:05,220 --> 00:10:08,939 S1: extracting all the ideas out of string theory insights. Okay, 179 00:10:08,970 --> 00:10:15,360 S1: it's doing actual insights. Yeah. Quotes. Nothing. Facts. Nothing. Habits. Nothing. Recommendations. Nice. Okay, 180 00:10:15,390 --> 00:10:17,699 S1: I'm pretty happy with this, actually. See what we get 181 00:10:17,700 --> 00:10:21,960 S1: from the insights potential unified framework for understanding all fundamental 182 00:10:21,960 --> 00:10:26,400 S1: forces and particles. Well, yeah, that's that's why string theory 183 00:10:26,400 --> 00:10:29,940 S1: was exciting and or is still exciting. Not sure if 184 00:10:29,940 --> 00:10:32,910 S1: it actually still is. Extra dimensions. A key element of 185 00:10:32,910 --> 00:10:36,420 S1: string theory could reshape our understanding of the universe. See, 186 00:10:36,450 --> 00:10:41,880 S1: this is really cool, right? Look at this. Look at ideas. Insights. 187 00:10:41,880 --> 00:10:45,840 S1: This is really cool and powerful, but it's not as 188 00:10:45,840 --> 00:10:48,160 S1: cool as this. If you're trying to explain it to 189 00:10:48,190 --> 00:10:52,480 S1: just a regular person or or heck to me on 190 00:10:52,480 --> 00:10:55,630 S1: any given day, I would like to hear it this way. 191 00:10:55,630 --> 00:10:57,820 S1: And then if I really want to go deep on it, 192 00:10:57,820 --> 00:11:01,270 S1: like then maybe extract wisdom to pull this stuff out 193 00:11:01,270 --> 00:11:05,380 S1: and definitely extract wisdom if I want to extract, you know, 194 00:11:05,410 --> 00:11:08,350 S1: purely the ideas or whatever, which I actually have separate 195 00:11:08,350 --> 00:11:11,830 S1: patterns for those. But anyway, bottom line, the reason I 196 00:11:11,830 --> 00:11:15,370 S1: got on to this is that I did an extraction 197 00:11:15,400 --> 00:11:19,510 S1: of the techniques that were that they said they were 198 00:11:19,510 --> 00:11:23,050 S1: using the zero one model from in this interview, they 199 00:11:23,050 --> 00:11:25,929 S1: said we here's what we love doing. And they went 200 00:11:25,929 --> 00:11:27,910 S1: around the room to all those different people and it 201 00:11:27,910 --> 00:11:31,449 S1: was like, hey, what? What do you do? What what 202 00:11:31,480 --> 00:11:33,820 S1: have you found that oh, one can do that GPT 203 00:11:33,850 --> 00:11:37,180 S1: four can't. And so they came up with this list. 204 00:11:37,179 --> 00:11:42,819 S1: Generate code part pass error messages for debugging. Learn complex 205 00:11:42,820 --> 00:11:51,420 S1: technical subjects asking it to explain concepts without hallucinations. Brainstorm solutions. Okay. Um, okay. 206 00:11:51,450 --> 00:11:55,080 S1: So I think this might be my Twitter post about it. Yeah. Yeah. 207 00:11:55,080 --> 00:11:57,329 S1: So I did a thread here. I'll just walk through 208 00:11:57,330 --> 00:12:00,089 S1: these real quick. So some thoughts. A lot of these 209 00:12:00,090 --> 00:12:04,200 S1: takeaways are similar to GPT four sonnet that can already do. 210 00:12:04,200 --> 00:12:07,830 S1: But the main theme is simply asking more of the 211 00:12:07,830 --> 00:12:11,309 S1: model and expecting more out of it. And another key 212 00:12:11,309 --> 00:12:15,479 S1: thing seems to be idea generation brainstorming creativity. So it's 213 00:12:15,480 --> 00:12:18,450 S1: more like a creative partner. Oh, one is rather than 214 00:12:18,480 --> 00:12:21,660 S1: previous models. So that was one thing they were really 215 00:12:21,660 --> 00:12:24,720 S1: excited about. One of the ones was I loved the 216 00:12:24,720 --> 00:12:28,080 S1: bit about asking to connect ideas. So if we go 217 00:12:28,080 --> 00:12:31,559 S1: back to this list, look at this, um, ask oh 218 00:12:31,559 --> 00:12:37,199 S1: one deep questions about abstract concepts to receive thoughtful, detailed answers. Um, 219 00:12:37,200 --> 00:12:40,559 S1: implement projects with little prior knowledge. Oh, here we go. 220 00:12:40,590 --> 00:12:44,970 S1: Organize unstructured thoughts by asking oh, one to connect ideas 221 00:12:44,970 --> 00:12:49,660 S1: and identify missing elements. That is cool. I'll just open 222 00:12:49,660 --> 00:12:55,150 S1: this up, make text more creative by requesting multiple alternative ideas. 