WEBVTT - UL NO. 440: RAID (Real World AI Definitions) 0:00:00.050 --> 0:00:03.440 Whether you're starting or scaling your company's security program, demonstrating 0:00:03.470 --> 0:00:06.980 top notch security practices and establishing trust is more important 0:00:06.980 --> 0:00:12.110 than ever. Vanta automates compliance for Soc2, ISO 27,001 and more, 0:00:12.110 --> 0:00:16.070 saving you time and money while helping you build customer trust. Plus, 0:00:16.070 --> 0:00:20.329 you can streamline security reviews by automating questionnaires and demonstrating 0:00:20.329 --> 0:00:24.050 your security posture with a customer facing trust center, all 0:00:24.050 --> 0:00:28.610 powered by advanced AI. Over 7000 global companies like Atlassian, 0:00:28.610 --> 0:00:31.750 Flow Health and Quora use Vanta to manage risk and 0:00:31.750 --> 0:00:35.740 prove security in real time. Get $1,000 off Vanta when 0:00:35.740 --> 0:00:42.159 you go to Vanta comm slash unsupervised. That's vanta.com/supervised for 0:00:42.159 --> 0:00:47.379 $1,000 off. Welcome to Unsupervised Learning, a security, AI, and 0:00:47.380 --> 0:00:50.110 meaning focused podcast that looks at how best to thrive 0:00:50.110 --> 0:00:55.050 as humans in a post AI world. It combines original ideas, analysis, 0:00:55.050 --> 0:00:58.290 and mental models to bring not just the news, but 0:00:58.290 --> 0:01:05.880 why it matters and how to respond. All right, welcome 0:01:05.880 --> 0:01:08.370 to unsupervised learning. This is Daniel. All right. What do 0:01:08.370 --> 0:01:11.819 we have here? All right. So saw this funny thing. 0:01:11.819 --> 0:01:18.260 So basically, um, some terrorist funding news. Al Qaeda rebrands 0:01:18.260 --> 0:01:23.360 as I. Qaeda raises a $1 billion. Cool. I am 0:01:23.360 --> 0:01:28.910 rereading Alex Hormozi. Two business books. Uh, really? I think 0:01:28.910 --> 0:01:32.479 Alex Hormozi is, like the best business content generator right now, 0:01:32.480 --> 0:01:35.479 at least for my type of business, although he's like 0:01:35.480 --> 0:01:39.140 a gym person. Um, so it's not the same type 0:01:39.140 --> 0:01:44.470 of business, but he's just extraordinary at, like, user acquisition, 0:01:44.590 --> 0:01:47.650 like all sorts of stuff. And I love the format 0:01:47.650 --> 0:01:49.390 of his content. Like he's doing a whole bunch of 0:01:49.390 --> 0:01:53.440 shorts now. He's just making content probably, I don't know, 0:01:53.440 --> 0:01:56.050 it's got to be making 4 or 5 pieces a day. 0:01:56.200 --> 0:01:59.230 It's just so much and it's really good. My buddy 0:01:59.230 --> 0:02:03.730 Jacoby is an Army veteran, really cool security guy, and 0:02:03.730 --> 0:02:06.060 he's struggling really hard right now and he's got a 0:02:06.090 --> 0:02:09.540 gofund me. So if you could, uh, add to that, 0:02:09.540 --> 0:02:12.480 I would really appreciate it. Fabric made it to the 0:02:12.480 --> 0:02:16.500 front page of Hacker News. Uh, next run of augmented 0:02:16.500 --> 0:02:19.590 is on July 26th. You want to go sign up there? 0:02:19.620 --> 0:02:23.100 Got to spend some in-person time with a couple of 0:02:23.100 --> 0:02:27.690 fellow entrepreneurs creators over the weekend. Doing this stuff in 0:02:27.690 --> 0:02:31.970 person is invaluable. Like, we're talking about wanting to do 0:02:31.970 --> 0:02:35.270 a digital version of The thing that we built, and 0:02:35.270 --> 0:02:38.510 I'm reticent to do that because I. I want it 0:02:38.510 --> 0:02:42.440 to happen only in person, but we can make it 0:02:42.440 --> 0:02:45.200 digitally and then just interact with it in person. So 0:02:45.200 --> 0:02:47.570 that's what we're going to do, because it's too valuable 0:02:47.570 --> 0:02:51.350 not to be digital, but highly recommend setting up some 0:02:51.350 --> 0:02:55.630 similar sessions with your friends because it's really, really powerful. 0:02:55.780 --> 0:02:59.079 And my friend Monica Verma is running our CSO masterclass 0:02:59.080 --> 0:03:01.630 soon and she also has a newsletter, so go check 0:03:01.630 --> 0:03:04.450 those out. My work just finished a resource I've been 0:03:04.450 --> 0:03:07.750 working on for over a week called Raid. Real world 0:03:07.750 --> 0:03:13.450 AI definitions, not redundant array of inexpensive disks, which is 0:03:13.450 --> 0:03:15.549 the first thing I think of whenever I hear Raid. 0:03:15.550 --> 0:03:20.370 Either that or Bug Killer. But yeah, real world AI definitions. 0:03:20.490 --> 0:03:22.980 And I'm actually going to go into it and talk 0:03:22.980 --> 0:03:25.860 about it. So I'm going to go ahead and jump 0:03:25.860 --> 0:03:29.310 in here. Okay. So I see a lot of definitions 0:03:29.310 --> 0:03:31.380 for different AI terms out there. And I want to 0:03:31.380 --> 0:03:34.680 put my own thoughts into a long form version. And 0:03:34.680 --> 0:03:37.260 that's basically why I did this. And it's mostly for 0:03:37.260 --> 0:03:41.340 me because I love this quote. I forget the name 0:03:41.340 --> 0:03:44.570 of the woman. Uh, I need to memorize that. But 0:03:44.570 --> 0:03:48.680 she basically says, um, if you are thinking without writing, 0:03:48.680 --> 0:03:51.440 you only think that you're thinking. And I think that 0:03:51.440 --> 0:03:55.610 is such a powerful