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 --> 00:00:22,369 S1: why it matters and how to respond. All right, welcome 6 00:00:22,370 --> 00:00:24,860 S1: to unsupervised learning. This is Daniel. All right. What do 7 00:00:24,860 --> 00:00:28,310 S1: we have here? All right. So saw this funny thing. 8 00:00:28,310 --> 00:00:34,740 S1: So basically, um, some terrorist funding news. Al Qaeda rebrands 9 00:00:34,740 --> 00:00:40,500 S1: as I. Qaeda raises $1 billion. Cool. I am rereading 10 00:00:40,500 --> 00:00:45,690 S1: Alex Ramos's two business books. Uh, really? I think Alex 11 00:00:45,690 --> 00:00:48,960 S1: Hormozi is, like, the best business content generator right now, 12 00:00:48,960 --> 00:00:51,960 S1: at least for my type of business, although he's like 13 00:00:51,960 --> 00:00:55,630 S1: a gym person. Um, so it's not the same type 14 00:00:55,630 --> 00:01:00,970 S1: of business, but he's just extraordinary at like user acquisition, 15 00:01:01,090 --> 00:01:04,149 S1: like all sorts of stuff. And I love the format 16 00:01:04,150 --> 00:01:05,890 S1: of his content. Like he's doing a whole bunch of 17 00:01:05,890 --> 00:01:09,940 S1: shorts now. He's just making content probably, I don't know, 18 00:01:09,940 --> 00:01:12,550 S1: it's got to be making 4 or 5 pieces a day. 19 00:01:12,700 --> 00:01:15,730 S1: It's just so much and it's really good. My buddy 20 00:01:15,730 --> 00:01:20,250 S1: Jacoby is an Army veteran, really cool security guy, and 21 00:01:20,250 --> 00:01:22,560 S1: he's struggling really hard right now and he's got a 22 00:01:22,590 --> 00:01:26,040 S1: gofund me. So if you could, uh, add to that, 23 00:01:26,040 --> 00:01:28,980 S1: I would really appreciate it. Fabric made it to the 24 00:01:28,980 --> 00:01:33,000 S1: front page of Hacker News. Uh, next run of augmented 25 00:01:33,000 --> 00:01:36,060 S1: is on July 26th. You want to go sign up there? 26 00:01:36,120 --> 00:01:39,600 S1: Got to spend some in-person time with a couple of 27 00:01:39,600 --> 00:01:44,170 S1: fellow entrepreneurs creators over the weekend. Doing this stuff in 28 00:01:44,170 --> 00:01:48,460 S1: person is invaluable. Like, we're talking about wanting to do 29 00:01:48,460 --> 00:01:51,790 S1: a digital version of The Thing that we built, and 30 00:01:51,790 --> 00:01:55,030 S1: I'm reticent to do that because I. I want it 31 00:01:55,030 --> 00:01:58,960 S1: to happen only in person, but we can make it 32 00:01:58,960 --> 00:02:01,720 S1: digitally and then just interact with it in person. So 33 00:02:01,720 --> 00:02:04,060 S1: that's what we're going to do, because it's too valuable 34 00:02:04,060 --> 00:02:07,860 S1: not to be digital, but highly recommend setting up some 35 00:02:07,860 --> 00:02:12,120 S1: similar sessions with your friends because it's really, really powerful. 36 00:02:12,120 --> 00:02:15,570 S1: And my friend Monica Verma is running her CSO masterclass 37 00:02:15,570 --> 00:02:18,150 S1: soon and she also has a newsletter, so go check 38 00:02:18,150 --> 00:02:20,970 S1: those out. My work just finished a resource I've been 39 00:02:20,970 --> 00:02:24,239 S1: working on for over a week called Raid Real World. 40 00:02:24,270 --> 00:02:29,940 S1: I definition's not redundant array of inexpensive disks, which is 41 00:02:29,940 --> 00:02:32,010 S1: the first thing I think of whenever I hear raid. 42 00:02:32,040 --> 00:02:36,870 S1: Either that or bug killer. But yeah, real world AI definitions. 43 00:02:36,990 --> 00:02:39,480 S1: And I'm actually going to go into it and talk 44 00:02:39,480 --> 00:02:42,360 S1: about it. So I'm going to go ahead and jump 45 00:02:42,360 --> 00:02:45,810 S1: in here. Okay. So I see a lot of definitions 46 00:02:45,810 --> 00:02:47,880 S1: for different AI terms out there. And I want to 47 00:02:47,880 --> 00:02:51,180 S1: put my own thoughts into a long form version. And 48 00:02:51,180 --> 00:02:53,750 S1: that's basically why I did this. And it's mostly for 49 00:02:53,750 --> 00:02:57,830 S1: me because I love this quote. I forget the name 50 00:02:57,830 --> 00:03:01,070 S1: of the woman. Uh, I need to memorize that. But 51 00:03:01,070 --> 00:03:05,180 S1: she basically says, um, if you are thinking without writing, 52 00:03:05,180 --> 00:03:07,970 S1: you only think that you're thinking. And I think that 53 00:03:07,970 --> 00:03:12,109 S1: is such a powerful statement, because when you are writing 54 00:03:12,110 --> 00:03:14,180 S1: and this is, um, this is why I call this 55 00:03:14,180 --> 00:03:19,589 S1: a Feynman approach to learning, it's basically Feynman said, if 56 00:03:19,590 --> 00:03:22,110 S1: you think you know something, try to teach it, and 57 00:03:22,110 --> 00:03:24,480 S1: you will figure out pretty quickly that you probably don't 58 00:03:24,480 --> 00:03:26,940 S1: know it as well as you thought you did. And 59 00:03:26,940 --> 00:03:30,450 S1: I've always known this for a very long time. So 60 00:03:30,450 --> 00:03:36,390 S1: starting back in 1999, I started writing tutorials to myself 61 00:03:36,390 --> 00:03:39,720 S1: to explain things. So when I was learning tech, I 62 00:03:39,720 --> 00:03:42,230 S1: literally would try to learn the tech from textbooks and 63 00:03:42,230 --> 00:03:44,120 S1: then it wouldn't make sense to me. So I would 64 00:03:44,120 --> 00:03:46,850 S1: go write my own explanation, and then I would just 65 00:03:46,850 --> 00:03:50,330 S1: use that whenever I wanted to remember something. And that 66 00:03:50,330 --> 00:03:53,270 S1: turned into my tutorials page, which is how I got 67 00:03:53,270 --> 00:03:56,780 S1: started on the internet. So table of contents here, the 68 00:03:56,780 --> 00:04:01,130 S1: expanded definitions table, the one liner definitions table, and then 69 00:04:01,130 --> 00:04:05,420 S1: I break out AGI and ASI into two different sections. 70 00:04:05,420 --> 00:04:08,120 S1: So we're going to start with the one liner AI 71 00:04:08,120 --> 00:04:12,140 S1: definitions table. And this is coming off of Hamel Hussein's 72 00:04:12,170 --> 00:04:16,789 S1: I bullshit knife, which gave violently short definitions for a 73 00:04:16,790 --> 00:04:19,940 S1: bunch of AI terms. And this is my expanded version 74 00:04:19,940 --> 00:04:24,109 S1: of that. Okay, so this is made to like crystallize 75 00:04:24,110 --> 00:04:27,260 S1: in your brain what these things are in very, very 76 00:04:27,260 --> 00:04:30,770 S1: crisp conversational form. And we're going to go into more 77 00:04:30,770 --> 00:04:34,250 S1: detail in the further ones. But I tech that does 78 00:04:34,250 --> 00:04:38,779 S1: cognitive tasks that only humans could do before. And the 79 00:04:38,779 --> 00:04:42,410 S1: reason I have that only humans could do before is 80 00:04:42,410 --> 00:04:45,770 S1: because that's a moving bar. So before there was a 81 00:04:45,770 --> 00:04:51,980 S1: chess playing robot or a chess playing AI, everyone said, well, yeah, 82 