223 00:12:55,150 --> 00:12:58,120 S1: So again, all of these are very thinking based and 224 00:12:58,120 --> 00:13:01,810 S1: very creativity based. I generate ideas for structure and content 225 00:13:01,809 --> 00:13:06,429 S1: of writing. Yeah, really really cool stuff. So that's what 226 00:13:06,429 --> 00:13:09,400 S1: that was about. Um, yeah. And it's in the newsletter, 227 00:13:09,400 --> 00:13:11,440 S1: so you can go check it out. All right. Wrote 228 00:13:11,440 --> 00:13:16,120 S1: a piece for AT&T business about AI transforming real time monitoring. 229 00:13:16,120 --> 00:13:20,890 S1: So this is basically a security post about many AIS. And, uh, yeah, 230 00:13:20,920 --> 00:13:23,319 S1: I think you'll like it. It's, uh, a piece I've 231 00:13:23,320 --> 00:13:24,910 S1: been thinking about, so I was happy to do it 232 00:13:24,910 --> 00:13:29,410 S1: for AT&T business. Who, uh, who sponsored the piece. So 233 00:13:29,410 --> 00:13:32,949 S1: I'm excited to be a keynote speaker at Swiss Cyberstorm, 234 00:13:32,950 --> 00:13:36,250 S1: which is a really, really cool, uh, security conference. I've 235 00:13:36,250 --> 00:13:39,400 S1: been reading more about it. They're not about, like, all 236 00:13:39,400 --> 00:13:42,160 S1: the latest hacks and like, oh, we we hacked this thing. 237 00:13:42,190 --> 00:13:43,839 S1: Let me show you how we can break into this 238 00:13:43,840 --> 00:13:46,390 S1: thing or whatever, which again, I've done a million of 239 00:13:46,390 --> 00:13:49,089 S1: those talks. I'm all for them. At this stage of 240 00:13:49,090 --> 00:13:51,520 S1: my career, though, it's like I'm not finding that as 241 00:13:51,520 --> 00:13:55,240 S1: interesting just because I feel like it's too easy to 242 00:13:55,270 --> 00:13:58,030 S1: hack things. There are lots of kids standing at the 243 00:13:58,030 --> 00:14:00,400 S1: top of a stairway, and it's like, easy to walk 244 00:14:00,400 --> 00:14:02,620 S1: by and push them. Why would you want to walk 245 00:14:02,620 --> 00:14:06,850 S1: by and push them? The whole world is super fragile. 246 00:14:06,850 --> 00:14:10,809 S1: It's a surprise that any of this works like I just. 247 00:14:10,809 --> 00:14:13,060 S1: I'm not getting the same enjoyment that I used to 248 00:14:13,059 --> 00:14:18,070 S1: get from breaking into things or knocking something over, you 249 00:14:18,070 --> 00:14:20,110 S1: know what I mean? It's like it doesn't give me 250 00:14:20,110 --> 00:14:22,420 S1: a good feeling. So the thing that's giving me a 251 00:14:22,420 --> 00:14:25,510 S1: good feeling now is building and trying to figure out 252 00:14:25,510 --> 00:14:29,800 S1: what the most difficult problems are in the world. Um, anyway, 253 00:14:30,070 --> 00:14:33,880 S1: bottom line for this thing, this conference here keep going 254 00:14:33,910 --> 00:14:37,840 S1: on tangents. Um, this conference focuses on that. It focuses 255 00:14:37,840 --> 00:14:41,080 S1: on the ideas around security as opposed to like the 256 00:14:41,080 --> 00:14:45,290 S1: latest hacking thing. And I like a mix of both, honestly. 257 00:14:45,290 --> 00:14:48,500 S1: But I prefer more the the thoughts and the trends 258 00:14:48,500 --> 00:14:51,800 S1: and stuff like that of like, where is this going? 259 00:14:51,800 --> 00:14:54,530 S1: Are we moving to a place where there's no trust? 260 00:14:54,560 --> 00:14:58,700 S1: How does all security change when there's no trust? What 261 00:14:58,730 --> 00:15:01,340 S1: sorts of infrastructure are we going to need in that world? 262 00:15:01,340 --> 00:15:03,740 S1: That kind of thing. And you could use code unsupervised 263 00:15:03,740 --> 00:15:09,110 S1: learning for 15% off for your registration. And the theme 264 00:15:09,110 --> 00:15:14,600 S1: for this one is AI Revolution. So yeah. Cyberstorm Swiss Cyberstorm. 265 00:15:14,630 --> 00:15:17,930 S1: All right. Security Israel launched an extraordinary attack on Hezbollah 266 00:15:17,960 --> 00:15:21,500 S1: using a combination of supply chain and remote triggering techniques. 