1 00:00:01,129 --> 00:00:04,460 S1: Welcome to Unsupervised Learning, a security, AI and meaning focused 2 00:00:04,460 --> 00:00:07,340 S1: podcast that looks at how best to thrive as humans 3 00:00:07,340 --> 00:00:11,570 S1: in a post AI world. It combines original ideas, analysis, 4 00:00:11,570 --> 00:00:14,810 S1: and mental models to bring not just the news, but 5 00:00:14,810 --> 00:00:22,250 S1: why it matters and how to respond. All right, welcome 6 00:00:22,250 --> 00:00:25,279 S1: to unsupervised learning. This is Daniel Miessler. All right. Episode 7 00:00:25,280 --> 00:00:30,740 S1: 441 441 of these. That's pretty insane. Okay. Friend of UL, 8 00:00:30,740 --> 00:00:33,290 S1: Ray ulnar is looking for a new position as a 9 00:00:33,290 --> 00:00:37,550 S1: systems engineer. You should reach out to Ray directly and 10 00:00:37,760 --> 00:00:41,600 S1: got his LinkedIn there. He's one of the best members 11 00:00:41,600 --> 00:00:45,739 S1: of UL and, uh, super, super smart, super kind and 12 00:00:45,740 --> 00:00:48,920 S1: very competent as well as you could see from his LinkedIn. And, uh, 13 00:00:48,920 --> 00:00:52,100 S1: he just came onto the market. So I would grab 14 00:00:52,100 --> 00:00:55,900 S1: him before he goes. Added a couple of new levels 15 00:00:56,020 --> 00:00:59,440 S1: to the AGI definitions. So if we click through there 16 00:00:59,440 --> 00:01:03,400 S1: and we go down here, here are the definitions or 17 00:01:03,400 --> 00:01:07,690 S1: the levels. So AGI one is better but with significant drawbacks. 18 00:01:07,690 --> 00:01:13,480 S1: Two is competent but imperfect. Three is a full $80,000 19 00:01:13,480 --> 00:01:18,400 S1: a year white collar worker replacement. Four Okay, this is 20 00:01:18,400 --> 00:01:21,639 S1: where it gets cool for is a world class employee 21 00:01:21,640 --> 00:01:26,080 S1: such as Andre Karpathy or Jeff Dean. So imagine 1% 22 00:01:26,080 --> 00:01:30,370 S1: of 1% and vision creativity programming execution ability. This is 23 00:01:30,370 --> 00:01:33,820 S1: like top top tier. So imagine there's only like 100 24 00:01:33,819 --> 00:01:37,060 S1: or 1000 of these employees on the whole entire planet 25 00:01:37,060 --> 00:01:39,789 S1: at any given time. Wherever you want to set that bar, 26 00:01:39,790 --> 00:01:45,060 S1: it's something like that. Level five is a pinnacle human employee. Okay, 27 00:01:45,060 --> 00:01:48,690 S1: so this is somebody like the smartest person who's ever lived. 28 00:01:48,690 --> 00:01:51,660 S1: So a John von Neumann or an Isaac Newton or 29 00:01:51,660 --> 00:01:54,960 S1: a Richard Feynman or a Claude Shannon. So what this 30 00:01:54,960 --> 00:01:58,680 S1: offers over tier four is the ability to invent completely 31 00:01:58,680 --> 00:02:01,350 S1: new things when they don't exist, or see and explain 32 00:02:01,350 --> 00:02:04,410 S1: the world in a completely new way. So Newton didn't 33 00:02:04,410 --> 00:02:08,970 S1: have calculus, so he invented it. That's always, like, astounded me. 34 00:02:09,000 --> 00:02:12,480 S1: Feynman was a supreme teacher, like a lot of people 35 00:02:12,480 --> 00:02:14,520 S1: consider him to be one of the best teachers ever. 36 00:02:14,550 --> 00:02:19,680 S1: Von Neumann innovated in mathematics, physics, engineering, game theory, like 37 00:02:19,680 --> 00:02:22,709 S1: tons of stuff, and Claude Shannon gave us information theory, 38 00:02:22,710 --> 00:02:26,790 S1: the foundations of cryptography, and a million other things. So basically, 39 00:02:26,790 --> 00:02:29,040 S1: at this level, you not only have the creativity and 40 00:02:29,040 --> 00:02:32,660 S1: execution of a top like whatever, only 100 or 1000 41 00:02:32,660 --> 00:02:35,210 S1: in the entire world at any given moment. But you 42 00:02:35,210 --> 00:02:39,290 S1: also have the once in a generation level innovation capabilities. 43 00:02:39,290 --> 00:02:42,860 S1: And once you're at Pinnacle Human, okay, the level after 44 00:02:42,860 --> 00:02:46,220 S1: that is where I cross over into ASI, which is 45 00:02:46,220 --> 00:02:49,940 S1: defined logically as a level of general AI that's smarter 46 00:02:49,940 --> 00:02:53,210 S1: and more capable than any human that's ever lived. And 47 00:02:53,210 --> 00:02:55,220 S1: that's the reason I added these, is so we have 48 00:02:55,220 --> 00:02:59,540 S1: a smoother transition from the end of AGI right into 49 00:02:59,540 --> 00:03:02,870 S1: the beginning of AC one. Right. And that's where we 50 00:03:02,870 --> 00:03:08,660 S1: have right here AC one. So that was that. Uh, 51 00:03:08,660 --> 00:03:11,419 S1: I feel like Apple Notes is my actual operating system 52 00:03:11,419 --> 00:03:14,150 S1: and Mac OS is just the window manager, just, uh, 53 00:03:14,150 --> 00:03:19,040 S1: stupid jokes. Um, how many notes do I have in 54 00:03:19,050 --> 00:03:21,750 S1: Apple Notes? I'm going to look right now to try 55 00:03:21,750 --> 00:03:28,139 S1: to emphasize this point. 