00:04:51,980 --> 00:04:54,350 S1: but it can't play chess. It never will be able 83 00:04:54,350 --> 00:04:56,810 S1: to play chess because that's a human thing. And of 84 00:04:56,810 --> 00:05:00,440 S1: course you could do that now. Image identification. Being able 85 00:05:00,440 --> 00:05:03,230 S1: to tell the difference between a dog and a cat, 86 00:05:03,230 --> 00:05:05,779 S1: that was a very, very hard problem for a very 87 00:05:05,779 --> 00:05:08,150 S1: long time. You have to think in terms of like 88 00:05:08,150 --> 00:05:11,600 S1: what currently is a thing that only humans can do 89 00:05:11,600 --> 00:05:15,289 S1: that I cannot do yet that is the barrier which 90 00:05:15,290 --> 00:05:18,870 S1: determines AI or not, because the moment you cross over 91 00:05:18,870 --> 00:05:22,500 S1: into that and I suddenly becomes available to do that, 92 00:05:22,500 --> 00:05:25,979 S1: that's what's currently called I. Now what's interesting is as 93 00:05:25,980 --> 00:05:29,280 S1: it moves further and further out, people forget that that's 94 00:05:29,279 --> 00:05:32,520 S1: I like image recognition. A lot of people are like, 95 00:05:32,520 --> 00:05:35,609 S1: that's not AI, that's ML. Well, it's still AI, first 96 00:05:35,610 --> 00:05:38,640 S1: of all, because machine learning is a subset of AI, 97 00:05:38,640 --> 00:05:42,390 S1: but also because it's doing something that only humans could 98 00:05:42,390 --> 00:05:47,289 S1: do before. So I love that definition. Very flexible machine 99 00:05:47,290 --> 00:05:50,860 S1: learning AI that can improve just by seeing more data. 100 00:05:50,860 --> 00:05:53,260 S1: And we're going to talk about this more, uh, later 101 00:05:53,260 --> 00:05:57,250 S1: on prompting clearly articulate what you want from an AI. 102 00:05:57,279 --> 00:06:01,030 S1: Rag provide context to an AI that's too big or 103 00:06:01,029 --> 00:06:04,089 S1: expensive to fit in a prompt. An agent, an AI 104 00:06:04,089 --> 00:06:07,800 S1: component that does more than just LM call to respond. 105 00:06:07,800 --> 00:06:11,760 S1: So it's not just call and response, that's just regular AI. 106 00:06:11,760 --> 00:06:14,610 S1: An agent does more than that, taking on more of 107 00:06:14,610 --> 00:06:17,549 S1: the work chain of thought. Tell the AI to walk 108 00:06:17,550 --> 00:06:20,940 S1: through its thinking and steps. Zero shot. Ask an AI 109 00:06:20,970 --> 00:06:24,510 S1: to do something without any examples. Multi-shot. Ask an AI 110 00:06:24,540 --> 00:06:28,680 S1: to do something and provide multiple examples. Prompt injection tricking 111 00:06:28,680 --> 00:06:32,990 S1: an AI into doing something bad. Jailbreaking, bypassing security controls 112 00:06:32,990 --> 00:06:35,659 S1: to get full execution ability, or at least as full 113 00:06:35,660 --> 00:06:40,400 S1: as possible, doesn't necessarily need to be full. AGI general 114 00:06:40,400 --> 00:06:44,510 S1: AI smart enough to replace an $80,000 white collar worker, 115 00:06:44,510 --> 00:06:49,730 S1: and ASI, which is superintelligence. General AI that's smarter and 116 00:06:49,730 --> 00:06:52,640 S1: or more capable than any human. So I think that's 117 00:06:52,640 --> 00:06:55,520 S1: a pretty clean list for thinking about these concepts. And 118 00:06:55,520 --> 00:06:58,010 S1: now I want to go into more detail for each. 119 00:06:58,010 --> 00:07:01,460 S1: So I, I is technology that does cognitive tasks or 120 00:07:01,460 --> 00:07:04,520 S1: work that could previously only be done by humans. So 121 00:07:04,520 --> 00:07:07,609 S1: cognitive tasks or work and I might clean that up 122 00:07:07,610 --> 00:07:12,110 S1: at some point and just make it work or cognitive tasks. Anyway, 123 00:07:12,110 --> 00:07:14,480 S1: lots of different ways to define AI. So this is 124 00:07:14,480 --> 00:07:17,120 S1: going to be controversial. And this is what I talked about. 125 00:07:17,120 --> 00:07:19,530 S1: Well yeah of course I can do this. But it 126 00:07:19,530 --> 00:07:21,660 S1: still can't do this and it probably never will be 127 00:07:21,660 --> 00:07:24,150 S1: able to. And then the narrator voice says, yeah, that 128 00:07:24,150 --> 00:07:26,730 S1: happened seven months later. And this is why I like 129 00:07:26,730 --> 00:07:30,270 S1: the definition is because it moves over time. Machine learning 130 00:07:30,300 --> 00:07:33,420 S1: a subset of AI that enables a system to learn 131 00:07:33,420 --> 00:07:36,900 S1: from data alone rather than needing to be explicitly programmed. 132 00:07:36,900 --> 00:07:41,460 S1: This is super exciting to me as a definition. Actually, 133 00:07:41,460 --> 00:07:45,170 S1: as a concept. I can't believe the program is the same, 134 00:07:45,170 --> 00:07:48,590 S1: but you change the data and the program gets smarter. 135 00:07:48,620 --> 00:07:51,650 S1: That blows my mind and it's like the most important. 136 00:07:51,650 --> 00:07:54,440 S1: In fact, if you want to go tighter learning from 137 00:07:54,440 --> 00:07:57,830 S1: data alone or a technology system that's able to learn 138 00:07:57,830 --> 00:08:00,560 S1: from data alone, like lots of different ways you can 139 00:08:00,560 --> 00:08:04,070 S1: skin this thing. Prompt engineering. Prompt engineering is the art 140 00:08:04,070 --> 00:08:07,680 S1: and science of using language, usually text, to get AI 141 00:08:07,710 --> 00:08:10,200 S1: to do precisely what you want it to do. Yeah, 142 00:08:10,230 --> 00:08:12,900 S1: a lot of people think prompt engineering is like this 143 00:08:12,900 --> 00:08:16,080 S1: really special thing. Others think it's not important at all. 144 00:08:16,080 --> 00:08:19,980 S1: I think it's absolutely an art and a science because 145 00:08:19,980 --> 00:08:23,640 S1: it's more about clear thinking than about text itself. Just 146 00:08:23,640 --> 00:08:26,850 S1: like writing, the best prompt engineer is the same. It 147 00:08:26,850 --> 00:08:29,850 S1: comes from deeply understanding the problem and being able to 148 00:08:29,850 --> 00:08:32,650 S1: break out your instructions to the AI in a very 149 00:08:32,650 --> 00:08:36,040 S1: methodical and clear way. So the clearer you can think 150 00:08:36,040 --> 00:08:40,089 S1: and the clearer you can think about problems and describe methodologies. 151 00:08:40,090 --> 00:08:42,130 S1: That's that's what's going to make you really good at 152 00:08:42,130 --> 00:08:45,250 S1: prompt engineering. Don't think of it as an AI thing. 153 00:08:45,250 --> 00:08:49,480 S1: Think of it as a thinking thing and a writing thing. Retrieval. 