267 00:15:21,500 --> 00:15:23,900 S1: And I've got a whole bunch here. You know, I 268 00:15:23,900 --> 00:15:26,180 S1: don't think I'm going to go into this. It's it's 269 00:15:26,180 --> 00:15:30,080 S1: political and it's a bit long and we're already like 270 00:15:30,110 --> 00:15:32,300 S1: pretty far into this podcast. So I think I'm going 271 00:15:32,330 --> 00:15:34,760 S1: to let you go and read that one security researcher 272 00:15:34,760 --> 00:15:39,890 S1: named X, y, z three VA found a catastrophic flaw 273 00:15:39,890 --> 00:15:43,850 S1: in the Arc browser That lets attackers inject arbitrary code 274 00:15:43,850 --> 00:15:48,290 S1: into user sessions using just a user ID. That is, 275 00:15:48,530 --> 00:15:53,210 S1: that is nasty. So presumably I haven't looked deeply at 276 00:15:53,210 --> 00:15:56,960 S1: this one, but it's like presumably each user has some 277 00:15:56,960 --> 00:16:00,859 S1: sort of guide. And then there's infrastructure that controls like 278 00:16:00,860 --> 00:16:03,680 S1: the guides. And then if you attack that infrastructure, you're 279 00:16:03,680 --> 00:16:07,370 S1: now attacking the user who's using the browser if they're 280 00:16:07,370 --> 00:16:11,990 S1: in as that guide. That is just so gross. So gross. 281 00:16:12,020 --> 00:16:19,190 S1: All right. Hacker named Adkar 72 424 has leaked a 282 00:16:19,190 --> 00:16:23,600 S1: massive database of 3.3 billion unique email addresses or claiming 283 00:16:23,600 --> 00:16:26,720 S1: they're unique. And he said, yeah, it's not dupes. It's 284 00:16:26,720 --> 00:16:32,510 S1: not old stuff. It's a new captcha. And evidently it's 21.8GB. 285 00:16:32,540 --> 00:16:36,590 S1: We're thinking about putting this into Suckless, but that's pretty huge. 286 00:16:36,590 --> 00:16:41,000 S1: Suckless is already massive as it is. Chinese scientists have 287 00:16:41,000 --> 00:16:43,880 S1: figured out how to use Starlink satellites to detect stealth 288 00:16:43,880 --> 00:16:49,520 S1: aircraft and drones, which are designed to dodge radar. Presumably, 289 00:16:49,520 --> 00:16:54,380 S1: they're basically it's background radiation coming from the satellites. And 290 00:16:54,380 --> 00:16:58,220 S1: if you look up and you detect the satellite radiation, 291 00:16:58,370 --> 00:17:01,760 S1: the signals coming from the satellites, and you see things missing, 292 00:17:01,760 --> 00:17:05,330 S1: you see little cutouts. Well, that's where the thing is. 293 00:17:05,330 --> 00:17:08,480 S1: That's my guess. Either that or there's bounce off the 294 00:17:08,480 --> 00:17:12,740 S1: thing that produces like interference, and they're detecting that interference. 295 00:17:12,740 --> 00:17:15,320 S1: I'm betting it's one of those two things. Thanks to 296 00:17:15,350 --> 00:17:21,680 S1: Nudge Security for sponsoring nuclei templates. Version ten is out, 297 00:17:21,710 --> 00:17:25,340 S1: including new Azure Config stuff. Google is making it easier 298 00:17:25,340 --> 00:17:29,390 S1: to use Passkeys by syncing them automatically via Google Password 299 00:17:29,390 --> 00:17:34,160 S1: and across Chrome on windows, Mac and Linux and Android 300 00:17:34,160 --> 00:17:37,460 S1: and iOS. Support is coming soon. Greynoise has been tracking 301 00:17:37,460 --> 00:17:43,109 S1: mysterious noise Storms of spoofed internet traffic since January of 2020, 302 00:17:43,109 --> 00:17:46,590 S1: but still remains unknown. They're still trying to figure it out, 303 00:17:46,619 --> 00:17:51,540 S1: but it includes this love Ascii string in ICMP, and 304 00:17:51,540 --> 00:17:55,290 S1: they might be like covert communications or DDoS coordination. Kind 305 00:17:55,320 --> 00:17:57,930 S1: of interesting. That's a man that reminds me of like 306 00:17:57,960 --> 00:18:01,260 S1: old school, like ham radio stuff or CB radio where 307 00:18:01,290 --> 00:18:04,560 S1: like you, you hear this signal and it sounds like 308 00:18:04,590 --> 00:18:07,020 S1: a pulse or something, and then you research and you 309 00:18:07,020 --> 00:18:10,440 S1: find out it's like some Russian submarine somewhere, or some 310 00:18:10,470 --> 