3867 Apple notes my operating system. 56 00:03:28,139 --> 00:03:31,020 S1: Not really, but you see what I'm saying? All right, 57 00:03:31,020 --> 00:03:36,480 S1: let's get into it. Okay, so substrate I've been thinking 58 00:03:36,480 --> 00:03:39,120 S1: about and working on this one for months now and 59 00:03:39,120 --> 00:03:42,240 S1: finally announced it. And actually there's a separate whole substrate 60 00:03:42,240 --> 00:03:45,850 S1: video coming out very soon you should check out. So, uh, yeah, 61 00:03:45,850 --> 00:03:47,530 S1: go look at that when it comes out. I'm not 62 00:03:47,530 --> 00:03:49,660 S1: going to go into detail here, because I just did 63 00:03:49,660 --> 00:03:52,090 S1: in that other piece, got a new piece on dynamic 64 00:03:52,090 --> 00:03:54,370 S1: content summaries and how I think they're going to be 65 00:03:54,370 --> 00:03:56,800 S1: the way we view content in the future. Essentially, the 66 00:03:56,800 --> 00:04:01,180 S1: idea here is that rather than get raw content from 67 00:04:01,180 --> 00:04:05,170 S1: different places, I will produce content for us dynamic. So 68 00:04:05,170 --> 00:04:08,170 S1: imagine that we love to see Richard Feynman explain things 69 00:04:08,170 --> 00:04:11,619 S1: to us. Well, anytime anything complex happens in the world, 70 00:04:11,620 --> 00:04:14,170 S1: we're going to have a video. It's going to be 71 00:04:14,170 --> 00:04:17,320 S1: a YouTube video or something like a YouTube video, and 72 00:04:17,320 --> 00:04:21,430 S1: it's going to be Richard Feynman fully deepfaked, fully Deepfaked. 73 00:04:21,430 --> 00:04:24,489 S1: It looks exactly like Richard Feynman, and he's actually got 74 00:04:24,490 --> 00:04:28,120 S1: a piece of chalk, and he's actually in his Caltech 75 00:04:28,120 --> 00:04:33,419 S1: classroom with a giant whiteboard, and he's dynamically whiteboarding out 76 00:04:33,420 --> 00:04:36,210 S1: this concept to you and explaining with his voice and 77 00:04:36,210 --> 00:04:39,420 S1: his jokes and his humor or whatever. And he's explaining 78 00:04:39,420 --> 00:04:41,850 S1: everything to you. In fact, it might be like if 79 00:04:41,850 --> 00:04:46,770 S1: it's science related or complexity related or whatever, something Feynman ish, 80 00:04:46,770 --> 00:04:49,680 S1: then use Feynman. But maybe you've got this other like 81 00:04:49,680 --> 00:04:53,219 S1: cute influencer girl who you like to explain, like modern 82 00:04:53,220 --> 00:04:56,279 S1: cultural topics or whatever, and she is deepfaked to do 83 00:04:56,279 --> 00:04:58,170 S1: that for you there. And of course, there's going to 84 00:04:58,170 --> 00:05:00,510 S1: be all sorts of considerations around like, well, do they 85 00:05:00,510 --> 00:05:02,520 S1: have permission to do that? And a lot of people 86 00:05:02,520 --> 00:05:04,799 S1: will say, yeah, go ahead, do it. As long as 87 00:05:04,800 --> 00:05:08,430 S1: it falls within these criteria. Some people won't care. They'll 88 00:05:08,430 --> 00:05:11,430 S1: just deepfake anybody and do that. Um, other times it'll 89 00:05:11,430 --> 00:05:14,339 S1: be like the avatar of your actual Da. So your 90 00:05:14,339 --> 00:05:17,160 S1: Da is going to your digital assistant, your personal AI 91 00:05:17,160 --> 00:05:20,160 S1: is going to have an avatar as well. So maybe 92 00:05:20,160 --> 00:05:22,520 S1: you like that avatar to tell it to you, right? 93 00:05:22,520 --> 00:05:25,880 S1: Because maybe that one is of a coach, or honestly, 94 00:05:25,880 --> 00:05:27,770 S1: it's just going to become like one of your best friends. 95 00:05:27,770 --> 00:05:31,130 S1: It knows you the closest. It's very, you know, intimately 96 00:05:31,130 --> 00:05:34,400 S1: connected with you. So maybe it's the one teaching you. 97 00:05:34,400 --> 00:05:37,430 S1: Maybe it's you and your own voice teaching you. Okay. 98 00:05:37,430 --> 00:05:40,190 S1: It's going to be very easy to deepfake yourself. So 99 00:05:40,190 --> 00:05:43,909 S1: maybe you're explaining it to yourself like you created the content. 