statement, because when you are writing 0:03:55.610 --> 0:03:57.680 and this is, um, this is why I call this 0:03:57.680 --> 0:04:03.080 a Feynman approach to learning, it's basically Feynman said, if 0:04:03.080 --> 0:04:05.630 you think you know something, try to teach it and 0:04:05.630 --> 0:04:08.000 you will figure out pretty quickly that you probably don't 0:04:08.000 --> 0:04:10.460 know it as well as you thought you did. And 0:04:10.460 --> 0:04:13.940 I've always known this for a very long time. So 0:04:13.940 --> 0:04:19.880 starting back in 1999, I started writing tutorials to myself 0:04:19.880 --> 0:04:23.210 to explain things. So when I was learning tech, I 0:04:23.210 --> 0:04:25.729 literally would try to learn the tech from textbooks and 0:04:25.730 --> 0:04:27.620 then it wouldn't make sense to me. So I would 0:04:27.620 --> 0:04:30.330 go write my own explanation and then I would just 0:04:30.330 --> 0:04:33.810 use that whenever I wanted to remember something. And that 0:04:33.810 --> 0:04:36.750 turned into my tutorials page, which is how I got 0:04:36.750 --> 0:04:40.260 started on the internet. So table of contents here, the 0:04:40.260 --> 0:04:44.610 expanded definitions table, the one liner definitions table, and then 0:04:44.610 --> 0:04:48.900 I break out AGI and ASI into two different sections. 0:04:48.900 --> 0:04:51.630 So we're going to start with the one liner AI 0:04:51.630 --> 0:04:55.640 definitions table. And this is coming off of Hamel Hussein's 0:04:55.670 --> 0:05:00.289 I bullshit knife which gave violently short definitions for a 0:05:00.290 --> 0:05:03.440 bunch of I terms. And this is my expanded version 0:05:03.440 --> 0:05:07.609 of that. Okay. So this is made to like crystallize 0:05:07.610 --> 0:05:10.760 in your brain what these things are in very, very 0:05:10.760 --> 0:05:14.270 crisp conversational form. And we're going to go into more 0:05:14.270 --> 0:05:17.740 detail in the further ones. But I tech that does 0:05:17.740 --> 0:05:22.270 cognitive tasks that only humans could do before. And the 0:05:22.270 --> 0:05:25.900 reason I have that only humans could do before is 0:05:25.900 --> 0:05:29.260 because that's a moving bar. So before there was a 0:05:29.260 --> 0:05:35.470 chess playing robot or a chess playing AI, everyone said, well, yeah, 0:05:35.470 --> 0:05:37.870 but it can't play chess. It never will be able 0:05:37.870 --> 0:05:40.299 to play chess because that's a human thing. And of 0:05:40.300 --> 0:05:43.920 course you could do that now. Image identification. Being able 0:05:43.920 --> 0:05:46.740 to tell the difference between a dog and a cat, 0:05:46.740 --> 0:05:49.290 that was a very, very hard problem for a very 0:05:49.290 --> 0:05:51.660 long time. You have to think in terms of like 0:05:51.660 --> 0:05:55.080 what currently is a thing that only humans can do 0:05:55.080 --> 0:05:58.800 that I cannot do. Yet that is the barrier which 0:05:58.800 --> 0:06:02.370 determines AI or not, because the moment you cross over 0:06:02.370 --> 0:06:06.000 into that and I suddenly becomes available to do that, 0:06:06.000 --> 0:06:09.479 that's what's currently called I. Now, what's interesting is as 0:06:09.480 --> 0:06:12.779 it moves further and further out, people forget that that's 0:06:12.779 --> 0:06:16.049 I like image recognition. A lot of people are like, 0:06:16.050 --> 0:06:19.110 that's not AI, that's ML. Well, it's still AI, first 0:06:19.110 --> 0:06:22.140 of all, because machine learning is a subset of AI, 0:06:22.140 --> 0:06:25.890 but also because it's doing something that only humans could 0:06:25.890 --> 0:06:30.140 do before. So I, I love that definition. Very flexible 0:06:30.290 --> 0:06:34.370 machine learning AI that can improve just by seeing more data. 0:06:34.370 --> 0:06:36.770 And we're going to talk about this more, uh, later 0:06:36.770 --> 0:06:41.630 on prompting clearly articulate what you want from an AI rag. 0:06:41.630 --> 0:06:45.080 Provide context to an AI that's too big or expensive 0:06:45.080 --> 0:06:48.140 to fit in a prompt, an agent, an AI component 0:06:48.140 --> 0:06:51.650 that does more than just LM call to respond. So 0:06:51.650 --> 0:06:55.260 it's not just call and response, that's just regular. I, 0:06:55.260 --> 0:06:58.110 an agent, does more than that, taking on more of 0:06:58.110 --> 0:07:01.049 the work chain of thought. Tell the I to walk 0:07:01.050 --> 0:07:04.440 through its thinking and steps. Zero shot ask an AI 0:07:04.470 --> 0:07:08.010 to do something without any examples. Multi-shot. Ask an AI 0:07:08.040 --> 0:07:12.180 to do something and provide multiple examples. Prompt injection tricking 0:07:12.180 --> 0:07:16.410 an AI into doing something bad. Jailbreaking, bypassing security controls 0:07:16.410 --> 0:07:19.100 to get full execution ability, or at least as full 0:07:19.100 --> 0:07:23.840 as possible, doesn't necessarily need to be full AGI general 0:07:23.840 --> 0:07:27.950 AI smart enough to replace an $80,000 white collar worker, 0:07:27.950 --> 0:07:33.170 and ASI, which is superintelligence. General AI that's smarter and 0:07:33.170 --> 0:07:36.050 or more capable than any human. So I think that's 0:07:36.050 --> 0:07:39.020 a pretty clean list for thinking about these concepts. And 0:07:39.020 --> 0:07:41.510 now I want to go into more detail for each. 0:07:41.530 --> 0:07:44.950 So AI. AI is technology that does cognitive tasks or 0:07:44.950 --> 0:07:48.010 work that could previously only be done by humans. So 0:07:48.010 --> 0:07:51.130 cognitive tasks or work, and I might clean that up 0:07:51.130 --> 0:07:55.630 at some point and just make it work or cognitive tasks. Anyway, 0:07:55.630 --> 0:07:58.000 lots of different ways to define AI. So this is 0:07:58.000 --> 0:08:00.610 going to be controversial. And this is what I talked about. 