00:18:14,340 S1: like Siberian radio station. But there's all this digging and 311 00:18:14,340 --> 00:18:16,830 S1: this is like pre-internet. So it's like really hard to 312 00:18:16,859 --> 00:18:20,520 S1: find information. Then you find out, oh, Jim. Oh, you 313 00:18:20,550 --> 00:18:22,770 S1: got to talk to Jim. Jim knows about this. He's 314 00:18:22,770 --> 00:18:26,040 S1: in whatever Saskatchewan, Idaho. And you talk to Jim and 315 00:18:26,040 --> 00:18:28,409 S1: he's like, oh yeah, well, there's this one time. I 316 00:18:28,410 --> 00:18:32,850 S1: love that idea of like digging up archaic knowledge. And 317 00:18:32,850 --> 00:18:36,180 S1: I find this really interesting where it's like, Greynoise has 318 00:18:36,180 --> 00:18:39,179 S1: a bunch of sensors and they're seeing this love Ascii 319 00:18:39,210 --> 00:18:42,630 S1: thing and they're like, what is this? We've been tracking 320 00:18:42,630 --> 00:18:46,649 S1: it since, you know, January 2020. We don't know what 321 00:18:46,650 --> 00:18:48,000 S1: it is. We're trying to figure it out. I think 322 00:18:48,000 --> 00:18:51,090 S1: that's cool. And thanks to Threatlocker for sponsoring the podcast. 323 00:18:51,119 --> 00:18:53,910 S1: All right, AI tech. So Sam Altman just dropped an 324 00:18:53,910 --> 00:18:56,220 S1: essay called The Intelligence Age. I'm going to pull this 325 00:18:56,220 --> 00:18:58,260 S1: thing up. We're not going to read it. But first 326 00:18:58,260 --> 00:19:02,220 S1: of all, interesting website. I mean, it's just like super 327 00:19:02,250 --> 00:19:05,280 S1: got this piece of art, which is no doubt AI art, 328 00:19:05,280 --> 00:19:07,859 S1: I'm guessing probably a doll image. Who knows, maybe he 329 00:19:07,859 --> 00:19:10,890 S1: made it, but it's like this super basic website. It's 330 00:19:10,890 --> 00:19:15,780 S1: got a date. It's IoT. Sam Altman. Com and essentially 331 00:19:15,780 --> 00:19:18,119 S1: what he does. Well, actually, here's what we're going to do. 332 00:19:18,150 --> 00:19:21,660 S1: Watch this. See, I just did, uh, command a and 333 00:19:21,660 --> 00:19:23,970 S1: we're going to go back here and we're going to 334 00:19:23,970 --> 00:19:27,780 S1: do uh, we're going to paste that in. So I 335 00:19:27,780 --> 00:19:32,220 S1: did a create story explanation of the Sam Altman piece. 336 00:19:32,220 --> 00:19:36,389 S1: So here we go. Society's collective intelligence has always brought 337 00:19:36,390 --> 00:19:40,150 S1: us forward and I supercharges that. AI tools will help 338 00:19:40,150 --> 00:19:44,469 S1: us solve hard problems. Imagine personal AI teams of virtual 339 00:19:44,470 --> 00:19:48,760 S1: experts helping us create anything shared prosperity where everyone's life 340 00:19:48,760 --> 00:19:52,090 S1: is better. Deep learning is the key. AI will become 341 00:19:52,090 --> 00:19:56,260 S1: increasingly capable. AI assistants will handle complex tasks like coordinating 342 00:19:56,260 --> 00:19:58,780 S1: medical care for us. It will help us create better 343 00:19:58,780 --> 00:20:02,920 S1: AI and accelerate scientific discovery. It's going to need a 344 00:20:02,950 --> 00:20:06,790 S1: lot of computing power and energy. There will be challenges. 345 00:20:06,790 --> 00:20:11,619 S1: The potential benefits are immense, similar to solving climate change 346 00:20:11,619 --> 00:20:15,760 S1: and even with job market shifts. AI will amplify our 347 00:20:15,760 --> 00:20:19,510 S1: abilities and create new opportunities. So it's basically a rah 348 00:20:19,510 --> 00:20:22,660 S1: rah for AI, which is not surprising coming from somebody 349 00:20:22,660 --> 00:20:25,510 S1: running an AI company. But it was it was interesting. 