154 00:08:49,480 --> 00:08:53,439 S1: Augmented generation Rag is the process of taking large quantities 155 00:08:53,440 --> 00:08:55,970 S1: of data which are either too large or too expensive 156 00:08:55,970 --> 00:08:58,490 S1: to put in a prompt directly, and making that data 157 00:08:58,490 --> 00:09:02,990 S1: usable as vectorized embeddings to AI at runtime. And I 158 00:09:02,990 --> 00:09:06,230 S1: say it's important to understand that Rag is a hack 159 00:09:06,230 --> 00:09:09,260 S1: that solves a specific problem, which is that people and 160 00:09:09,260 --> 00:09:13,130 S1: companies have way too much data gigabytes, terabytes, petabytes of 161 00:09:13,130 --> 00:09:15,650 S1: data that they want their AI to be aware of 162 00:09:15,650 --> 00:09:19,160 S1: when performing tasks. But I can only handle small amounts 163 00:09:19,160 --> 00:09:22,240 S1: of that data per interaction. And that's what Rag is for. 164 00:09:22,240 --> 00:09:24,670 S1: So the solution we come up with is to use 165 00:09:24,670 --> 00:09:28,600 S1: embeddings and vector databases to encode relevant information and send 166 00:09:28,600 --> 00:09:31,059 S1: that along with the prompts. And it's not really clear 167 00:09:31,059 --> 00:09:33,670 S1: what the successor is going to be. But one option 168 00:09:33,670 --> 00:09:37,000 S1: is to just have massive context windows and put the 169 00:09:37,000 --> 00:09:39,700 S1: raw content in there. But then you have cost problems, 170 00:09:39,700 --> 00:09:41,500 S1: so you would need that to be very cheap for 171 00:09:41,530 --> 00:09:45,050 S1: that to work. Agents An AI agent is an AI 172 00:09:45,080 --> 00:09:49,040 S1: component that interprets instructions and takes on more of the 173 00:09:49,040 --> 00:09:53,930 S1: work in a total AI workflow than just LLM response, 174 00:09:53,929 --> 00:10:00,229 S1: for example, executing functions, performing data lookups, etc. before passing 175 00:10:00,230 --> 00:10:04,460 S1: on results. Yeah, so there are lots of different definitions 176 00:10:04,460 --> 00:10:07,589 S1: for agent. I think the trick is to go back 177 00:10:07,590 --> 00:10:12,000 S1: to the Latin. The agents or agents, which is to 178 00:10:12,000 --> 00:10:15,510 S1: do or to act. So I think eventually this is 179 00:10:15,510 --> 00:10:17,730 S1: going to be like an AI component that has its 180 00:10:17,730 --> 00:10:20,700 S1: own mission and goals and can use its resources and 181 00:10:20,700 --> 00:10:24,150 S1: capabilities to accomplish them in a self-directed way. That is 182 00:10:24,150 --> 00:10:28,229 S1: the true nature of agent or agents from the Latin. 183 00:10:28,230 --> 00:10:31,170 S1: But I think in this current version, I think the 184 00:10:31,170 --> 00:10:33,700 S1: best way to say it is that anything that acts 185 00:10:33,700 --> 00:10:36,760 S1: on behalf of the mission and something that performs multiple 186 00:10:36,760 --> 00:10:40,420 S1: steps towards the final goal, and that's why I'm using 187 00:10:40,420 --> 00:10:43,240 S1: this definition for agent chain of thought, a way of 188 00:10:43,240 --> 00:10:45,520 S1: interacting with AI in which you don't just say what 189 00:10:45,520 --> 00:10:47,500 S1: you want, but you give the steps that you would 190 00:10:47,500 --> 00:10:50,140 S1: take to accomplish the task. So you're basically teaching it 191 00:10:50,140 --> 00:10:52,959 S1: how you think, which which is essentially teaching it how 192 00:10:52,960 --> 00:10:55,569 S1: to think like a human. And then it uses that 193 00:10:55,570 --> 00:10:58,199 S1: template to sort of do that every time it's going 194 00:10:58,200 --> 00:11:00,510 S1: to solve the same type of problem. And to me, 195 00:11:00,510 --> 00:11:02,609 S1: it's kind of an example of what we talked about 196 00:11:02,610 --> 00:11:06,540 S1: in prompt engineering, clear thinking. It's walking the AI through 197 00:11:06,540 --> 00:11:09,390 S1: how you think when you're solving a problem yourself. Prompt 198 00:11:09,390 --> 00:11:12,810 S1: injection a method of using language, usually text, to get 199 00:11:12,809 --> 00:11:14,970 S1: AI to do something it's not supposed to do. And 200 00:11:14,970 --> 00:11:17,819 S1: a lot of people confuse prompt injection with jailbreaking. And 201 00:11:17,820 --> 00:11:19,800 S1: I think the best way to think about this. And 202 00:11:19,809 --> 00:11:22,990 S1: thanks to Jason Haddox for talking through this with me, 203 00:11:22,990 --> 00:11:26,260 S1: because he had a very similar definition, uh, is to 204 00:11:26,260 --> 00:11:30,069 S1: say prompt injection is a method. It's an attack avenue. 205 00:11:30,070 --> 00:11:33,760 S1: It's not a payload, whereas jailbreaking is a goal. We're 206 00:11:33,760 --> 00:11:37,360 S1: trying to bypass the security on a system. Prompt injection 207 00:11:37,360 --> 00:11:40,300 S1: is not bypassing the security on a system. Well, it's 208 00:11:40,300 --> 00:11:43,240 S1: a method for doing that. It's a way of tricking 209 00:11:43,240 --> 00:11:45,829 S1: the eye into doing something it's not supposed to do. 210 00:11:45,830 --> 00:11:49,309 S1: But that could be telling a joke that it's not 211 00:11:49,309 --> 00:11:51,830 S1: supposed to tell. That could be showing a picture it's 212 00:11:51,830 --> 00:11:54,200 S1: not supposed to show. It doesn't mean it's bypassing the 213 00:11:54,200 --> 00:11:57,109 S1: security on a system so you can execute commands. So 214 00:11:57,110 --> 00:11:59,510 S1: it could be data exfil, it could be code execution, 215 00:11:59,510 --> 00:12:02,780 S1: it could be image generation. It could be anything. So easiest, 216 00:12:02,780 --> 00:12:04,880 S1: cleanest way, at least in my mind, to think about this. 217 00:12:04,880 --> 00:12:08,140 S1: Prompt injection is a method. Jailbreaking is a goal. All right. 218 00:12:08,140 --> 00:12:11,440 S1: We talked about jailbreaking going to past that one. Artificial 219 00:12:11,440 --> 00:12:15,459 S1: general intelligence. The ability for an AI system, whether a model, 220 00:12:15,460 --> 00:12:18,220 S1: a product or a system to fully do the job 221 00:12:18,220 --> 00:12:21,339 S1: of an average US based knowledge worker. In other words, 222 00:12:21,340 --> 00:12:24,069 S1: it's not just competent at doing a specific thing, which 223 00:12:24,070 --> 00:12:28,689 S1: is called narrow AI oftentimes, but many different things, which 224 00:12:28,690 --> 00:12:31,750 S1: is why it's called general. So basically it's some combination 225 00:12:31,750 --> 00:12:35,500 S1: of sufficiently general and sufficiently competent and the amounts of 226 00:12:35,500 --> 00:12:38,260 S1: which are, you know, going to be debated so generally 227 00:12:38,260 --> 00:12:42,370 S1: competent at many other cognitive tasks, sample efficient. So we 228 00:12:42,370 --> 00:12:45,520 S1: can learn quickly from like one example or a low 229 00:12:45,520 --> 00:12:47,949 S1: number of examples. And the system should be able to 230 00:12:47,950 --> 00:12:51,880 S1: apply new information, concepts and skills that learns to additional 231 00:12:51,880 --> 00:12:55,930 S1: new problems. So it's not just generally intelligent from its training, 232 00:12:55,929 --> 00:12:58,859 S1: but when it learns new things, it's generally intelligent