100 00:05:43,910 --> 00:05:46,729 S1: That would be insane. Point is, if there is a 101 00:05:46,730 --> 00:05:51,650 S1: four hour video explaining this very complex topic, okay, it's 102 00:05:51,650 --> 00:05:54,049 S1: a new type of machine learning. It's a new type 103 00:05:54,050 --> 00:05:56,690 S1: of AI. It's a new hacking technique, but it's an 104 00:05:56,690 --> 00:05:59,089 S1: hour long. It's three hours long. It's four hours long. 105 00:05:59,089 --> 00:06:03,380 S1: It's five hours long. It's a 20 video course. The 106 00:06:03,380 --> 00:06:06,320 S1: AI will go and watch it, and then it'll be like, hey, um, 107 00:06:06,320 --> 00:06:08,570 S1: I created a thing. I know you don't have much 108 00:06:08,570 --> 00:06:12,950 S1: time this week, but I created a 14 minute summary 109 00:06:12,950 --> 00:06:16,640 S1: of this 20 video series, and I'm going to show 110 00:06:16,640 --> 00:06:19,760 S1: it to you. It's actually done by Richard Feynman. So 111 00:06:19,760 --> 00:06:23,510 S1: now Richard Feynman explains that entire 20 videos to you 112 00:06:23,510 --> 00:06:26,839 S1: in 14 minutes, perfectly taught by Richard Feynman. So this 113 00:06:26,839 --> 00:06:29,870 S1: is the concept of dynamic content summaries that the raw 114 00:06:29,870 --> 00:06:32,930 S1: version of a thing is usually not going to be 115 00:06:32,930 --> 00:06:35,770 S1: ideal for a person, the ideal will be to have 116 00:06:35,770 --> 00:06:38,770 S1: it pitched in the way they like to receive things. Okay, 117 00:06:38,770 --> 00:06:40,900 S1: because it won't always be a video. Sometimes it will 118 00:06:40,900 --> 00:06:43,330 S1: be an audio book. It'll just be audio in your ear. 119 00:06:43,330 --> 00:06:45,520 S1: Other times it'll be like a visual story, and it 120 00:06:45,520 --> 00:06:48,070 S1: won't actually have an avatar or a person teaching it 121 00:06:48,070 --> 00:06:51,010 S1: on a whiteboard. It'll just be like animation, and maybe 122 00:06:51,010 --> 00:06:52,990 S1: people will learn better from that. Maybe people will learn 123 00:06:52,990 --> 00:06:55,720 S1: better from a story, whatever they prefer. That content will 124 00:06:55,720 --> 00:06:59,620 S1: be dynamically created by AI to best give you the 125 00:06:59,620 --> 00:07:02,710 S1: content from the raw version. And of course, I'm a 126 00:07:02,710 --> 00:07:05,620 S1: huge fan of slow, slow learning, and there's a whole 127 00:07:05,620 --> 00:07:08,680 S1: bunch of stuff in fabric already for this. Like sometimes 128 00:07:08,680 --> 00:07:11,680 S1: you don't want that summary, and sometimes you should stop 129 00:07:11,680 --> 00:07:15,010 S1: watching those summaries and say in the AI should say, 130 00:07:15,010 --> 00:07:17,680 S1: you know what? I created you a 14 minute summary 131 00:07:17,680 --> 00:07:20,200 S1: of this thing, but this you need to go watch 132 00:07:20,200 --> 00:07:22,750 S1: the full thing because Because oftentimes because I've been using 133 00:07:22,750 --> 00:07:27,100 S1: these summaries with fabric for almost two years now and 134 00:07:27,100 --> 00:07:30,460 S1: I often, well, pretty much always I always get something 135 00:07:30,460 --> 00:07:33,590 S1: more from doing the slow version, from watching the four 136 00:07:33,590 --> 00:07:36,880 S1: hour version, from watching their facial expressions. I get value 137 00:07:36,880 --> 00:07:39,430 S1: from the summary itself, but I get more value from 138 00:07:39,430 --> 00:07:42,790 S1: watching it happen slowly and watching it seep in slowly. 139 00:07:42,790 --> 00:07:44,980 S1: So maybe you do both, maybe you do one or 140 00:07:44,980 --> 00:07:47,490 S1: the other, but don't give up the slow just because 141 00:07:47,490 --> 00:07:49,710 S1: you could do the fast. Exploring the idea of personal 142 00:07:49,710 --> 00:07:52,980 S1: and business brands. Yeah, that was another short piece. Okay. 143 00:07:52,980 --> 00:07:56,640 S1: Kaspersky is shut down US operations because of banning. AT&T 144 00:07:56,640 --> 00:08:00,510 S1: says nearly all cellular customers and some landline users have 145 00:08:00,510 --> 00:08:03,780 S1: had their data stolen. But Kim Zetter at Wired said 146 00:08:03,960 --> 00:08:09,690 S1: somebody got paid like $300,000 from the hacking team by 147 00:08:09,690 --> 00:08:14,400 S1: AT&T to delete what was supposedly a copy of all 148 00:08:14,400 --> 00:08:18,120 S1: the data, and supposedly the only copy of all the data. 149 00:08:18,120 --> 00:08:21,180 S1: So we'll see how that plays out. But good reporting 150 00:08:21,180 --> 00:08:24,270 S1: from Kim Zetter. Russia is using AI enhanced software called 151 00:08:24,270 --> 00:08:30,270 S1: Meliorator to create fake online personas for disinformation campaigns. Ooh, 152 00:08:30,300 --> 00:08:33,209 S1: love this, because the better this gets, the more it's 153 00:08:33,210 --> 00:08:35,780 S1: going to look like a real human. And then if 154 00:08:35,780 --> 00:08:38,510 S1: the AI starts acting like a human with like clicking 155 00:08:38,510 --> 00:08:40,910 S1: on things, being able to tell the difference is going 156 00:08:40,910 --> 00:08:42,739 S1: to get a whole lot more difficult. And this is 157 00:08:42,740 --> 00:08:44,450 S1: actually one of the things I'm most worried about from 158 00:08:44,450 --> 00:08:47,630 S1: AI is disinformation bots, not just the sheer number of them, 159 00:08:47,630 --> 00:08:52,010 S1: but also their how advanced they are, how sophisticated they are. Basically, 160 00:08:52,010 --> 00:08:55,460 S1: the better AI gets, and especially from agent frameworks, the 161 00:08:55,460 --> 00:08:58,700 S1: more it's going to be exactly the same or almost 162 00:08:58,700 --> 00:09:03,140 S1: exactly the same, as if our enemies have millions of 163 00:09:03,140 --> 00:09:07,130 S1: employees in their intelligence agencies. Okay, let me say that again. 