0:08:00.610 --> 0:08:03.040 Well yeah of course I can do this. But it 0:08:03.040 --> 0:08:05.170 still can't do this and it probably never will be 0:08:05.170 --> 0:08:07.650 able to. And then the narrator voice says, yeah, that 0:08:07.650 --> 0:08:10.230 happened seven months later. And this is why I like 0:08:10.230 --> 0:08:13.770 the definition is because it moves over time. Machine learning 0:08:13.800 --> 0:08:16.920 a subset of AI that enables a system to learn 0:08:16.920 --> 0:08:20.400 from data alone, rather than needing to be explicitly programmed. 0:08:20.400 --> 0:08:24.990 This is a super exciting to me as a definition. Actually, 0:08:24.990 --> 0:08:28.679 as a concept. I can't believe the program is the same, 0:08:28.680 --> 0:08:32.100 but you change the data and the program gets smarter. 0:08:32.130 --> 0:08:35.160 That blows my mind. And it's like the most important. 0:08:35.160 --> 0:08:37.950 In fact, if you want to go tighter learning from 0:08:37.950 --> 0:08:41.340 data alone or a technology system that's able to learn 0:08:41.340 --> 0:08:44.069 from data alone, like lots of different ways you can 0:08:44.070 --> 0:08:47.580 skin this thing. Prompt engineering. Prompt engineering is the art 0:08:47.580 --> 0:08:51.180 and science of using language, usually text, to get AI 0:08:51.210 --> 0:08:53.730 to do precisely what you want it to do. Yeah, 0:08:53.760 --> 0:08:56.410 a lot of people think prompt engineering is like this 0:08:56.410 --> 0:08:59.590 really special thing. Others think it's not important at all. 0:08:59.590 --> 0:09:03.490 I think it's absolutely an art and a science because 0:09:03.490 --> 0:09:07.150 it's more about clear thinking than about text itself. Just 0:09:07.150 --> 0:09:10.330 like writing, the best prompt engineer is the same. It 0:09:10.330 --> 0:09:13.330 comes from deeply understanding the problem and be able to 0:09:13.330 --> 0:09:16.150 break out your instructions to the AI in a very 0:09:16.150 --> 0:09:19.550 methodical and clear way, So the clearer you can think 0:09:19.550 --> 0:09:23.599 and the clearer you can think about problems and describe methodologies. 0:09:23.600 --> 0:09:25.640 That's that's what's going to make you really good at 0:09:25.640 --> 0:09:28.760 prompt engineering. Don't think of it as an AI thing. 0:09:28.760 --> 0:09:32.960 Think of it as a thinking thing and a writing thing. Retrieval. 0:09:32.960 --> 0:09:36.920 Augmented generation Rag is the process of taking large quantities 0:09:36.920 --> 0:09:39.470 of data, which are either too large or too expensive 0:09:39.470 --> 0:09:41.990 to put in a prompt directly, and making that data 0:09:42.010 --> 0:09:46.510 usable as vectorized embeddings to AI at runtime. And I 0:09:46.510 --> 0:09:49.720 say it's important to understand that Rag is a hack 0:09:49.720 --> 0:09:52.750 that solves a specific problem, which is that people and 0:09:52.750 --> 0:09:56.650 companies have way too much data gigabytes, terabytes, petabytes of 0:09:56.650 --> 0:09:59.140 data that they want their AI to be aware of 0:09:59.140 --> 0:10:02.650 when performing tasks. But I can only handle small amounts 0:10:02.650 --> 0:10:05.740 of that data per interaction. And that's what Rag is for. 0:10:05.740 --> 0:10:08.189 So the solution we come up with is to use 0:10:08.190 --> 0:10:12.090 embeddings and vector databases to encode relevant information and send 0:10:12.090 --> 0:10:14.550 that along with the prompts. And it's not really clear 0:10:14.550 --> 0:10:17.160 what the successor is going to be. But one option 0:10:17.160 --> 0:10:20.520 is to just have massive context windows and put the 0:10:20.520 --> 0:10:23.189 raw content in there. But then you have cost problems, 0:10:23.190 --> 0:10:24.990 so you would need that to be very cheap for 0:10:25.020 --> 0:10:28.560 that to work. Agents. An AI agent is an AI 0:10:28.590 --> 0:10:32.540 component that interprets instructions and takes on more of the 0:10:32.540 --> 0:10:37.430 work in a total AI workflow than just LLM response, 0:10:37.429 --> 0:10:43.729 for example, executing functions, performing data lookups, etc. before passing 0:10:43.730 --> 0:10:47.960 on results. Yeah, so there are lots of different definitions 0:10:47.960 --> 0:10:51.079 for agent. I think the trick is to go back 0:10:51.080 --> 0:10:55.510 to the Latin, the agents or Arjun's, which is to 0:10:55.510 --> 0:10:59.020 do or to act. So I think eventually this is 0:10:59.020 --> 0:11:01.210 going to be like an AI component that has its 0:11:01.210 --> 0:11:04.210 own mission and goals and can use its resources and 0:11:04.210 --> 0:11:07.630 capabilities to accomplish them in a self-directed way. That is 0:11:07.630 --> 0:11:11.740 the true nature of agent or agents from the Latin. 