at 233 00:12:58,860 --> 00:13:01,470 S1: applying those new things that learn to everything else in 234 00:13:01,470 --> 00:13:05,370 S1: the future. And I like this definition focusing on the 235 00:13:05,370 --> 00:13:10,890 S1: worker replacing a worker, because I think this is arguably 236 00:13:10,890 --> 00:13:14,520 S1: what matters most to humans when we talk about AI discussion, 237 00:13:14,520 --> 00:13:17,730 S1: which is the future of humans in a world of AI, 238 00:13:17,760 --> 00:13:22,000 S1: regular people don't care about processing speed or model size 239 00:13:22,000 --> 00:13:24,429 S1: or model weights or any of that crap. What they 240 00:13:24,429 --> 00:13:26,560 S1: care about is if and when any of this is 241 00:13:26,559 --> 00:13:29,890 S1: going to actually affect them. And that means job replacement mostly, 242 00:13:29,890 --> 00:13:32,830 S1: or at least that's the most important thing. So these 243 00:13:32,830 --> 00:13:36,760 S1: are the levels that I see within that job replacement 244 00:13:36,760 --> 00:13:40,270 S1: of this average white collar worker. So first level is 245 00:13:40,270 --> 00:13:43,750 S1: better but with significant drawbacks. And I'm not going to 246 00:13:43,750 --> 00:13:46,340 S1: go through all of these. But it's like the interface, 247 00:13:46,340 --> 00:13:50,420 S1: the language, the confusion, the errors, flexibility, basically. Basically, real 248 00:13:50,420 --> 00:13:54,530 S1: world work environments are kludgy and lame and they change 249 00:13:54,530 --> 00:13:57,860 S1: all the time. And it's basically a giant mess. This 250 00:13:57,860 --> 00:14:01,160 S1: level one is basically saying, yes, we brought in an 251 00:14:01,160 --> 00:14:04,220 S1: AI worker to replace that person, or we didn't hire 252 00:14:04,220 --> 00:14:07,280 S1: a person. We used an AI worker instead for for 253 00:14:07,280 --> 00:14:10,020 S1: growth if you want to be nice. So that's what happened. 254 00:14:10,020 --> 00:14:14,220 S1: We brought in this level one AGI one. And here's 255 00:14:14,220 --> 00:14:17,939 S1: the problem. It's kludgy. So so the problem is that 256 00:14:17,940 --> 00:14:20,400 S1: they're doing the work. But you have to clean up 257 00:14:20,400 --> 00:14:24,450 S1: after them. So the interface proprietary cumbersome the language that 258 00:14:24,450 --> 00:14:27,240 S1: you have to use. It's still kind of like AI language. 259 00:14:27,240 --> 00:14:30,510 S1: It's almost like you're prompting. You're not just normally talking 260 00:14:30,510 --> 00:14:33,390 S1: to them. Um, it gets confused a decent amount of 261 00:14:33,390 --> 00:14:36,330 S1: the time. There are frequent mistakes which need to be 262 00:14:36,330 --> 00:14:39,630 S1: fixed by humans, and if you want to change what 263 00:14:39,630 --> 00:14:43,229 S1: this thing is doing, it needs to be largely retooled. 264 00:14:43,800 --> 00:14:46,530 S1: Not not like in a major way, but you've got 265 00:14:46,530 --> 00:14:48,300 S1: to have a conversation with it. You've got to change 266 00:14:48,300 --> 00:14:51,030 S1: some documentation, you've got to do some stuff. So going 267 00:14:51,030 --> 00:14:54,510 S1: back to the top level of this one better but 268 00:14:54,510 --> 00:14:57,660 S1: with significant drawbacks. So it is right at the line 269 00:14:57,660 --> 00:14:59,850 S1: of being too much work and not good enough to 270 00:14:59,850 --> 00:15:05,670 S1: replace a human right. That's the trick. It's just barely better. Barely. 271 00:15:05,670 --> 00:15:08,040 S1: And sometimes you can't tell if it's better or not. 272 00:15:08,040 --> 00:15:11,190 S1: It might be too much work. Next one. It removes 273 00:15:11,190 --> 00:15:13,890 S1: a lot of those problems. So it's like in the middle. 274 00:15:13,890 --> 00:15:20,430 S1: It's like decent, competent but imperfect. So mostly normal workflows, 275 00:15:20,440 --> 00:15:24,670 S1: mostly human with exceptions, doesn't get confused very often. Fewer 276 00:15:24,670 --> 00:15:27,400 S1: mistakes and the mistakes aren't as bad and it just 277 00:15:27,400 --> 00:15:30,760 S1: decently well to new direction from leaders. So that's like 278 00:15:30,760 --> 00:15:36,280 S1: the middle level. And then AGI three is full worker replacement. 279 00:15:36,280 --> 00:15:42,580 S1: So this can replace this average 80 K worker in 2022. 280 00:15:42,610 --> 00:15:46,100 S1: White collar worker in the US. It can fully replace them. 281 00:15:46,100 --> 00:15:49,130 S1: So when you talk to them, it's you talk to 282 00:15:49,130 --> 00:15:51,440 S1: them just like any other employee. You could text them, 283 00:15:51,440 --> 00:15:53,630 S1: you could voice them, you could do video with them. 284 00:15:53,630 --> 00:15:55,910 S1: It's just like an employee. You talk to them exactly 285 00:15:55,910 --> 00:15:59,450 S1: like an employee. They're confused the same amount or less 286 00:15:59,450 --> 00:16:02,990 S1: than the $80,000 employee. Keep in mind, humans have all 287 00:16:02,990 --> 00:16:05,750 S1: these problems too, right? Humans also get confused and need 288 00:16:05,750 --> 00:16:07,760 S1: to be retasked or whatever. So it's not like we're 289 00:16:07,760 --> 00:16:10,700 S1: competing with perfection here. Same or fewer mistakes than an 290 00:16:10,700 --> 00:16:14,720 S1: $80,000 employee. Employees also make mistakes and just as flexible 291 00:16:14,720 --> 00:16:18,560 S1: or more than $80,000 employee employees also have trouble like 292 00:16:18,560 --> 00:16:21,020 S1: completely being retests. It's like I thought it was working 293 00:16:21,020 --> 00:16:23,030 S1: on that. What are you doing? Like, so this is 294 00:16:23,030 --> 00:16:26,300 S1: basically saying it's just as good as them or or better. 295 00:16:26,300 --> 00:16:31,160 S1: So this is the top level of complete worker replacement. 296 00:16:31,160 --> 00:16:32,930 S1: I think this is going to take a decent amount 297 00:16:32,930 --> 00:16:35,060 S1: of time. And then the next question is like, okay, 298 00:16:35,060 --> 00:16:38,420 S1: but what if it's like that? It's a full replacement 299 00:16:38,420 --> 00:16:41,210 S1: for that worker. So it's AGI three, but with an 300 00:16:41,210 --> 00:16:45,410 S1: IQ of 150. Well that's that's really powerful. And I'm 301 00:16:45,410 --> 00:16:48,950 S1: not working on that axis. I'm only working on how 302 00:16:48,950 --> 00:16:52,190 S1: good of a worker are they. And that's the three 303 00:16:52,190 --> 00:16:55,490 S1: levels for AGI. Okay. Let's do ASI a level of 304 00:16:55,490 --> 00:16:59,670 S1: general AI that's smarter and more capable than any human 305 00:16:59,670 --> 00:17:02,340 S1: that's ever lived. So this is interesting for a number 306 00:17:02,340 --> 00:17:05,670 S1: of reasons. One, it's a threshold that sits above AGI 307 00:17:05,670 --> 00:17:08,580 S1: and people don't really agree on that definition. And the 308 00:17:08,580 --> 00:17:11,190 S1: second thing is, at least as I'm defining it, it 309 00:17:11,190 --> 00:17:15,000 S1: has a massive range. And third, it blends with AGI 310 00:17:15,000 --> 00:17:19,470 S1: because AGI just really means general incompetent. And ASI is 311 00:17:19,470 --> 00:17:24,619 S1: also general incompetent. So really ASI is also AGI. It's 312 00:17:24,619 --> 00:17:28,550 S1: just uh, at a level beyond any human. So AGI 313 00:17:28,550 --> 00:17:32,000 S1: is replacement for a human worker making SDK in the 314 00:17:32,000 --> 00:17:36,770 S1: US in 2022, which is pretty. I present AI and 315 00:17:36,770 --> 00:17:41,960 S1: ASI is more smart and more competent than any human, 316 00:17:41,960 --> 00:17:44,359 S1: and the model I'm using there is like John von 317 00:17:44,359 --> 00:17:50,440 S1: Neumann because unlike Einstein and Newton, he was, uh, very broad. 