164 00:09:07,130 --> 00:09:09,439 S1: That was that was too convoluted. One of the things 165 00:09:09,440 --> 00:09:12,500 S1: I'm most worried about is that the better AI gets, 166 00:09:12,500 --> 00:09:19,850 S1: it'll be like our foreign intelligence agencies for our enemies China, Russia, Iran, 167 00:09:19,850 --> 00:09:23,680 S1: North Korea. Instead of having a few hundred elite people, 168 00:09:23,679 --> 00:09:26,320 S1: they're going to have a few hundred elite people and 169 00:09:26,320 --> 00:09:30,309 S1: 40,000 right below them, elite people. And then the better 170 00:09:30,309 --> 00:09:33,520 S1: the AI gets, those move up into the elite, and 171 00:09:33,520 --> 00:09:37,030 S1: then the elite humans become augmented with that AI and 172 00:09:37,030 --> 00:09:39,910 S1: they become like super elite. And the better the AI gets, 173 00:09:39,910 --> 00:09:42,610 S1: I mean, it just keeps moving up that chain. So 174 00:09:42,610 --> 00:09:46,570 S1: if you think it's bad to have ten, 50, 100 175 00:09:46,570 --> 00:09:51,040 S1: elite people in these different countries who are both Intel 176 00:09:51,040 --> 00:09:55,030 S1: trained and manipulation trained, but they're also like super hackers 177 00:09:55,030 --> 00:09:58,090 S1: and they're good at all these different skills. But there's 178 00:09:58,090 --> 00:10:00,640 S1: I mean, there's only so many in the world. Well, 179 00:10:00,640 --> 00:10:05,860 S1: now imagine them with a thousand of them, 10,000 of them, 100,000, 180 00:10:05,890 --> 00:10:10,030 S1: 100 million. And they're working consistently with no sleep to 181 00:10:10,030 --> 00:10:13,350 S1: go after all your targets. That that's what I'm personally 182 00:10:13,350 --> 00:10:16,770 S1: worried about right now. And if they are producing content 183 00:10:16,770 --> 00:10:19,679 S1: and the content is all manipulative, the internet is going 184 00:10:19,679 --> 00:10:22,680 S1: to have to switch from blocklist to an allow list. 185 00:10:22,679 --> 00:10:25,050 S1: I mean, think about that. The internet currently is an 186 00:10:25,050 --> 00:10:30,090 S1: allow list. I'm sorry, it's a block list. Everything is available, 187 00:10:30,090 --> 00:10:32,520 S1: and you have to specifically go into a place and say, 188 00:10:32,520 --> 00:10:34,050 S1: I don't want to see this. I don't want to 189 00:10:34,050 --> 00:10:36,120 S1: see this. I don't want to see this because of 190 00:10:36,120 --> 00:10:38,579 S1: this problem right here that I'm talking about. We're going 191 00:10:38,580 --> 00:10:40,590 S1: to have to switch. We're going to have to switch 192 00:10:40,590 --> 00:10:44,460 S1: to the other method, which is to say my I, 193 00:10:44,490 --> 00:10:51,420 S1: my internet platform, my mobile OS, okay, Android, macOS, iOS, 194 00:10:51,420 --> 00:10:54,059 S1: they're not going to allow anything in unless you have 195 00:10:54,059 --> 00:10:56,250 S1: the green check mark. What is the green check mark? 196 00:10:56,250 --> 00:11:00,340 S1: Green check mark is an agreement between the the tech platform, 197 00:11:00,340 --> 00:11:04,660 S1: the mobile platform, the OS and trusted providers like a YouTube, 198 00:11:04,660 --> 00:11:10,360 S1: which has approved sources from approved individuals and approved organizations. 199 00:11:10,360 --> 00:11:12,400 S1: So your news is going to have to come that 200 00:11:12,400 --> 00:11:14,650 S1: way because it's going to have to be vetted. So 201 00:11:14,650 --> 00:11:17,350 S1: it gets the green check mark. This might be something 202 00:11:17,350 --> 00:11:19,479 S1: that's required. It's going to happen anyway, but it might 203 00:11:19,480 --> 00:11:23,439 S1: be something that's required within settings and within operating systems 204 00:11:23,440 --> 00:11:26,910 S1: because it's simply too hard to drown out everything. Because 205 00:11:26,910 --> 00:11:30,330 S1: this problem right here, the amount of disinformation bots, pretty 206 00:11:30,330 --> 00:11:35,010 S1: soon it's going to be like, whatever, 50% of all 207 00:11:35,010 --> 00:11:37,140 S1: content is going to be manipulation, then it's going to 208 00:11:37,140 --> 00:11:40,170 S1: be like 75%, then