0:11:11.740 --> 0:11:14.650 But I think in this current version, I think the 0:11:14.650 --> 0:11:17.230 best way to say it is that anything that acts 0:11:17.230 --> 0:11:20.250 on behalf of the mission and something that performs multiple 0:11:20.250 --> 0:11:23.910 steps towards the final goal, and that's why I'm using 0:11:23.910 --> 0:11:26.730 this definition for agent chain of thought, a way of 0:11:26.730 --> 0:11:29.040 interacting with AI in which you don't just say what 0:11:29.040 --> 0:11:30.990 you want, but you give the steps that you would 0:11:30.990 --> 0:11:33.660 take to accomplish the task. So you're basically teaching it 0:11:33.660 --> 0:11:36.449 how you think, which which is essentially teaching it how 0:11:36.450 --> 0:11:39.059 to think like a human. And then it uses that 0:11:39.059 --> 0:11:41.699 template to sort of do that every time it's going 0:11:41.700 --> 0:11:44.000 to solve the same type of problem. And to me, 0:11:44.000 --> 0:11:46.100 it's kind of an example of what we talked about 0:11:46.100 --> 0:11:50.059 in prompt engineering, clear thinking. It's walking the AI through 0:11:50.059 --> 0:11:52.910 how you think when you're solving a problem yourself. Prompt 0:11:52.910 --> 0:11:56.300 injection a method of using language, usually text, to get 0:11:56.300 --> 0:11:58.460 AI to do something it's not supposed to do. And 0:11:58.460 --> 0:12:01.310 a lot of people confuse prompt injection with jailbreaking. And 0:12:01.309 --> 0:12:03.290 I think the best way to think about this. And 0:12:03.290 --> 0:12:06.530 thanks to Jason Haddox for talking through this with me, 0:12:06.530 --> 0:12:09.800 because he had a very similar definition, uh, is to 0:12:09.800 --> 0:12:13.610 say prompt injection is a method. It's an attack avenue. 0:12:13.610 --> 0:12:17.300 It's not a payload, whereas jailbreaking is a goal. We're 0:12:17.300 --> 0:12:20.900 trying to bypass the security on a system. Prompt injection 0:12:20.900 --> 0:12:23.840 is not bypassing the security on a system. Well, it's 0:12:23.840 --> 0:12:26.780 a method for doing that. It's a way of tricking 0:12:26.780 --> 0:12:29.359 the eye into doing something it's not supposed to do. 0:12:29.360 --> 0:12:32.790 But that could be telling a joke that it's not 0:12:32.790 --> 0:12:35.340 supposed to tell. That could be showing a picture it's 0:12:35.340 --> 0:12:37.710 not supposed to show. It doesn't mean it's bypassing the 0:12:37.710 --> 0:12:40.590 security on a system so you can execute commands. So 0:12:40.590 --> 0:12:42.990 it could be data exfil, it could be code execution, 0:12:42.990 --> 0:12:46.260 it could be image generation. It could be anything. So easiest, 0:12:46.260 --> 0:12:48.390 cleanest way, at least in my mind, to think about this. 0:12:48.390 --> 0:12:51.630 Prompt injection is a method. Jailbreaking is a goal. All right. 0:12:51.630 --> 0:12:54.920 We talked about jailbreaking. Going to pass that one. Artificial 0:12:54.920 --> 0:12:58.939 general intelligence. The ability for an AI system, whether a model, 0:12:58.940 --> 0:13:01.700 a product or a system to fully do the job 0:13:01.700 --> 0:13:04.849 of an average US based knowledge worker. In other words, 0:13:04.850 --> 0:13:07.580 it's not just competent at doing a specific thing, which 0:13:07.580 --> 0:13:12.170 is called narrow AI oftentimes, but many different things, which 0:13:12.170 --> 0:13:15.260 is why it's called general. So basically it's some combination 0:13:15.260 --> 0:13:19.000 of sufficiently general and sufficiently competent, and the amounts of 0:13:19.000 --> 0:13:21.760 which are, you know, going to be debated so generally 0:13:21.760 --> 0:13:25.870 competent at many other cognitive tasks, sample efficient so we 0:13:25.870 --> 0:13:29.020 can learn quickly from like one example or a low 0:13:29.020 --> 0:13:31.449 number of examples. And the system should be able to 0:13:31.450 --> 0:13:35.380 apply new information, concepts and skills that learns to additional 0:13:35.380 --> 0:13:39.430 new problems. So it's not just generally intelligent from its training, 0:13:39.429 --> 0:13:42.340 but when it learns new things, it's generally intelligent at 0:13:42.360 --> 0:13:44.970 applying those new things that learn to everything else in 0:13:44.970 --> 0:13:48.870 the future. And I like this definition focusing on the 0:13:48.870 --> 0:13:54.390 worker replacing a worker, because I think this is arguably 0:13:54.390 --> 0:13:58.020 what matters most to humans when we talk about AI discussion, 0:13:58.020 --> 0:14:01.230 which is the future of humans in a world of AI, 0:14:01.260 --> 0:14:05.520 regular people don't care about processing speed or model size 0:14:05.520 --> 0:14:07.949 or model weights or any of that crap. What they 0:14:07.950 --> 0:14:10.080 care about is if and when any of this is 0:14:10.080 --> 0:14:13.380 going to actually affect them. And that means job replacement mostly, 0:14:13.380 --> 0:14:16.320 or at least that's the most important thing. So these 0:14:16.320 --> 0:14:20.280 are the levels that I see within that job replacement 0:14:20.280 --> 0:14:23.790 of this average white collar worker. So first level is 0:14:23.790 --> 0:14:27.239 better but with significant drawbacks. And I'm not going to 0:14:27.240 --> 0:14:29.850 go through all of these. But it's like the interface 0:14:29.850 --> 0:14:33.910 the language, the confusion, the errors, flexibility, basically. Basically, real 0:14:33.910 --> 0:14:38.020 world work environments are kludgy and lame and they change 0:14:38.020 --> 0:14:41.350 all the time. And it's basically a giant mess. This 0:14:41.350 --> 0:14:44.650 level one is basically saying, yes, we brought in an 0:14:44.650 --> 0:14:47.710 AI worker to replace that person, or we didn't hire 0:14:47.710 --> 0:14:50.770 a person. We used an AI worker instead for for 0:14:50.770 --> 0:14:53.530 growth if you want to be nice. So that's what happened. 