350 00:20:25,510 --> 00:20:27,850 S1: It was a good read. And, uh, yeah, a good 351 00:20:27,850 --> 00:20:33,580 S1: summary there from Google. Gemini 1.5, Pro Dash 002 and 352 00:20:33,580 --> 00:20:38,440 S1: the create Story explanation pattern in fabric. South Korea's research 353 00:20:38,440 --> 00:20:43,540 S1: institute has unveiled Deja Vu AI system that analyzes CCTV 354 00:20:43,540 --> 00:20:48,130 S1: footage to predict and potentially prevent crimes. Yep, basically every 355 00:20:48,130 --> 00:20:51,430 S1: sci fi ever preventing crimes. Like. I'm not saying this 356 00:20:51,430 --> 00:20:54,700 S1: is not possible, but I think just because it's possible 357 00:20:54,700 --> 00:20:57,130 S1: doesn't mean you should run full speed, you know, carrying 358 00:20:57,130 --> 00:21:00,790 S1: scissors towards that possibility. Look, a lot of good can 359 00:21:00,790 --> 00:21:03,640 S1: come from anything that's even starts out a little bad. 360 00:21:03,640 --> 00:21:06,699 S1: And the opposite is absolutely true as well. Like if 361 00:21:06,700 --> 00:21:11,590 S1: somebody is an obvious, obvious criminal, they're walking like a criminal. 362 00:21:11,740 --> 00:21:13,510 S1: What does that mean to walk like a criminal? I 363 00:21:13,510 --> 00:21:16,840 S1: don't know, maybe it means like having a hoodie on 364 00:21:16,840 --> 00:21:20,379 S1: and glancing suspiciously from side to side at three in 365 00:21:20,410 --> 00:21:24,670 S1: the morning while carrying a crowbar in a neighborhood full 366 00:21:24,670 --> 00:21:28,660 S1: of cars that get broken into at roughly 3 a.m. 367 00:21:28,660 --> 00:21:31,600 S1: with crowbars. Right? That would be an example of like 368 00:21:31,630 --> 00:21:35,650 S1: a high correlation where, like, you have little much less 369 00:21:35,650 --> 00:21:39,820 S1: chance of bias, much less chance of jumping to conclusions. 370 00:21:39,940 --> 00:21:43,449 S1: It's like yeah. Oh, and also they're limping on the 371 00:21:43,450 --> 00:21:46,300 S1: right side in this particular group. Like puts all the 372 00:21:46,300 --> 00:21:48,310 S1: loot in their right side or whatever. You know what 373 00:21:48,310 --> 00:21:50,020 S1: I mean? It's like the more you add up the 374 00:21:50,020 --> 00:21:55,060 S1: evidence or whatever from like actual data and not random 375 00:21:55,060 --> 00:21:58,510 S1: things of like matching the clothing or matching the skin 376 00:21:58,510 --> 00:22:03,159 S1: color or matching whatever, you know, human bias based thing, 377 00:22:03,190 --> 00:22:06,160 S1: wherever you can have as much data as possible and 378 00:22:06,160 --> 00:22:11,290 S1: really good analysis and retroactively and currently look at it 379 00:22:11,320 --> 00:22:14,800 S1: and evaluate it and make sure you're not being like, 380 00:22:14,830 --> 00:22:17,800 S1: biased in a, in a in a bad way. Sure. 381 00:22:17,830 --> 00:22:21,399 S1: The problem is that's not where it will stop, okay. 382 00:22:21,430 --> 00:22:24,340 S1: They will people will deploy that and it will be 383 00:22:24,340 --> 00:22:28,540 S1: cool and it will be useful to so-called predict crimes. 384 00:22:28,540 --> 00:22:31,000 S1: But the problem is they'll be like okay, cool. So 385 00:22:31,000 --> 00:22:34,430 S1: let's open up that faucet. And now suddenly we're detecting 386 00:22:34,430 --> 00:22:37,430 S1: all people who look different in the neighborhood because they're 387 00:22:37,430 --> 00:22:40,370 S1: in the wrong neighborhood. Like, well, people who look like 388 00:22:40,369 --> 00:22:42,980 S1: that don't ever come into this place. And then unless 389 00:22:42,980 --> 00:22:45,980 S1: they're up to no good and some I could actually learn, 390 00:22:45,980 --> 00:22:48,410 S1: that doesn't mean the good ones will, but there's going 391 00:22:48,410 --> 00:22:51,740 S1: to be lots of bad and halfway bad eyes in 392 00:22:51,740 --> 00:22:54,859 S1: addition to the really, really good ones. So I think 393 00:22:54,859 --> 00:22:58,670 S1: any time you start predicting crimes and then affecting people's 394 00:22:58,670 --> 00:23:01,880 S1: lives as a result of those predictions, like that's sci 395 00:23:01,880 --> 00:23:04,910 S1: fi shit and it's not a good thing. Johnny IV 396 00:23:04,940 --> 00:23:08,450 S1: has confirmed he's building a hardware AI device with OpenAI. 397 00:23:08,480 --> 00:23:10,910 S1: This thing actually might crush the others. This is what 398 00:23:10,910 --> 00:23:13,790 S1: I was talking about, because the synergy between hardware, software 399 00:23:13,790 --> 00:23:17,419 S1: and aesthetics is really important in AI gadgets. Like something 400 00:23:17,420 --> 00:23:19,910 S1: you're going to have on you, something you're wearing, whatever. 