318 00:17:50,440 --> 00:17:54,490 S1: He was a generalist. Right? So he did game theory, physics, computing, 319 00:17:54,490 --> 00:17:57,280 S1: lots of other stuff. But I don't think, uh, being 320 00:17:57,280 --> 00:18:00,250 S1: smarter is the only thing that matters, uh, because I 321 00:18:00,250 --> 00:18:02,469 S1: think it comes down to multiple things. So if I 322 00:18:02,470 --> 00:18:05,050 S1: look at all the different things that I'm putting into AC, 323 00:18:05,050 --> 00:18:08,410 S1: it's like your ability to model abstractions of reality at 324 00:18:08,410 --> 00:18:14,080 S1: these different levels. It's the ability to take action, perceive, understand, improve, solve, create, destroy. 325 00:18:14,080 --> 00:18:17,110 S1: And then it's like, okay, what fields are we innovating in? 326 00:18:17,109 --> 00:18:21,490 S1: What problems are we able to solve and what scale 327 00:18:21,520 --> 00:18:25,120 S1: are we actually working? You know, thinking about, I guess 328 00:18:25,119 --> 00:18:28,030 S1: I should add universe or whatever, but all the way 329 00:18:28,030 --> 00:18:31,629 S1: from Quark to Galaxy is like the scale. So what's 330 00:18:31,630 --> 00:18:33,830 S1: really cool about this is you can now turn these 331 00:18:33,830 --> 00:18:37,369 S1: into functional phrases. So you could say like an AI 332 00:18:37,400 --> 00:18:43,220 S1: capable of curing aging by creating new chemicals that affect DNA. Okay. 333 00:18:43,220 --> 00:18:46,670 S1: So we have a problem here creating new chemicals. That's 334 00:18:46,670 --> 00:18:50,810 S1: like a net new thing, right? So that you're seeing this, 335 00:18:50,810 --> 00:18:54,290 S1: you know, verbs and nouns here, maintaining a full city 336 00:18:54,290 --> 00:18:58,619 S1: by monitoring and adjusting all public resources in real time, 337 00:18:58,619 --> 00:19:01,230 S1: and the AI capable of taking over a country by 338 00:19:01,230 --> 00:19:05,609 S1: manufacturing a drone army and a new energy based weapon. 339 00:19:05,609 --> 00:19:08,010 S1: So this is like new science here, and an AI 340 00:19:08,040 --> 00:19:12,450 S1: capable of faster than light travel by discovering completely new physics. 341 00:19:12,450 --> 00:19:15,810 S1: So that's like a level above, right? So let's look 342 00:19:15,810 --> 00:19:19,919 S1: at the different levels here. AC one superior an AI 343 00:19:19,950 --> 00:19:24,290 S1: capable of making incremental improvements in multiple fields and managing 344 00:19:24,290 --> 00:19:28,250 S1: up to city size entities on its own. So here 345 00:19:28,250 --> 00:19:30,230 S1: are the things that it has. Smarter and more capable 346 00:19:30,230 --> 00:19:33,800 S1: than any human on many topics. Able to move progress 347 00:19:33,800 --> 00:19:37,939 S1: forward in multiple scientific fields. Able to recommend novel solutions 348 00:19:37,940 --> 00:19:41,570 S1: to many of our main problems. Able to copy and 349 00:19:41,570 --> 00:19:45,470 S1: surpass the creativity of many top artists. Able to fully 350 00:19:45,470 --> 00:19:49,379 S1: manage a large company or city itself and keep, keep 351 00:19:49,380 --> 00:19:53,490 S1: in mind, copy and surpass some of the best artists. Now, 352 00:19:53,490 --> 00:19:57,000 S1: what this really means is like in a superhuman way, right? 353 00:19:57,000 --> 00:20:00,540 S1: So it's not just like better in one area. It's like, okay, 354 00:20:00,540 --> 00:20:04,800 S1: it's Eminem, but it's way better than Eminem. It's Kendrick Lamar, 355 00:20:04,800 --> 00:20:08,880 S1: but five times better, right? Same with these types of things. 356 00:20:08,880 --> 00:20:13,420 S1: Novel solutions, moving progress forward, which obviously no human has done. 357 00:20:13,420 --> 00:20:16,300 S1: So they're doing it better, but it doesn't have these. 358 00:20:16,300 --> 00:20:20,710 S1: It can't create. Net new physics or net new material science. 359 00:20:20,710 --> 00:20:23,980 S1: It's unable to fundamentally improve itself by orders of magnitude. 360 00:20:23,980 --> 00:20:28,659 S1: So next phase dominant okay. First one was superior. Next 361 00:20:28,660 --> 00:20:31,389 S1: one is dominant. This one has all of AC one. 362 00:20:31,390 --> 00:20:34,030 S1: But it's also able to completely change how we see 363 00:20:34,040 --> 00:20:37,970 S1: multiple fields able to completely solve most of our current problems. 364 00:20:38,000 --> 00:20:41,870 S1: Able to fully manage a country by itself. So this 365 00:20:41,869 --> 00:20:44,270 S1: is a huge jump over the first one, able to 366 00:20:44,270 --> 00:20:48,260 S1: fundamentally improve itself by orders of magnitude, but still can't create. 367 00:20:48,260 --> 00:20:50,600 S1: Net new physics or run an entire planet. And then 368 00:20:50,600 --> 00:20:54,980 S1: you have AC three, another massive jump AI capable of creating. 369 00:20:54,980 --> 00:21:00,490 S1: Net new physics, completely new materials manipulation of near fundamental 370 00:21:00,490 --> 00:21:04,960 S1: reality or fundamental reality. If it's possible to do that 371 00:21:04,960 --> 00:21:08,770 S1: and can run an entire planet. And keep in mind, 372 00:21:08,770 --> 00:21:12,280 S1: I might add ASI for um, if I have a 373 00:21:12,280 --> 00:21:17,350 S1: like a clear picture of what that looks like. Um, Yoshi. No. Rishi, 374 00:21:17,380 --> 00:21:21,879 S1: my buddy Rishi mentioned, um, what about someone? An AI 375 00:21:21,880 --> 00:21:24,919 S1: that's smarter than all of humanity put together? And I 376 00:21:24,920 --> 00:21:29,330 S1: was like, well, isn't that confusing knowledge with intelligence and reason? 377 00:21:29,330 --> 00:21:31,760 S1: But I haven't heard back from him yet. Other things 378 00:21:31,760 --> 00:21:34,940 S1: to consider here. So this one adds the ability to 379 00:21:34,940 --> 00:21:38,840 S1: modify reality at a fundamental or near fundamental level, able 380 00:21:38,840 --> 00:21:42,830 S1: to manage an entire planet simultaneously. And maybe its primary 381 00:21:42,830 --> 00:21:46,800 S1: concerns become like sun expansion, which is going to like 382 00:21:46,800 --> 00:21:50,010 S1: eat up the earth or populating the galaxy and beyond, 383 00:21:50,010 --> 00:21:52,500 S1: or the heat death of the universe, which is like 384 00:21:52,500 --> 00:21:55,680 S1: escaping this reality. Yeah, or breaking out of the simulation. 