0:14:53.530 --> 0:14:57.720 We brought in this level one AGI one. And here's 0:14:57.720 --> 0:15:01.470 the problem. It's kludgy. So so the problem is that 0:15:01.470 --> 0:15:03.900 they're doing the work. But you have to clean up 0:15:03.900 --> 0:15:07.950 after them. So the interface proprietary cumbersome the language that 0:15:07.950 --> 0:15:10.740 you have to use. It's still kind of like AI language. 0:15:10.740 --> 0:15:14.010 It's almost like you're prompting. You're not just normally talking 0:15:14.010 --> 0:15:16.890 to them. Um, it gets confused a decent amount of 0:15:16.890 --> 0:15:19.820 the time. There are frequent mistakes which need to be 0:15:19.820 --> 0:15:23.120 fixed by humans, and if you want to change what 0:15:23.120 --> 0:15:26.720 this thing is doing, it needs to be largely retooled. 0:15:27.260 --> 0:15:30.020 Not not like in a major way, but you've got 0:15:30.020 --> 0:15:31.790 to have a conversation with it. You've got to change 0:15:31.790 --> 0:15:34.520 some documentation, you've got to do some stuff. So going 0:15:34.520 --> 0:15:38.000 back to the top level of this one better but 0:15:38.000 --> 0:15:41.150 with significant drawbacks. So it is right at the line 0:15:41.150 --> 0:15:43.330 of being too much work and not good enough to 0:15:43.330 --> 0:15:49.150 replace a human right. That's the trick. It's just barely better. Barely. 0:15:49.150 --> 0:15:51.520 And sometimes you can't tell if it's better or not. 0:15:51.520 --> 0:15:54.700 It might be too much work. Next one. It removes 0:15:54.700 --> 0:15:57.400 a lot of those problems. So it's like in the middle. 0:15:57.400 --> 0:16:03.940 It's like decent, competent but imperfect. So mostly normal workflows, 0:16:03.940 --> 0:16:08.160 mostly human with exceptions. Doesn't get confused very often. Fewer 0:16:08.160 --> 0:16:10.920 mistakes and the mistakes aren't as bad, and it just 0:16:10.920 --> 0:16:14.250 decently well to new direction from leaders. So that's like 0:16:14.250 --> 0:16:19.770 the middle level. And then AGI three is full worker replacement. 0:16:19.770 --> 0:16:26.070 So this can replace this average 80 K worker in 2022. 0:16:26.100 --> 0:16:29.610 White collar worker in the US. It can fully replace them. 0:16:29.610 --> 0:16:32.630 So when you talk to them, it's you talk to 0:16:32.630 --> 0:16:34.940 them just like any other employee. You could text them, 0:16:34.940 --> 0:16:37.130 you could voice them, you could do video with them. 0:16:37.130 --> 0:16:39.410 It's just like an employee. You talk to them exactly 0:16:39.410 --> 0:16:42.950 like an employee. They're confused the same amount or less 0:16:42.950 --> 0:16:46.490 than the $80,000 employee. Keep in mind, humans have all 0:16:46.490 --> 0:16:49.250 these problems too, right? Humans also get confused and need 0:16:49.250 --> 0:16:51.230 to be retasked or whatever. So it's not like we're 0:16:51.230 --> 0:16:54.200 competing with perfection here. Same or fewer mistakes than an 0:16:54.200 --> 0:16:58.210 $80,000 employee. Employees also make mistakes and just as flexible 0:16:58.210 --> 0:17:02.050 or more than $80,000 employee employees also have trouble like 0:17:02.050 --> 0:17:04.510 completely being retests. It's like I thought it was working 0:17:04.510 --> 0:17:06.520 on that. What are you doing? Like, so this is 0:17:06.520 --> 0:17:09.790 basically saying it's just as good as them or or better. 0:17:09.790 --> 0:17:14.649 So this is the top level of complete worker replacement. 0:17:14.650 --> 0:17:16.420 I think this is going to take a decent amount 0:17:16.420 --> 0:17:18.550 of time. And then the next question is like okay, 0:17:18.550 --> 0:17:21.940 but what if it's like that? It's a full replacement 0:17:21.940 --> 0:17:24.730 for that worker. So it's AGI three, but with an 0:17:24.730 --> 0:17:28.930 IQ of 150. Well that's that's really powerful. And I'm 0:17:28.930 --> 0:17:32.440 not working on that axis. I'm only working on how 0:17:32.440 --> 0:17:35.710 good of a worker are they. And that's the three 0:17:35.710 --> 0:17:39.010 levels for AGI okay. Let's do AC a level of 0:17:39.010 --> 0:17:43.180 general AI that's smarter and more capable than any human 0:17:43.180 --> 0:17:45.830 that's ever lived. So this is interesting for a number 0:17:45.830 --> 0:17:49.160 of reasons. One, it's a threshold that sits above AGI 0:17:49.160 --> 0:17:52.100 and people don't really agree on that definition. And the 0:17:52.100 --> 0:17:54.709 second thing is, at least as I'm defining it, it 0:17:54.710 --> 0:17:58.490 has a massive range. And third, it blends with AGI 0:17:58.490 --> 0:18:02.960 because AGI just really means general incompetent. And ASI is 0:18:02.960 --> 0:18:08.139 also general incompetent. So really ASI is also AGI. It's 0:18:08.140 --> 0:18:12.040 just uh, at a level beyond any human. So AGI 0:18:12.040 --> 0:18:15.489 is replacement for a human worker making SDK in the 0:18:15.490 --> 0:18:20.260 US in 2022, which is pretty AI, prision AI and 0:18:20.260 --> 0:18:25.449 ASI is more smart and more competent than any human. 