it's going to be 95%. So 209 00:11:40,170 --> 00:11:42,719 S1: if you are naive and you just walk onto the internet, 210 00:11:42,720 --> 00:11:45,720 S1: especially as like a regular person or like a grandma 211 00:11:45,720 --> 00:11:48,300 S1: or something, you're going to be screwed. You're going to 212 00:11:48,300 --> 00:11:50,449 S1: be just like, pulled around. Like all your money will 213 00:11:50,450 --> 00:11:53,090 S1: disappear because you had to save your kids from some 214 00:11:53,090 --> 00:11:55,550 S1: horrible accident. And all of a sudden there was an 215 00:11:55,550 --> 00:11:58,220 S1: asteroid that hit the thing and aliens came. And pretty 216 00:11:58,220 --> 00:12:00,380 S1: soon you're in a bunker and you have no money, 217 00:12:00,380 --> 00:12:03,020 S1: and it's like none of it was real. The amount 218 00:12:03,020 --> 00:12:06,920 S1: of and quality of manipulation. Think about that as well. 219 00:12:06,920 --> 00:12:10,910 S1: The quality of the pitches and the manipulation and the 220 00:12:10,910 --> 00:12:13,959 S1: deepfakes and the trickery is going to get so good. 221 00:12:13,960 --> 00:12:16,480 S1: It's going to trick us too. It's going to trick experts, 222 00:12:16,480 --> 00:12:20,200 S1: not just regular people. There's a new XM vulnerability with 223 00:12:20,200 --> 00:12:23,829 S1: Cvss of 9.1. Google is now offering Passkeys for high 224 00:12:23,830 --> 00:12:27,219 S1: risk users who join Advanced Protection. Previously, they had to 225 00:12:27,220 --> 00:12:33,790 S1: get a physical token foreign influence campaigns analysis from US intelligence. Basically, 226 00:12:33,790 --> 00:12:37,450 S1: they say Russia is backing Trump, probably mostly because of Ukraine. 227 00:12:37,450 --> 00:12:40,510 S1: Iran is acting as a chaos agent in its campaigns, 228 00:12:40,510 --> 00:12:44,320 S1: focusing on exploiting US political and social tensions rather than 229 00:12:44,320 --> 00:12:48,100 S1: backing one side or another. And China's mostly staying out 230 00:12:48,100 --> 00:12:52,510 S1: of the elections just trying to data collect for future 231 00:12:52,510 --> 00:12:57,640 S1: influence operations. Think OPM, uh, Marriott. Yeah, though don't get 232 00:12:57,640 --> 00:13:01,090 S1: me started on those. Thanks to OtterBox for sponsoring. GitLab 233 00:13:01,090 --> 00:13:05,040 S1: has a critical flaw in its CI CD pipelines. Thanks 234 00:13:05,040 --> 00:13:10,380 S1: to Project Discovery for sponsoring and OpenAI's AGI levels, OpenAI 235 00:13:10,380 --> 00:13:13,350 S1: has published their five tier ladder for AI progress. I'm 236 00:13:13,350 --> 00:13:15,959 S1: honestly not a fan. Uh, other than level five, I 237 00:13:15,960 --> 00:13:18,870 S1: find level five really interesting, but the problem is they're 238 00:13:18,870 --> 00:13:21,900 S1: going from like chat bots to human level reasoners and 239 00:13:21,900 --> 00:13:24,809 S1: then to AI agents, then to innovators that can aid 240 00:13:24,809 --> 00:13:27,839 S1: in invention. So level two and level four are way 241 00:13:27,840 --> 00:13:33,300 S1: too close. Okay, so human level reasoners to agents. It's 242 00:13:33,300 --> 00:13:37,199 S1: not a big enough jump. Uh, and both are already possible. 243 00:13:37,200 --> 00:13:39,569 S1: That's the other problem. Level two and level four. They're 244 00:13:39,570 --> 00:13:42,089 S1: kind of already possible. Then you have this really interesting 245 00:13:42,090 --> 00:13:44,460 S1: jump at level five to something that can do the 246 00:13:44,460 --> 00:13:48,690 S1: work of an organization. So for OpenAI level five, AGI 247 00:13:48,690 --> 00:13:51,970 S1: is something that could do the work of an entire organization. 248 00:13:51,970 --> 00:13:56,290 S1: That's not super helpful. Okay. There are lots of very 249 00:13:56,290 --> 00:14:00,160 S1: dumb organizations, okay. An organization could be for people. An 250 00:14:00,160 --> 00:14:04,240 S1: organization can be Apple, an organization could be OpenAI in 251 00:14:04,240 --> 00:14:08,350 S1: the year 2027. So that is not super descriptive in 252 00:14:08,350 --> 00:14:11,230 S1: my opinion. And even worse, the problem is they're mixing 253 00:14:11,230 --> 00:14:14,960 S1: criteria here. Right? So one is I don't know what 254 00:14:14,960 --> 00:14:19,550 S1: what criteria number one is Reasoners is about thinking quality 255 00:14:19,550 --> 00:14:22,700 S1: agents is just an attribute, which means the thing can 256 00:14:22,700 --> 00:14:24,830 S1: take an action. It doesn't say anything about the quality 257 00:14:24,830 --> 00:14:28,940 S1: of the action. So that's not necessarily above one or 258 00:14:28,940 --> 00:14:32,300 S1: below four or above four. It's it doesn't put it 259 00:14:32,300 --> 00:14:35,480 