385 00:21:55,680 --> 00:21:58,439 S1: I should add that. So as a summary, I terms 386 00:21:58,440 --> 00:22:01,830 S1: are confusing. It's nice to have simple, practical versions. It's 387 00:22:01,830 --> 00:22:05,100 S1: useful to crystallize your own definitions on paper, both for 388 00:22:05,100 --> 00:22:07,410 S1: your own reference and to see if your definitions are 389 00:22:07,410 --> 00:22:11,620 S1: consistent with each other. And I think these eye definitions 390 00:22:11,619 --> 00:22:15,850 S1: work best when they're human focused and practically worded, because 391 00:22:15,850 --> 00:22:18,129 S1: it is very easy to try to be technical with 392 00:22:18,130 --> 00:22:21,910 S1: these definitions, and you instantly open up like this giant 393 00:22:21,910 --> 00:22:26,080 S1: mess the moment you go a level down in technicality. 394 00:22:26,080 --> 00:22:29,949 S1: And what you'll find is if you get 20 eye 395 00:22:29,950 --> 00:22:33,730 S1: experts in in a room and a bunch of ML experts, 396 00:22:33,730 --> 00:22:36,660 S1: they've been doing this 30 years and you ask them 397 00:22:36,660 --> 00:22:39,600 S1: their opinions. They don't agree. The textbooks don't agree. The 398 00:22:39,600 --> 00:22:42,690 S1: experts don't agree. The moment you go a layer down, 399 00:22:42,690 --> 00:22:47,040 S1: it's a giant mess. And that giant mess stops conversations 400 00:22:47,040 --> 00:22:49,860 S1: from happening because they end up, they'll get two hours 401 00:22:49,859 --> 00:22:51,840 S1: into a podcast and be like, oh, that's what you 402 00:22:51,840 --> 00:22:54,060 S1: meant by I. That's not what I mean by I. 403 00:22:54,060 --> 00:22:57,780 S1: And I'm like, okay, let's stop messing with let's stop 404 00:22:57,780 --> 00:23:00,790 S1: wasting two hours in a conversation because we're not talking 405 00:23:00,790 --> 00:23:04,480 S1: about the same thing. Let's level up to an abstraction 406 00:23:04,480 --> 00:23:07,719 S1: level where it's useful in a conversation. That's why it's 407 00:23:07,720 --> 00:23:10,810 S1: called real world AI definition. All right. So that's that one. 408 00:23:10,810 --> 00:23:13,960 S1: All right. Stories. Cloudflare has a new free tool to 409 00:23:13,960 --> 00:23:17,170 S1: stop bots from scraping websites. Remember I told you about 410 00:23:17,170 --> 00:23:20,140 S1: Cloudflare how they just do like the coolest stuff and 411 00:23:20,140 --> 00:23:23,450 S1: they just Cloudflare is a beast. I am telling you. 412 00:23:23,450 --> 00:23:27,440 S1: They just they see a problem that is annoying everybody 413 00:23:27,440 --> 00:23:29,780 S1: and they go and just pour a little bit of 414 00:23:29,780 --> 00:23:32,780 S1: like cement in that seam or that crack or that 415 00:23:32,780 --> 00:23:35,659 S1: hole or whatever, and they're just like, we're going to 416 00:23:35,660 --> 00:23:38,150 S1: solve this. And if no one else does it, they're 417 00:23:38,150 --> 00:23:40,220 S1: just going to be the solution for this. Like they're 418 00:23:40,220 --> 00:23:42,680 S1: doing this with Captcha. They're doing this now with AI 419 00:23:42,680 --> 00:23:46,550 S1: scraping like somebody over there is going, this is going 420 00:23:46,550 --> 00:23:48,330 S1: to be a problem in the future or this is 421 00:23:48,330 --> 00:23:51,060 S1: already becoming a problem. Let's get there first. And I 422 00:23:51,060 --> 00:23:54,030 S1: massively respect it. And before too long, the whole whole 423 00:23:54,030 --> 00:23:57,030 S1: internet's going to be Cloudflare, because over the course of time, 424 00:23:57,030 --> 00:24:00,570 S1: they've made like 500 of these tools at some point. Right. 425 00:24:00,570 --> 00:24:03,930 S1: And it's like all those random little tools that they 426 00:24:03,930 --> 00:24:07,560 S1: made become the structure of the internet. Twilio says, uh, 427 00:24:07,560 --> 00:24:12,010 S1: somebody's got phone numbers because of an unsecured auth API endpoint, 428 00:24:12,010 --> 00:24:16,359 S1: 33 million phone numbers. Russian experts say they fully analyzed 429 00:24:16,359 --> 00:24:21,700 S1: the structure of the American attack Miss Atacms missile, and 430 00:24:21,700 --> 00:24:25,630 S1: they believe that's going to help them fighting Ukrainian missiles, 431 00:24:25,630 --> 00:24:29,590 S1: which is not good. Thanks to Tynes for sponsoring. North 432 00:24:29,590 --> 00:24:32,560 S1: Korea switched its state TV broadcast from a Chinese satellite 433 00:24:32,560 --> 00:24:35,580 S1: to a Russian one, and now fewer people can watch 434 00:24:35,609 --> 00:24:39,449 S1: Hezbollah launch over 200 rockets and drones at Israel after 435 00:24:39,450 --> 00:24:43,920 S1: Israel killed a senior Hezbollah official. Thanks to nudge security 436 00:24:43,920 --> 00:24:48,240 S1: for sponsoring. US intelligence community is diving into generative AI 437 00:24:48,270 --> 00:24:55,560 S1: to enhance intelligence operations, using it for search, discovery, counterargument generation. Uh, 438 00:24:55,560 --> 00:24:58,440 S1: I'm going to make so much Intel stuff I cannot 439 00:24:58,440 --> 00:25:02,000 S1: wait to build that product in substrate. I cannot wait. 440 00:25:02,000 --> 00:25:05,750 S1: There's some analysis saying I cannot be funny from Anya, 441 00:25:05,750 --> 00:25:10,550 S1: Jeremy Greenwald, and I think I disagree. And here's an example. 442 00:25:10,550 --> 00:25:14,660 S1: My buddy Joseph posted this one. It's a glyph app 443 00:25:14,660 --> 00:25:18,440 S1: and it makes memes. Look at this bug bounty, bro. 444 00:25:18,440 --> 00:25:22,580 S1: Like I legitimately laughed at some of these. Show me. 445 00:25:23,270 --> 00:25:27,380 S1: Show you my leet skills. Use the inspect element I 446 00:25:27,380 --> 00:25:31,760 S1: only hack ethically dos school website in eighth grade. I mean, 447 00:25:31,760 --> 00:25:35,330 S1: first of all, these are real. They're good in terms 448 00:25:35,330 --> 00:25:37,400 S1: of like they actually apply to the field that they're 449 00:25:37,400 --> 00:25:41,389 S1: making fun of. And, uh, my POC is flawless. It's 450 00:25:41,390 --> 00:25:45,620 S1: a Rick roll link. I generated this. Okay. I've been 451 00:25:45,619 --> 00:25:48,450 S1: saying for a while that being funny is a major 452 00:25:48,450 --> 00:25:53,940 S1: milestone in human AI. In AI becoming real. And just 453 00:25:53,940 --> 00:25:56,159 S1: like a lot of other things in AI, there are 454 00:25:56,160 --> 00:25:57,930 S1: so many people who are like, oh yeah, it can. 455 00:25:57,930 --> 00:25:59,760 S1: It could tell a cat from a dog, but it's 456 00:25:59,760 --> 00:26:02,610 S1: never going to make me laugh. Whatever the stuff's been 457 00:26:02,609 --> 00:26:06,480 S1: out for like 18 minutes, this just happened at the 458 00:26:06,480 --> 00:26:09,990 S1: end of 2022. Like this is still day one. I'm 459 00:26:09,990 --> 00:26:12,859 S1: telling you right now, there will be funny I very 460 00:26:12,859 --> 00:26:15,380 S1: soon and this is the first version of it. I mean, 461 00:26:15,380 --> 00:26:19,370 S1: this is already funny. I technically the real standard is 462 00:26:19,369 --> 00:26:22,550 S1: going to be a stand up routine, a deepfake of 463 00:26:22,550 --> 00:26:26,690 S1: a face, or actually a deepfake. Okay, I am watching 464 00:26:26,690 --> 00:26:30,620 S1: a stage. There is a deepfake, uh, person standing there. 