0:18:25.450 --> 0:18:28.659 And the model I'm using there is like John von Neumann, 0:18:28.660 --> 0:18:33.900 because unlike Einstein and Newton, he was, uh, very broad. 0:18:33.930 --> 0:18:37.980 He was a generalist, right. So he did game theory, physics, computing, 0:18:37.980 --> 0:18:41.580 lots of other stuff. But I don't think being smarter 0:18:41.580 --> 0:18:44.070 is the only thing that matters, because I think it 0:18:44.070 --> 0:18:46.260 comes down to multiple things. So if I look at 0:18:46.260 --> 0:18:48.720 all the different things that I'm putting into ASI, it's 0:18:48.720 --> 0:18:52.080 like your ability to model abstractions of reality at these 0:18:52.080 --> 0:18:57.590 different levels. It's the ability to take action, to perceive, understand, improve, solve, create, destroy. 0:18:57.590 --> 0:19:00.620 And then it's like, okay, what fields are we innovating in? 0:19:00.619 --> 0:19:05.000 What problems are we able to solve and what scale 0:19:05.030 --> 0:19:08.630 are we actually working? You know, thinking about, I guess 0:19:08.630 --> 0:19:11.540 I should add universe or whatever, but all the way 0:19:11.540 --> 0:19:15.109 from Quark to Galaxy is like the scale. So what's 0:19:15.109 --> 0:19:17.330 really cool about this is you can now turn these 0:19:17.330 --> 0:19:20.859 into functional phrases. So you could say, like an AI 0:19:20.890 --> 0:19:26.710 capable of curing aging by creating new chemicals that affect DNA. Okay. 0:19:26.710 --> 0:19:30.160 So we have a problem here creating new chemicals. That's 0:19:30.160 --> 0:19:34.330 like a net new thing, right? So that you're seeing this, 0:19:34.330 --> 0:19:37.810 you know, verbs and nouns here, maintaining a full city 0:19:37.810 --> 0:19:42.129 by monitoring and adjusting all public resources in real time. 0:19:42.130 --> 0:19:44.740 And the AI capable of taking over a country by 0:19:44.740 --> 0:19:49.119 manufacturing a drone army and a new energy based weapon. 0:19:49.119 --> 0:19:51.520 So this is like new science here, and an AI 0:19:51.550 --> 0:19:55.960 capable of faster than light travel by discovering completely new physics. 0:19:55.960 --> 0:19:59.320 So that's like a level above, right? So let's look 0:19:59.320 --> 0:20:03.430 at the different levels here. AC one superior, an AI 0:20:03.460 --> 0:20:07.790 capable of making incremental improvements in multiple fields and managing 0:20:07.790 --> 0:20:11.750 up to city size entities on its own. So here 0:20:11.750 --> 0:20:13.730 are the things that it has smarter and more capable 0:20:13.730 --> 0:20:17.300 than any human on many topics. Able to move progress 0:20:17.300 --> 0:20:21.439 forward in multiple scientific fields. Able to recommend novel solutions 0:20:21.440 --> 0:20:25.070 to many of our main problems. Able to copy and 0:20:25.070 --> 0:20:28.970 surpass the creativity of many top artists. Able to fully 0:20:28.970 --> 0:20:32.889 manage a large company or city itself and keep, keep 0:20:32.890 --> 0:20:36.970 in mind, copy and surpass some of the best artists. Now, 0:20:36.970 --> 0:20:40.510 what this really means is like in a superhuman way, right? 0:20:40.510 --> 0:20:44.020 So it's not just like better in one area. It's like, okay, 0:20:44.020 --> 0:20:48.310 it's Eminem, but it's way better than Eminem. It's Kendrick Lamar, 0:20:48.310 --> 0:20:52.360 but five times better, right? Same with these types of things. 0:20:52.359 --> 0:20:56.910 Novel solutions, moving progress forward, which obviously no human has done. 0:20:56.910 --> 0:20:59.790 So they're doing it better, but it doesn't have these. 0:20:59.790 --> 0:21:04.200 It can't create. Net new physics or net new material science. 0:21:04.200 --> 0:21:07.500 It's unable to fundamentally improve itself by orders of magnitude. 0:21:07.500 --> 0:21:12.150 So next phase dominant okay. First one was superior. Next 0:21:12.150 --> 0:21:14.879 one is dominance. This one has all of AC one. 0:21:14.880 --> 0:21:17.550 But it's also able to completely change how we see 0:21:17.550 --> 0:21:21.440 multiple fields able to completely solve most of our current problems. 0:21:21.470 --> 0:21:25.370 Able to fully manage a country by itself. So this 0:21:25.369 --> 0:21:27.770 is a huge jump over the first one. Able to 0:21:27.770 --> 0:21:31.760 fundamentally improve itself by orders of magnitude, but still can't create. 0:21:31.760 --> 0:21:34.100 Net new physics or run an entire planet. And then 0:21:34.100 --> 0:21:38.030 you have AC three, another massive jump AI capable of 0:21:38.030 --> 0:21:43.429 creating net new physics, completely new materials, manipulation of near 0:21:43.450 --> 0:21:47.830 fundamental reality or fundamental reality. If it's possible to do 0:21:47.830 --> 0:21:52.270 that and can run an entire planet. And keep in mind, 0:21:52.270 --> 0:21:55.780 I might add AC for um, if I have a 0:21:55.780 --> 0:22:00.850 like a clear picture of what that looks like. Um, Yoshi. No. Rishi, 0:22:00.880 --> 0:22:05.379 my buddy Rishi mentioned, um, what about someone? An AI 0:22:05.380 --> 0:22:08.409 that's smarter than all of humanity put together? And I 0:22:08.410 --> 0:22:12.850 was like, well, isn't that confusing knowledge with intelligence and reason? 