S1: anywhere because it's apples to oranges. Innovators is just a 260 00:14:35,480 --> 00:14:40,010 S1: descriptive output. So it aids in invention. Here's the problem 261 00:14:40,010 --> 00:14:43,430 S1: with this chat bots already aid with invention and that's 262 00:14:43,430 --> 00:14:45,470 S1: what they have as their level. One is chat bots. 263 00:14:45,470 --> 00:14:48,950 S1: The chat bots already help inventors invent better things. Then 264 00:14:48,950 --> 00:14:52,520 S1: you have level five, which is actually about scale more 265 00:14:52,520 --> 00:14:55,850 S1: than quality. Um, yeah. And like I said, scale isn't 266 00:14:55,850 --> 00:14:58,880 S1: super important because you don't know if you're scaling like 267 00:14:58,880 --> 00:15:01,430 S1: an ant colony. They're just kind of doing the same thing, 268 00:15:01,430 --> 00:15:05,080 S1: or if you're scaling and the scale itself leads to 269 00:15:05,080 --> 00:15:09,010 S1: smarter things. So just not overall happy with it. I'm 270 00:15:09,010 --> 00:15:11,410 S1: happy they put it out and I feel bad because 271 00:15:11,410 --> 00:15:13,540 S1: I'm sure someone put some effort into it. And I'm 272 00:15:13,540 --> 00:15:16,450 S1: sure they're human and I don't want you to feel bad. 273 00:15:16,450 --> 00:15:20,110 S1: Just hopefully take this feedback and maybe the next one. 274 00:15:20,110 --> 00:15:24,050 S1: Just think about these ideas, okay? AI startups raising $100 275 00:15:24,050 --> 00:15:27,370 S1: million in 2024. Got a full list there. Anthropic has 276 00:15:27,370 --> 00:15:32,110 S1: more new features, uh, to help automate prompt engineering. And 277 00:15:32,110 --> 00:15:34,750 S1: I see a lot of my friends switching to cloud 278 00:15:34,750 --> 00:15:39,340 S1: over ChatGPT right now, especially sonnet 35. I'm actually using 279 00:15:39,340 --> 00:15:42,880 S1: sonnet 35 as my default within fabric right now, and 280 00:15:42,880 --> 00:15:45,040 S1: it's a leapfrog game because soon we're about to have 281 00:15:45,040 --> 00:15:50,200 S1: opus 35. Cannot wait for that llama 3.0 300 billion 282 00:15:50,200 --> 00:15:54,810 S1: and eventually GPT five, I'm guessing fall to right at 283 00:15:54,810 --> 00:15:59,100 S1: the very beginning of January. So I'm going to say August. September. 284 00:15:59,100 --> 00:16:04,680 S1: I'm going to say between October and February, 90% chance 285 00:16:04,680 --> 00:16:08,340 S1: of GPT five or the next GPT to come out, 286 00:16:08,340 --> 00:16:09,930 S1: whether they call it 4 or 5 or they call 287 00:16:09,930 --> 00:16:13,560 S1: it five, whatever, or some new name. And yeah, 2025 288 00:16:13,560 --> 00:16:16,620 S1: is going to be nuts. So we're gonna we're gonna 289 00:16:16,620 --> 00:16:19,380 S1: have a new president. Well, maybe a new president, but 290 00:16:19,380 --> 00:16:21,300 S1: either way, it's going to be a new whatever. There's 291 00:16:21,300 --> 00:16:23,010 S1: going to be an election and an outcome. So that's 292 00:16:23,010 --> 00:16:25,320 S1: going to be crazy by itself. Um, hopefully not the 293 00:16:25,320 --> 00:16:28,500 S1: bad crazy, but, uh, yeah, there's just going to be 294 00:16:28,500 --> 00:16:32,040 S1: a whole lot going on in 2025, not the least 295 00:16:32,040 --> 00:16:34,710 S1: of which is all these new models. New fiber optic 296 00:16:34,710 --> 00:16:41,430 S1: network transmits data at above 400TB per second. And importantly, 297 00:16:41,430 --> 00:16:45,060 S1: this is using existing fiber, not some, you know, adamantium. 298 00:16:45,060 --> 00:16:47,850 S1: Special stuff. YouTube music is testing. A new feature lets 299 00:16:47,850 --> 00:16:50,910 S1: you use AI to generate a playlist by just describing 300 00:16:50,910 --> 00:16:53,460 S1: what you want to hear. I would use that a lot. 301 00:16:53,460 --> 00:16:57,330 S1: There's a surge in delivery startups like Hail Ify, okay, 302 00:16:57,330 --> 00:17:02,310 S1: Hail Ify, and they're focusing on Shine and Temu, which 303 00:17:02,310 --> 00:17:05,990 S1: are the Chinese people competing with Amazon? Why women are 304 00:17:05,990 --> 00:17:09,290 S1: disappearing from tech Houston is on a path to an 305 00:17:09,290 --> 00:17:13,220 S1: all out power crisis. Tour de France riders are inhaling 306 00:17:13,220 --> 00:17:16,370 S1: carbon monoxide, which is a poison gas. It's the one 307 00:17:16,369 --> 00:17:18,830 S1: that kills you, I think, in your car, in a garage, 308 00:17:18,830 --> 00:17:22,219 S1: if you don't have, like, open ventilation, how how is 309 00:17:22,220 --> 00:17:28,250 S1: a toxic gas improving performance that okay. Yeah. 130 million 310 00:17:28,260 --> 00:17:33,540 S1: US adults have low literacy skills, over half of Americans 311 00:17:33,540 --> 00:17:38,550 S1: aged 16 to 74. Read below. A sixth grade level. 