401 00:23:19,940 --> 00:23:23,510 S1: Who gets that better than the, you know, British Apple guy? 402 00:23:23,540 --> 00:23:26,960 S1: Not many people. Blackrock and Microsoft are teaming up to 403 00:23:26,990 --> 00:23:32,730 S1: raise $30 billion for AI infrastructure, putting it into $100 404 00:23:32,730 --> 00:23:36,380 S1: billion in investments, hopefully. And they're building data centers and 405 00:23:36,380 --> 00:23:39,229 S1: energy projects primarily in the US. But this is a 406 00:23:39,230 --> 00:23:44,270 S1: partnership between Blackrock, Microsoft and the UAE to build data 407 00:23:44,300 --> 00:23:48,980 S1: centers and energy products projects in the US. Cool Canadian 408 00:23:48,980 --> 00:23:52,310 S1: study found an AI tool can reduce unexpected deaths in 409 00:23:52,310 --> 00:23:56,990 S1: hospitals by 26%. LinkedIn has quietly opted users into using 410 00:23:56,990 --> 00:23:59,869 S1: their data to trade generative AI models. I think they 411 00:23:59,869 --> 00:24:02,330 S1: might already be rolling this back because there was so 412 00:24:02,330 --> 00:24:05,120 S1: much backlash about it, and I think in the UK 413 00:24:05,150 --> 00:24:09,830 S1: they definitely said no. So I'm not sure how much 414 00:24:09,830 --> 00:24:13,310 S1: that's still continuing. But they tried. They tried and they 415 00:24:13,310 --> 00:24:15,650 S1: did the right thing by telling people and they freaked out. 416 00:24:15,680 --> 00:24:21,650 S1: Recent study by ring rover. No ring over found that 76.5%. 417 00:24:21,650 --> 00:24:26,840 S1: So 77% of recruiters preferred AI generated headshots over real ones. 418 00:24:26,840 --> 00:24:29,480 S1: Pardon the drinking there or edit it out. Okay, a 419 00:24:29,480 --> 00:24:31,939 S1: lot of Amazon employees are upset about the requirement to 420 00:24:31,970 --> 00:24:34,400 S1: go back to five days in the office. Basically, people 421 00:24:34,400 --> 00:24:38,000 S1: are freaking out. They're like, this is super lame. Whatever. 422 00:24:38,000 --> 00:24:40,640 S1: I said my piece on that a million times. Sure, 423 00:24:40,640 --> 00:24:43,700 S1: I'll say it again. Apple's iOS update has RCS support, 424 00:24:43,700 --> 00:24:46,399 S1: which means you're going to get to do more iPhone 425 00:24:46,400 --> 00:24:53,840 S1: like things with Android users with green bubbles. So typing indications, emojis, 426 00:24:53,840 --> 00:24:55,640 S1: stuff like that. It's all going to be a lot 427 00:24:55,640 --> 00:25:00,290 S1: more smooth now that you're on iOS 18. Apple's A16 428 00:25:00,290 --> 00:25:03,710 S1: mobile processors are being produced in the US at TSMC's 429 00:25:03,740 --> 00:25:09,950 S1: Arizona facility. Love this self-reliance American manufacturing push. Absolutely love it. 430 00:25:09,980 --> 00:25:14,870 S1: Apple's iPhone 16 now supports wireless firmware restoration. I'm guessing 431 00:25:14,869 --> 00:25:18,470 S1: Android has had this since like 2002. The Apple's Watch 432 00:25:18,500 --> 00:25:21,410 S1: Apple Watch's Remote app now lets you adjust volume with 433 00:25:21,410 --> 00:25:25,220 S1: Digital Crown, so it's basically like a better remote control 434 00:25:25,220 --> 00:25:28,460 S1: now for your Apple TV or whatever. And somebody shares 435 00:25:28,460 --> 00:25:33,030 S1: his journey of turning blog content into an e-book. Facundo Solano. 436 00:25:33,060 --> 00:25:37,890 S1: Turning blog content into an e-book. Humans Rick Beato argues 437 00:25:37,890 --> 00:25:40,979 S1: that music is getting worse because technology has made it 438 00:25:40,980 --> 00:25:44,220 S1: too easy to produce and consume. There's a new study 439 00:25:44,220 --> 00:25:47,610 S1: showing that omega three fatty fatty acids can help reduce 440 00:25:47,609 --> 00:25:51,119 S1: system symptoms of anxiety and depression in mice. So if 441 00:25:51,119 --> 00:25:54,450 S1: you're a mouse, huge, huge news for you. US Department 442 00:25:54,450 --> 00:25:57,450 S1: of Energy is rolling out over $3 billion to fund 443 00:25:57,450 --> 00:26:01,710 S1: more than two dozen battery projects across 14 states. Love it, 444 00:26:01,740 --> 00:26:04,740 S1: love it, love it. Astronomers have discovered the largest black 445 00:26:04,740 --> 00:26:08,669 S1: hole jets ever observed. So basically, the black hole on 446 00:26:08,670 --> 00:26:12,270 S1: its axis is shooting out stuff. These are the largest 447 00:26:12,270 --> 00:26:18,150 S1: jets they've ever seen, stretching 23 million light years, which 448 00:26:18,150 --> 00:26:24,120 S1: is lining up 140 Milky Way galaxies. That is ridiculous. 