0:22:12.850 --> 0:22:15.280 But I haven't heard back from him yet. Other things 0:22:15.280 --> 0:22:18.460 to consider here. So this one adds the ability to 0:22:18.460 --> 0:22:22.360 modify reality at a fundamental or near fundamental level, able 0:22:22.359 --> 0:22:26.350 to manage an entire planet simultaneously. And maybe its primary 0:22:26.350 --> 0:22:30.310 concerns become like sun expansion, which is going to like 0:22:30.310 --> 0:22:33.530 eat up the earth or populating the galaxy and beyond, 0:22:33.530 --> 0:22:35.990 or the heat death of the universe, which is like 0:22:35.990 --> 0:22:39.170 escaping this reality. Yeah, or breaking out of the simulation. 0:22:39.170 --> 0:22:41.959 I should add that. So as a summary, I terms 0:22:41.960 --> 0:22:45.320 are confusing. It's nice to have simple, practical versions. It's 0:22:45.320 --> 0:22:48.620 useful to crystallize your own definitions on paper, both for 0:22:48.619 --> 0:22:50.900 your own reference and to see if your definitions are 0:22:50.900 --> 0:22:55.130 consistent with each other. And I think these eye definitions 0:22:55.130 --> 0:22:59.350 work best when they're human focused and practically worded, because 0:22:59.350 --> 0:23:01.629 it is very easy to try to be technical with 0:23:01.630 --> 0:23:05.410 these definitions, and you instantly open up like this giant 0:23:05.410 --> 0:23:09.580 mess the moment you go a level down in technicality. 0:23:09.580 --> 0:23:13.449 And what you'll find is if you get 20 AI 0:23:13.450 --> 0:23:17.230 experts in in a room and a bunch of ML experts, 0:23:17.230 --> 0:23:20.160 they've been doing this 30 years and you ask them 0:23:20.160 --> 0:23:23.100 their opinions. They don't agree. The textbooks don't agree, the 0:23:23.100 --> 0:23:26.190 experts don't agree. The moment you go a layer down, 0:23:26.190 --> 0:23:30.540 it's a giant mess. And that giant mess stops conversations 0:23:30.540 --> 0:23:33.360 from happening because they end up, they'll get two hours 0:23:33.359 --> 0:23:35.340 into a podcast and be like, oh, that's what you 0:23:35.340 --> 0:23:37.560 meant by I. That's not what I mean by I. 0:23:37.560 --> 0:23:41.280 And I'm like, okay, let's stop messing with let's stop 0:23:41.280 --> 0:23:44.300 wasting two hours in a conversation because we're not talking 0:23:44.300 --> 0:23:47.990 about the same thing. Let's level up to an abstraction 0:23:47.990 --> 0:23:51.230 level where it's useful in a conversation. That's why it's 0:23:51.230 --> 0:23:54.290 called real world AI definition. All right. So that's that one. 0:23:54.290 --> 0:23:57.439 All right. Stories. Cloudflare has a new free tool to 0:23:57.440 --> 0:24:00.680 stop bots from scraping websites. Remember I told you about 0:24:00.680 --> 0:24:03.620 Cloudflare how they just do like the coolest stuff and 0:24:03.619 --> 0:24:06.950 they just Cloudflare is a beast. I am telling you. 0:24:06.950 --> 0:24:10.930 They just they see a problem that is annoying everybody 0:24:10.930 --> 0:24:13.270 and they go and just pour a little bit of 0:24:13.270 --> 0:24:16.270 like cement in that seam or that crack or that 0:24:16.270 --> 0:24:19.149 hole or whatever, and they're just like, we're going to 0:24:19.150 --> 0:24:21.669 solve this. And if no one else does it, they're 0:24:21.670 --> 0:24:23.710 just going to be the solution for this. Like they're 0:24:23.710 --> 0:24:26.170 doing this with Captcha. They're doing this now with AI 0:24:26.200 --> 0:24:30.040 scraping like somebody over there is going, this is going 0:24:30.040 --> 0:24:31.810 to be a problem in the future or this is 0:24:31.830 --> 0:24:34.560 already becoming a problem. Let's get there first. And I 0:24:34.560 --> 0:24:37.530 massively respect it. And before too long, the whole whole 0:24:37.530 --> 0:24:40.530 internet's going to be Cloudflare, because over the course of time, 0:24:40.530 --> 0:24:44.070 they've made like 500 of these tools at some point. Right. 0:24:44.070 --> 0:24:47.430 And it's like all those random little tools that they 0:24:47.430 --> 0:24:51.060 made become the structure of the internet. Twilio says, uh, 0:24:51.060 --> 0:24:55.500 somebody's got phone numbers because of an unsecured auth API endpoint, 0:24:55.500 --> 0:24:59.869 33 million phone numbers. Russian experts say they fully analyzed 0:24:59.869 --> 0:25:05.210 the structure of the American attack Miss Atacms missile, and 0:25:05.210 --> 0:25:09.140 they believe that's going to help them fighting Ukrainian missiles, 0:25:09.140 --> 0:25:13.100 which is not good. Thanks to tines for sponsoring. North 0:25:13.100 --> 0:25:15.649 Korea has switched its state TV broadcast from a Chinese 0:25:15.650 --> 0:25:18.320 satellite to a Russian one. And now fewer people can 0:25:18.320 --> 0:25:22.369 watch Hezbollah launch over 200 rockets and drones at Israel 0:25:22.369 --> 0:25:26.930 