312 00:17:38,550 --> 00:17:40,530 S1: I'm gonna say that again for the people in the back, 313 00:17:40,530 --> 00:17:45,540 S1: over half of Americans aged 16 to 74 read below. 314 00:17:45,540 --> 00:17:49,440 S1: A sixth grade level moving on Colorado poultry workers test 315 00:17:49,440 --> 00:17:52,200 S1: positive for bird flu. But we're not talking about that 316 00:17:52,200 --> 00:17:55,510 S1: because we we're tired of talking about viruses. Uh, we're 317 00:17:55,510 --> 00:17:57,459 S1: tired of talking about Covid. We don't want to talk 318 00:17:57,460 --> 00:18:00,970 S1: about bird flu. That's just another Covid. It's, uh. We're exhausted. 319 00:18:00,970 --> 00:18:03,430 S1: You know what? I give people a break. People are exhausted. 320 00:18:03,460 --> 00:18:06,370 S1: They don't want to talk about negative stuff. They literally 321 00:18:06,369 --> 00:18:09,400 S1: are just going to ignore this with blinders until it 322 00:18:09,400 --> 00:18:12,550 S1: literally comes to their front door and, like, makes their 323 00:18:12,550 --> 00:18:15,790 S1: friends sick. And other than that, any other coverage of this, 324 00:18:15,790 --> 00:18:17,620 S1: they're just going to be like, yeah, I'll know about 325 00:18:17,619 --> 00:18:21,010 S1: it if it gets too bad. Yeah. And meanwhile, all 326 00:18:21,010 --> 00:18:24,040 S1: the experts are like, guys, this is kind of bad. 327 00:18:24,040 --> 00:18:28,570 S1: It's transferring it humans. It's killed 6 million birds and 328 00:18:28,570 --> 00:18:32,980 S1: is now infecting dairy cattle across the state, which means 329 00:18:32,980 --> 00:18:38,770 S1: it's going into milk and dairy workers are getting it. No, 330 00:18:38,770 --> 00:18:41,530 S1: poultry workers are now getting it. So it's like. I mean, 331 00:18:41,530 --> 00:18:44,679 S1: I kind of feel it along with everyone else. I'm like, uh, 332 00:18:44,680 --> 00:18:48,880 S1: I guess if my friends start getting it, I'll pay attention. 333 00:18:48,880 --> 00:18:52,750 S1: Very strange times we live in, we're so, so exhausted 334 00:18:52,750 --> 00:18:56,919 S1: by negativity in certain types of negativity. Just foreign. 10,000 335 00:18:56,920 --> 00:19:00,490 S1: galaxies may host intelligent aliens. This is a new study 336 00:19:00,490 --> 00:19:03,609 S1: basically trying to explain the Drake equation of like why 337 00:19:03,609 --> 00:19:07,889 S1: we're not seeing aliens, but four out of 10,000 galaxies. 338 00:19:07,890 --> 00:19:12,419 S1: I believe there's 200 billion galaxies. Unless I'm mixing up 339 00:19:12,420 --> 00:19:15,060 S1: my numbers. Is it 200 billion galaxies? I think it's 340 00:19:15,060 --> 00:19:20,010 S1: 200 billion. So 4 in 10,000. But there's actually. So 341 00:19:20,010 --> 00:19:23,490 S1: what is that ten K to 200 billion. That's still 342 00:19:23,490 --> 00:19:27,630 S1: a lot of galaxies. That's a lot of galaxies. And 343 00:19:27,630 --> 00:19:30,129 S1: that's just this is just one study okay Okay. So 344 00:19:30,130 --> 00:19:36,880 S1: it's saying 0.003% to 0.2% of exoplanets. Yeah. But there's 345 00:19:36,880 --> 00:19:40,090 S1: a lot of exoplanets like this is not super depressing 346 00:19:40,119 --> 00:19:42,700 S1: to me. But I guess it's interesting if it explains 347 00:19:42,700 --> 00:19:47,500 S1: the Drake equation. Okay, ideas VC's are buying medical practices. 348 00:19:47,500 --> 00:19:50,320 S1: I got a buddy who's talking about how VCs are 349 00:19:50,320 --> 00:19:53,590 S1: just coming into these practices, and they're just moving in 350 00:19:53,590 --> 00:19:55,609 S1: and being like, you've got to sell all this super 351 00:19:55,609 --> 00:19:59,780 S1: gross stuff. Order tests that you don't need, like sell 352 00:19:59,780 --> 00:20:02,659 S1: them miracle drugs that don't actually work. You've got to 353 00:20:02,660 --> 00:20:06,380 S1: do this to hype up the user base. They're basically 354 00:20:06,380 --> 00:20:11,390 S1: turning it into like sexy sales and, um, getting them 355 00:20:11,390 --> 00:20:15,950 S1: to do things other than fundamental, like high quality, proactive 356 00:20:15,950 --> 00:20:19,220 S1: health like diet and exercise. It's kind of going the 357 00:20:19,220 --> 00:20:25,669 S1: opposite direction towards more tests, more assessments, pills and things 358 00:20:25,670 --> 00:20:28,310 S1: that don't actually work. Why? Because they make more money, 359 00:20:28,310 --> 00:20:31,670 S1: which is good for VC's. So really, really bad. And 360 00:20:31,670 --> 00:20:34,880 S1: thanks to my buddy for calling that out. Most conspiracies 361 00:20:34,880 --> 00:20:38,090 S1: come from not realizing how often things fail. The tweet 362 00:20:38,090 --> 00:20:41,930 S1: thread I put out recently