449 00:26:24,150 --> 00:26:27,930 S1: Voyager one, my favorite like friend. Like, I don't know, 450 00:26:27,960 --> 00:26:30,570 S1: I'll talk to this guy like, hey, we're still watching you. 451 00:26:30,600 --> 00:26:33,240 S1: You're doing good. You're doing good, buddy. You're doing good. 452 00:26:33,270 --> 00:26:38,430 S1: 47 year old spacecraft cruising through space since the late 70s. 453 00:26:38,430 --> 00:26:41,310 S1: Fired up thrusters that he hasn't used in decades. I 454 00:26:41,340 --> 00:26:43,740 S1: right here. Meanwhile, the asphalt on our roads has to 455 00:26:43,740 --> 00:26:46,409 S1: be replaced, like, every 45 minutes. They don't make them 456 00:26:46,410 --> 00:26:48,510 S1: like they used to. Interesting piece in the Wall Street 457 00:26:48,510 --> 00:26:53,010 S1: Journal about how pediatricians might have sparked the peanut allergy epidemic. 458 00:26:53,010 --> 00:26:55,500 S1: So basically they they told parents, make sure they don't 459 00:26:55,530 --> 00:26:58,860 S1: eat peanuts because they're make sure you don't eat peanuts 460 00:26:58,859 --> 00:27:02,490 S1: because peanuts will give you allergies. Not eating the peanuts 461 00:27:02,520 --> 00:27:05,070 S1: is what gives you the peanut allergy. I think that's 462 00:27:05,070 --> 00:27:08,100 S1: what it's saying. I'm not a peanut doctor, but I'm 463 00:27:08,100 --> 00:27:10,170 S1: pretty sure that's what it's saying. And I've heard this 464 00:27:10,170 --> 00:27:14,580 S1: from a million places. Ridiculous. Is that my new favorite word? Probably. 465 00:27:14,609 --> 00:27:19,050 S1: Ohio is directly funding private schools with taxpayer money. I'm 466 00:27:19,050 --> 00:27:21,479 S1: all for more structure in schools, but I think we 467 00:27:21,480 --> 00:27:23,520 S1: need to be careful about how we get there, like 468 00:27:23,520 --> 00:27:27,640 S1: not with a theocracy. Hopefully, Motus is revolutionizing wildlife tracking, 469 00:27:27,640 --> 00:27:32,050 S1: using lightweight radio transmitters to monitor the movements of small 470 00:27:32,050 --> 00:27:35,199 S1: flying animals like birds, bats and insects. Putting a radio 471 00:27:35,200 --> 00:27:39,970 S1: transmitter on an insect that's a big insect. 50,000 animals 472 00:27:39,970 --> 00:27:45,490 S1: across 400 species since 2014. All right. Discovery reCAPTCHA fish. 473 00:27:45,520 --> 00:27:48,370 S1: My buddy John Hammond created a fishing tool that mimics 474 00:27:48,369 --> 00:27:52,240 S1: a reCAPTCHA form and steals your stuff. RGA rip grep 475 00:27:52,270 --> 00:27:56,919 S1: on growth hormone basically lets you search through not only 476 00:27:56,950 --> 00:27:59,230 S1: the normal stuff you could do with rip grep, because 477 00:27:59,230 --> 00:28:01,270 S1: this is a wrapper for rip grep, but lets you 478 00:28:01,270 --> 00:28:05,380 S1: search through PDFs, ebooks, office documents, and even compressed files. 479 00:28:05,380 --> 00:28:07,690 S1: So it's got a whole bunch of extras built in 480 00:28:07,690 --> 00:28:09,879 S1: so that rip grep is just way better. In fact, 481 00:28:09,880 --> 00:28:11,919 S1: I need to install that like right now. Anyway, I'll 482 00:28:11,920 --> 00:28:14,890 S1: do that later. All right. Nuclei templates. We talked about 483 00:28:14,890 --> 00:28:17,020 S1: that one Dune shell a new take on the command 484 00:28:17,020 --> 00:28:22,090 S1: line experiment experience. Aim to bring a cozy customizable feel 485 00:28:22,090 --> 00:28:27,250 S1: that bash lacks dam vulnerable drone drone hacking simulator sci 486 00:28:27,250 --> 00:28:30,700 S1: fi ideas. Someone compiled a massive CSV file containing every 487 00:28:30,730 --> 00:28:37,540 S1: sci fi idea imaginable. Eli Burzynski Bendersky Eli Bendersky talks 488 00:28:37,540 --> 00:28:41,890 S1: about building LLM powered applications in go fabric. Recently switched 489 00:28:41,890 --> 00:28:45,550 S1: to go. We got a sick go developer now helping 490 00:28:45,550 --> 00:28:49,330 S1: Jonathan and I out and yeah, it's going really well. 491 00:28:49,660 --> 00:28:52,060 S1: That's definitely the next language I'm going to be doing. 