465 00:26:30,950 --> 00:26:32,930 S1: It doesn't even have to convince me. It could be 466 00:26:32,930 --> 00:26:36,570 S1: like uncanny valley. But if there's a deep fake person 467 00:26:36,570 --> 00:26:40,740 S1: standing there talking, doing a stand up routine and it 468 00:26:40,740 --> 00:26:43,050 S1: makes me laugh the way it makes me laugh when 469 00:26:43,050 --> 00:26:46,590 S1: I watch other stand ups, that is a major milestone. 470 00:26:46,590 --> 00:26:51,810 S1: That is extraordinary. I don't know how long that's going to. Oh, 471 00:26:51,840 --> 00:26:54,330 S1: I should do that on a prediction market. Yeah, I'm 472 00:26:54,330 --> 00:26:56,340 S1: curious about that. I'm guessing like a year and a 473 00:26:56,340 --> 00:26:58,290 S1: half to two years. I think it's going to be 474 00:26:58,290 --> 00:27:01,480 S1: a little while because that one's really, really difficult. But 475 00:27:01,480 --> 00:27:04,510 S1: it could be, honestly, it could be opus. It could 476 00:27:04,510 --> 00:27:07,480 S1: be opus 35 or GPT five. It could happen in 477 00:27:07,480 --> 00:27:10,300 S1: this next iteration, which is going to be around the 478 00:27:10,300 --> 00:27:13,030 S1: end of this year or beginning of next year. Okay. 479 00:27:13,030 --> 00:27:16,090 S1: Nvidia is set to make 12 billion by selling over 480 00:27:16,090 --> 00:27:21,100 S1: a million HG, H20 GPUs. That sounded impressive. I thought 481 00:27:21,100 --> 00:27:23,080 S1: it was like, how are you going to send those? 482 00:27:23,080 --> 00:27:25,609 S1: I thought we had export controls, but these are within 483 00:27:25,609 --> 00:27:28,340 S1: the export controls, so they must be like pretty nerfed. 484 00:27:28,340 --> 00:27:31,580 S1: Apple might soon announce a deal to bring Google Gemini 485 00:27:31,580 --> 00:27:35,389 S1: to iOS 18, so that means they're working with all three. 486 00:27:35,390 --> 00:27:38,990 S1: They already announced OpenAI in the announcement. They mentioned meta 487 00:27:39,020 --> 00:27:41,600 S1: a while back and now Google. And by the way, 488 00:27:41,600 --> 00:27:47,060 S1: I'm bothered with Apple and also with OpenAI because they 489 00:27:47,060 --> 00:27:50,350 S1: both did demos of things that aren't in the betas. 490 00:27:50,350 --> 00:27:53,619 S1: Some reporting is saying this Siri stuff, and all the 491 00:27:53,619 --> 00:27:56,379 S1: AI stuff won't even be until next spring. And I'm like, 492 00:27:56,380 --> 00:27:58,450 S1: stop showing me demos of stuff that won't be out 493 00:27:58,450 --> 00:28:02,290 S1: for months, if ever. But that's just me being annoyed. 494 00:28:02,290 --> 00:28:05,980 S1: Greece is moving to a six day working week. Weird. Oh, 495 00:28:05,980 --> 00:28:08,980 S1: this is actually for a different story. That's a mistake here. 496 00:28:09,490 --> 00:28:14,570 S1: Do not like. Um, This was about somebody, uh, Apple 497 00:28:14,570 --> 00:28:19,700 S1: putting cameras on AirPods, which I was hoping was going 498 00:28:19,700 --> 00:28:22,490 S1: to be actual cameras. But of course, that's going to 499 00:28:22,490 --> 00:28:26,120 S1: be more headgear that's going to require like, uh, like 500 00:28:26,119 --> 00:28:29,060 S1: a collar or like a neck piece or like a 501 00:28:29,060 --> 00:28:33,350 S1: heavy earphone or headphone because you need to power this, uh, 502 00:28:33,350 --> 00:28:35,750 S1: this camera that's running all the time. So you're going 503 00:28:35,750 --> 00:28:39,000 S1: to have battery issues plus lenses and stuff like that. 504 00:28:39,000 --> 00:28:41,940 S1: But what I really want is I want the AirPods 505 00:28:41,940 --> 00:28:44,670 S1: or this headset or whatever it is, but it's got 506 00:28:44,670 --> 00:28:46,260 S1: to be every day. But I want it to see 507 00:28:46,260 --> 00:28:49,260 S1: behind me. That's super critical. If you look at my 508 00:28:49,260 --> 00:28:51,450 S1: big eye piece, I want it to see in front 509 00:28:51,450 --> 00:28:53,790 S1: of me. I want it to see behind me. It'd 510 00:28:53,820 --> 00:28:55,860 S1: be nice if it saw to the sides and it 511 00:28:55,860 --> 00:28:59,250 S1: could connect them for 360. That would be nice. But 512 00:28:59,250 --> 00:29:02,890 S1: most importantly, in front and behind. And then that all 513 00:29:02,890 --> 00:29:05,410 S1: being processed and it's like, hey, there's a dangerous thing 514 00:29:05,410 --> 00:29:07,300 S1: in front of you. Hey, watch out for this. This 515 00:29:07,300 --> 00:29:10,360 S1: car is coming in your direction. Whatever. Someone's sneaking up 516 00:29:10,360 --> 00:29:14,110 S1: behind you. Be great for women. Like walking at night. 517 00:29:14,230 --> 00:29:17,230 S1: It's like this person is following you. Yeah, but no, 518 00:29:17,230 --> 00:29:21,070 S1: the thing that actually is getting shipped, it's for gesture controls. 519 00:29:21,070 --> 00:29:23,500 S1: So you get like, you know, go like this and 520 00:29:23,500 --> 00:29:26,580 S1: you could do things with your AirPods. That's the rumor. 521 00:29:26,580 --> 00:29:29,340 S1: David Brooks sat down with Steve Bannon to understand his 522 00:29:29,340 --> 00:29:33,180 S1: vision for the global populist movement. Um, number of U.S. 523 00:29:33,180 --> 00:29:36,600 S1: high school graduates is expected to peak in 2025 and 524 00:29:36,600 --> 00:29:40,170 S1: then decline for years. I don't understand this. Schools and 525 00:29:40,170 --> 00:29:43,200 S1: colleges are closing. Faculty members are being laid off. I, 526 00:29:43,410 --> 00:29:46,770 S1: I what? I don't understand why there's going to be 527 00:29:46,770 --> 00:29:50,320 S1: fewer people graduating. Is there a few people going into 528 00:29:50,320 --> 00:29:53,920 S1: the system? Is the population declining? I'm not sure exactly 529 00:29:53,920 --> 00:29:56,350 S1: what's going on here. US dollar just hit a 38 530 00:29:56,350 --> 00:30:00,040 S1: year peak against the yen. Peoples whose eyes dilated more 531 00:30:00,040 --> 00:30:03,580 S1: performed better on tests of working memory, suggesting that pupil 532 00:30:03,580 --> 00:30:05,500 S1: size is linked to how well we can process and 533 00:30:05,500 --> 00:30:10,570 S1: remember information. And this is from Scientific American. The FDA 534 00:30:10,570 --> 00:30:15,410 S1: approved Eli Lilly's Alzheimer's drug Kazuna. Why was that hard 535 00:30:15,410 --> 00:30:20,600 S1: to pronounce? Kazuna. Donanemab is the real name which shows 536 00:30:20,600 --> 00:30:24,800 S1: disease progression, slows disease progression by about a third. These 537 00:30:24,800 --> 00:30:28,430 S1: drug companies are killing it with Ozempic and now Alzheimer's. 538 00:30:28,430 --> 00:30:31,880 S1: High work in progress is killing your business, innovation and morale. 