on X therapy, rumination, and 363 00:20:41,930 --> 00:20:46,030 S1: untying knots. This is about therapy. Recommend you click on it. 364 00:20:46,060 --> 00:20:48,010 S1: We're already going long so I'm not going to talk 365 00:20:48,010 --> 00:20:52,840 S1: about it here. Discovery fluff on Lambda. Really cool. Uh 366 00:20:52,840 --> 00:20:56,260 S1: really cool project by this guy here. Um, definitely want 367 00:20:56,260 --> 00:20:59,410 S1: to check it out. Bullfrog GitHub action secures your workflows 368 00:20:59,410 --> 00:21:03,159 S1: by controlling outbound network connections. Everything you see is a 369 00:21:03,160 --> 00:21:06,850 S1: computational process if you know how to look. Um. Love it. 370 00:21:06,850 --> 00:21:11,379 S1: Sean Purry emotional Elicitor. This is a prompt by Moritz 371 00:21:11,410 --> 00:21:14,620 S1: Kremp who's a really cool, uh, doing some really cool 372 00:21:14,619 --> 00:21:18,220 S1: prompt stuff. Got to check him out. WTF happened to blogs? 373 00:21:18,220 --> 00:21:21,550 S1: As an employee you are disposable. You never control the 374 00:21:21,550 --> 00:21:25,120 S1: arc of your career. Smoking versus lung cancer deaths. I'm 375 00:21:25,119 --> 00:21:27,129 S1: going to show this one. This is insane. Look at 376 00:21:27,130 --> 00:21:29,980 S1: this thing. Hmm. Wonder if there's a correlation. Well, we 377 00:21:29,980 --> 00:21:34,090 S1: see a correlation. I wonder if there's causation. That's that's compelling. 378 00:21:34,090 --> 00:21:38,710 S1: Look at this. Taxes increasing bans taxes. Look at this. 379 00:21:38,710 --> 00:21:42,400 S1: These are deaths going down. Great depression, World War two 380 00:21:42,400 --> 00:21:46,390 S1: boom US banned cigarette ads on the radio. And surgeon 381 00:21:46,390 --> 00:21:50,320 S1: General report linked smoking to death from cancer. That's the top. 382 00:21:50,320 --> 00:21:52,240 S1: And it starts going down from there with all these 383 00:21:52,240 --> 00:21:56,459 S1: other measures. Really interesting learning multiple concepts from a SQL image, 384 00:21:56,460 --> 00:22:00,390 S1: change detection in satellite imagery. This is a hobby. I 385 00:22:00,390 --> 00:22:02,550 S1: would be doing a lot more. I'm actually waiting for 386 00:22:02,550 --> 00:22:04,500 S1: the AI to get really good at this. I want 387 00:22:04,500 --> 00:22:08,130 S1: to create a bot that basically watches the latest satellite 388 00:22:08,130 --> 00:22:13,170 S1: images from choice locations, uses AI in a pipeline, and 389 00:22:13,320 --> 00:22:17,639 S1: sends me a constant flow of updates of what changed. 390 00:22:17,640 --> 00:22:22,200 S1: Like it looks like they did something underground. Um, a 391 00:22:22,200 --> 00:22:24,390 S1: whole bunch of trucks moved in. They appear to be 392 00:22:24,390 --> 00:22:27,929 S1: this kind of truck. Here are the possible explanations for that. Again, 393 00:22:27,930 --> 00:22:30,479 S1: going back to the Intel thing that I'm building, I 394 00:22:30,480 --> 00:22:35,670 S1: want continuous flows of information flowing into me based on 395 00:22:35,670 --> 00:22:40,919 S1: capturing artifacts, in this case images, and doing AI analysis 396 00:22:40,920 --> 00:22:44,150 S1: on them. 89 things I Know About Git Commits I 397 00:22:44,150 --> 00:22:47,630 S1: got the most comments, uh, about this one and substrate 398 00:22:47,630 --> 00:22:50,689 S1: this week, so definitely check this one out by Jamie 399 00:22:50,690 --> 00:22:53,840 S1: Tanner and the recommendation of the week. Check on your 400 00:22:53,840 --> 00:22:57,290 S1: friends if you haven't heard from them in a while, 401 00:22:57,290 --> 00:23:00,379 S1: send them a text. It's free and they will appreciate 402 00:23:00,380 --> 00:23:03,500 S1: being thought of and be consistent about it. Like push 403 00:23:03,500 --> 00:23:06,859 S1: it right? They they will appreciate that in the age 404 00:23:06,859 --> 00:23:12,290 S1: of infinite leverage, judgment is the most important skill. Naval 405 00:23:12,290 --> 00:23:15,590 S1: has been crushing it lately. Absolutely crushing it. I'm going 406 00:23:15,590 --> 00:23:18,770 S1: to say this one again. In the age of infinite leverage, 407 00:23:18,770 --> 00:23:25,370 S1: judgment is the most important skill. Naval. Unsupervised learning is 408 00:23:25,369 --> 00:23:28,520 S1: produced and edited by Daniel Miller on a Neumann U87 409 00:23:28,520 --> 00:23:32,590 S1: AI microphone using Hindenburg. Intro and outro music is by 410 00:23:32,590 --> 00:23:35,800 S1: Zomby with the Y, and to get the text and 411 00:23:35,800 --> 00:23:38,139 S1: links from this episode, sign up for the newsletter version 412 00:23:38,140 --> 00:23:43,840 S1: of the show at Daniel miessler.com/newsletter. We'll see you next time.