492 00:28:52,060 --> 00:28:54,550 S1: Everything in is all go. Asset node talks about their 493 00:28:54,550 --> 00:28:57,040 S1: approach to recon, and Paul Graham's one pager on how 494 00:28:57,040 --> 00:28:59,650 S1: to start a startup. Really like the format of this thing. 495 00:28:59,680 --> 00:29:04,030 S1: Look at this thing. Pretty clean idea people. What customers want. Yeah, 496 00:29:04,060 --> 00:29:07,390 S1: quite nice ideas. Don't call them LMS. Probably the biggest 497 00:29:07,390 --> 00:29:10,330 S1: idea that's exploded in my mind lately is Karpathy's point 498 00:29:10,330 --> 00:29:14,290 S1: about LMS being poorly named. So he basically says Transformers 499 00:29:14,290 --> 00:29:18,490 S1: are general purpose computer systems and that LMS are actually 500 00:29:18,490 --> 00:29:22,270 S1: sequence predictors. And crucially, it doesn't matter what the stream is. 501 00:29:22,300 --> 00:29:25,000 S1: We just happen to be sending language right now. But 502 00:29:25,000 --> 00:29:28,420 S1: you can actually send anything. And it actually that takes 503 00:29:28,420 --> 00:29:31,600 S1: us right into the recommendation of the week. Start reframing 504 00:29:31,600 --> 00:29:35,020 S1: your thinking about AI and specifically about llms away from 505 00:29:35,020 --> 00:29:38,590 S1: just the next token of text. Instead, think of it 506 00:29:38,590 --> 00:29:42,580 S1: as sequence prediction and then one level after that, think 507 00:29:42,580 --> 00:29:46,480 S1: of it as answer prediction. So as Karpathy talks about 508 00:29:46,510 --> 00:29:52,120 S1: transformer architecture works on sequences of anything languages was, you know, 509 00:29:52,150 --> 00:29:55,630 S1: language was just a natural start. It works on whatever 510 00:29:55,630 --> 00:29:58,660 S1: you feed it. It's just a sequence thing. So the 511 00:29:58,660 --> 00:30:01,540 S1: recommendation of the week is to update your model of 512 00:30:01,540 --> 00:30:06,880 S1: AI from specific text predictor to generalized answer predictor. And 513 00:30:06,880 --> 00:30:10,120 S1: in fact, that's my new favorite acronym for AI. I'm 514 00:30:10,120 --> 00:30:12,940 S1: going to try not seriously to try to convince people 515 00:30:12,940 --> 00:30:18,280 S1: to use this gap gap generalized answer predictor. I think 516 00:30:18,280 --> 00:30:21,310 S1: it's way better than LM. It doesn't mean it's going 517 00:30:21,340 --> 00:30:25,010 S1: to win it. You know, the best acronym doesn't win. 518 00:30:25,010 --> 00:30:28,100 S1: It's the one that people just decide to use. And 519 00:30:28,100 --> 00:30:30,980 S1: the aphorism for the week people are strange. They are 520 00:30:30,980 --> 00:30:34,070 S1: constantly angered by trivial things. But on a major matter 521 00:30:34,070 --> 00:30:37,670 S1: like totally wasting their lives, they hardly seem to notice. 522 00:30:37,700 --> 00:30:41,780 S1: People are strange. They're constantly angered by trivial things. But 523 00:30:41,780 --> 00:30:45,680 S1: on a major matter like totally wasting their lives, they 524 00:30:45,710 --> 00:30:51,740 S1: hardly seem to notice. Charles Bukowski. Unsupervised learning is produced 525 00:30:51,740 --> 00:30:54,800 S1: and edited by Daniel Meisler on a Neumann U87 AI 526 00:30:54,800 --> 00:30:59,000 S1: microphone using Hindenburg. Intro and outro music is by zombie 527 00:30:59,030 --> 00:31:02,030 S1: with a Y. And to get the text and links 528 00:31:02,030 --> 00:31:04,280 S1: from this episode, sign up for the newsletter version of 529 00:31:04,280 --> 00:31:09,920 S1: the show at Daniel meisler.com/newsletter. We'll see you next time.