539 00:30:31,910 --> 00:30:35,210 S1: The more tasks you juggle, the slower everything moves. Yeah, 540 00:30:35,210 --> 00:30:39,120 S1: tell me about it. I'm struggling with this right now. Ideas. 541 00:30:39,120 --> 00:30:42,810 S1: The I class gap. There's a separation that is really 542 00:30:42,810 --> 00:30:46,590 S1: freaking me out about AI right now, which is dramatically 543 00:30:46,590 --> 00:30:53,970 S1: increased separation between social groups or classes, socioeconomic classes because 544 00:30:53,970 --> 00:30:57,000 S1: of AI. So basically one small group at the top 545 00:30:57,000 --> 00:31:00,000 S1: of like 1% or whatever, how you want to call 546 00:31:00,000 --> 00:31:02,540 S1: it 5%, We'll use AI to have a staff of 547 00:31:02,540 --> 00:31:07,640 S1: 10,000 executive assistants, tutors and analysts working constantly to run 548 00:31:07,640 --> 00:31:12,170 S1: their businesses, optimize their entire lives, accountants doing their paperwork, 549 00:31:12,170 --> 00:31:16,370 S1: filing their taxes, optimizing everything. Just think about that. A 550 00:31:16,370 --> 00:31:19,100 S1: lot of people, this top 1%, are going to have 551 00:31:19,100 --> 00:31:23,280 S1: like a team of 10 or 50 or 100 or 552 00:31:23,280 --> 00:31:27,450 S1: 10,000 executive assistants, which are basically independent eyes, all working 553 00:31:27,450 --> 00:31:31,380 S1: towards the same goals, optimizing everything in your life. That's 554 00:31:31,380 --> 00:31:34,080 S1: the top 1%. And a lot of other people either 555 00:31:34,080 --> 00:31:36,420 S1: won't be using AI at all, or they'll only be 556 00:31:36,420 --> 00:31:41,640 S1: using it for gaming, watching media porn entertainment, watching videos, 557 00:31:41,640 --> 00:31:47,790 S1: watching Netflix, just wasting time like slightly more efficiently using AI. 558 00:31:47,790 --> 00:31:50,660 S1: And this is largely the same split as between like 559 00:31:50,660 --> 00:31:54,140 S1: voracious readers and everyone else today. But it's going to 560 00:31:54,140 --> 00:31:57,260 S1: be way worse with AI, because AI is going to 561 00:31:57,260 --> 00:32:01,190 S1: lift the top 1% way more. Even more than being 562 00:32:01,190 --> 00:32:05,450 S1: a heavy reader. Discovery self publishing a tech book mma AI. 563 00:32:05,450 --> 00:32:09,440 S1: This guy is predicting MMA fight outcomes with very high 564 00:32:09,440 --> 00:32:13,610 S1: accuracy just using AI, and all he's doing is he's 565 00:32:13,610 --> 00:32:16,610 S1: looking at the the people that they fought in the past. 566 00:32:16,610 --> 00:32:20,510 S1: Sam Parr launched Sam's List, a database of CPAs, accountants 567 00:32:20,510 --> 00:32:24,410 S1: and tax strategists, and he's already made $32,000 in just 568 00:32:24,410 --> 00:32:28,550 S1: two months. And illustrated transformer visual intuitive guide to understanding 569 00:32:28,550 --> 00:32:32,270 S1: how Transformers work. 11 labs has a new voice isolator. 570 00:32:32,270 --> 00:32:37,600 S1: This thing's scary. Rachel Toback, my buddy, talked about this, uh, talking. 571 00:32:37,600 --> 00:32:41,110 S1: I forgot how to pronounce this. Um. Tao te ching. Yeah, 572 00:32:41,470 --> 00:32:43,660 S1: I feel like I should know how to pronounce that. Anyway, 573 00:32:44,020 --> 00:32:47,920 S1: Ursula K Le Guin did a translation of this, which. 574 00:32:47,920 --> 00:32:50,800 S1: I've not read this thing yet. I really need to 575 00:32:50,800 --> 00:32:53,020 S1: read this thing in full. And the recommendation of the 576 00:32:53,020 --> 00:32:57,580 S1: week is ground news. Ground news. This is like the 577 00:32:57,580 --> 00:33:02,180 S1: best website right now for news, especially going into this, uh, 578 00:33:02,180 --> 00:33:08,960 S1: political election series, uh, situation in the US. So it 579 00:33:08,960 --> 00:33:13,550 S1: shows you bias in how stories are covered. In fact, 580 00:33:13,550 --> 00:33:15,170 S1: let me just pull this up. This is the whole 581 00:33:15,410 --> 00:33:19,310 S1: purpose of being on, uh, on video. Look at this thing. Okay. 582 00:33:19,310 --> 00:33:24,050 S1: I'm logging in, maybe and doxing myself while I log in. 583 00:33:24,050 --> 00:33:27,600 S1: That's cool. Okay, look at these. You see these outlines? 584 00:33:27,600 --> 00:33:30,989 S1: It's got an outline for the type of coverage that 585 00:33:30,990 --> 00:33:36,450 S1: it's mostly getting. Look at this bar. Left 17% center 16%. 586 00:33:36,450 --> 00:33:40,650 S1: Right 67%. Okay. And you have these bars all over 587 00:33:40,650 --> 00:33:45,150 S1: the place. And what's really cool blind spot. Look at 588 00:33:45,150 --> 00:33:48,840 S1: this blind spot thing for the left. This is exposing 589 00:33:48,840 --> 00:33:51,490 S1: you to things that the right is talking about, that 590 00:33:51,490 --> 00:33:54,010 S1: the left is not. And same for over here, for 591 00:33:54,010 --> 00:33:56,800 S1: the right exposing you to things that the left is 592 00:33:56,800 --> 00:33:59,740 S1: talking about, that the right is not. So this this 593 00:33:59,740 --> 00:34:03,400 S1: helps you basically balance yourself, or at least have more 594 00:34:03,400 --> 00:34:07,510 S1: of an option of balancing yourself. Highly recommended. Especially if 595 00:34:07,510 --> 00:34:10,240 S1: you know somebody who claims to be wanting to be 596 00:34:10,239 --> 00:34:12,700 S1: less biased. And they're like, well, I'm just looking for 597 00:34:12,700 --> 00:34:16,270 S1: a good source or whatever. Send them this. Okay. And 598 00:34:16,270 --> 00:34:19,240 S1: the aphorism of the week, clear writing and clear speaking 599 00:34:19,239 --> 00:34:23,830 S1: are simply outputs of clear thinking. Naval clear writing and 600 00:34:23,830 --> 00:34:30,160 S1: clear speaking are simply outputs of clear thinking. Naval Unsupervised 601 00:34:30,160 --> 00:34:32,680 S1: Learning is produced and edited by Daniel Miller on a 602 00:34:32,680 --> 00:34:37,540 S1: Neumann U87 AI microphone using Hindenburg. Intro and outro music 603 00:34:37,540 --> 00:34:40,630 S1: is by Zomby with the Y, and to get the 604 00:34:40,630 --> 00:34:42,700 S1: text and links from this episode, sign up for the 605 00:34:42,700 --> 00:34:48,340 S1: newsletter version of the show at Daniel meisler.com/newsletter. We'll see 606 00:34:48,340 --> 00:34:49,060 S1: you next time.