after Israel killed a senior Hezbollah official. Thanks to nudge 0:25:26.930 --> 0:25:31.280 security for sponsoring. US intelligence community is diving into generative 0:25:31.280 --> 0:25:36.050 AI to enhance intelligence operations, using it for search, discovery, 0:25:36.050 --> 0:25:41.149 counterargument generation. Uh, I'm going to make so much Intel stuff, 0:25:41.150 --> 0:25:44.700 I cannot wait to build that product in substrate. I 0:25:44.700 --> 0:25:47.970 cannot wait. There's some analysis saying I cannot be funny 0:25:47.970 --> 0:25:52.980 from Anya, Jeremy Greenwald, and I think I disagree. And 0:25:52.980 --> 0:25:56.459 here's an example. My buddy Joseph posted this one. It's 0:25:56.460 --> 0:26:00.330 a glyph app and it makes memes. Look at this 0:26:00.330 --> 0:26:05.160 bug bounty, bro. Like I legitimately laughed at some of these. 0:26:05.160 --> 0:26:10.190 Show me show you my leet skills. Use the inspect 0:26:10.190 --> 0:26:14.270 element I only hack ethically dos school website in eighth grade. 0:26:14.390 --> 0:26:18.230 I mean, first of all, these are real. They're good 0:26:18.230 --> 0:26:20.629 in terms of like they actually apply to the field 0:26:20.630 --> 0:26:24.619 that they're making fun of. And, uh, my POC is flawless. 0:26:24.650 --> 0:26:28.970 It's a Rick roll link. I generated this. Okay. I've 0:26:28.970 --> 0:26:31.490 been saying for a while that being funny is a 0:26:31.490 --> 0:26:37.209 major milestone in human AI, in AI becoming real. And 0:26:37.210 --> 0:26:39.580 just like a lot of other things in AI, there 0:26:39.580 --> 0:26:41.440 are so many people who are like, oh yeah, it can. 0:26:41.440 --> 0:26:43.270 It could tell a cat from a dog, but it's 0:26:43.270 --> 0:26:46.120 never going to make me laugh. Whatever. The stuff's been 0:26:46.119 --> 0:26:49.990 out for like 18 minutes, this just happened at the 0:26:49.990 --> 0:26:53.500 end of 2022. Like this is still day one. I'm 0:26:53.500 --> 0:26:56.359 telling you right now, there will be funny I very 0:26:56.359 --> 0:26:58.880 soon and this is the first version of it. I mean, 0:26:58.880 --> 0:27:02.870 this is already funny. I technically the real standard is 0:27:02.869 --> 0:27:06.050 going to be a stand up routine, a deepfake of 0:27:06.050 --> 0:27:10.190 a face, or actually a deepfake. Okay, I am watching 0:27:10.190 --> 0:27:14.090 a stage. There is a deepfake, uh, person standing there. 0:27:14.450 --> 0:27:16.430 It doesn't even have to convince me. It could be 0:27:16.430 --> 0:27:20.679 like uncanny valley. But if there's a deepfake person standing 0:27:20.680 --> 0:27:24.490 there talking, doing a stand up routine and it makes 0:27:24.490 --> 0:27:26.649 me laugh the way it makes me laugh when I 0:27:26.650 --> 0:27:30.399 watch other stand ups, that is a major milestone. That 0:27:30.400 --> 0:27:35.290 is extraordinary. I don't know how long that's going to. Oh, 0:27:35.320 --> 0:27:37.840 I should do that on a prediction market. Yeah, I'm 0:27:37.840 --> 0:27:39.820 curious about that. I'm guessing like a year and a 0:27:39.820 --> 0:27:41.800 half to two years. I think it's going to be 0:27:41.800 --> 0:27:44.970 a little while because that one's really, really difficult. But 0:27:44.970 --> 0:27:48.000 it could be, honestly, it could be opus, it could 0:27:48.000 --> 0:27:50.970 be opus 35 or GPT five. It could happen in 0:27:50.970 --> 0:27:53.820 this next iteration, which is going to be around the 0:27:53.820 --> 0:27:56.520 end of this year or beginning of next year. Okay. 0:27:56.520 --> 0:27:59.580 Nvidia is set to make 12 billion by selling over 0:27:59.580 --> 0:28:04.590 a million HG, H20 GPUs. That sounded impressive. I thought 0:28:04.590 --> 0:28:06.570 it was like, how are you going to send those? 0:28:06.570 --> 0:28:09.119 I thought we had export controls, but these are within 0:28:09.119 --> 0:28:11.850 the export controls, so they must be like pretty nerfed. 0:28:11.850 --> 0:28:15.090 Apple might soon announce a deal to bring Google Gemini 0:28:15.090 --> 0:28:18.899 to iOS 18, so that means they're working with all three. 0:28:18.900 --> 0:28:22.500 They already announced OpenAI in the announcement. They mentioned meta 0:28:22.530 --> 0:28:25.109 a while back and now Google. And by the way, 0:28:25.109 --> 0:28:30.570 I'm bothered with Apple and also with OpenAI because they 0:28:30.570 --> 0:28:33.850 both did demos of things that aren't in the betas. 0:28:33.850 --> 0:28:37.119 Some reporting is saying this Siri stuff, and all the 0:28:37.119 --> 0:28:39.880 AI stuff won't even be until next spring. And I'm like, 0:28:39.880 --> 0:28:41.950 stop showing me demos of stuff that won't be out 0:28:41.950 --> 0:28:45.790 for months, if ever. But that's just me being annoyed. 0:28:45.790 --> 0:28:49.480 Greece is moving to a six day working week. Weird. Oh, 0:28:49.480 --> 0:28:52.480 this is actually for a different story. That's a mistake here. 0:28:53.020 --> 0:28:58.080 Do not like, um, this was about somebody, uh, Apple 0:28:58.080 --> 0:29:03.210 putting cameras on AirPods, which I was hoping was going 0:29:03.210 --> 0:29:05.970 to be actual cameras. But of course, that's going to 0:29:05.970 --> 0:29:09.630 be more headgear that's going to require like, uh, like 0:29:09.630 --> 0:29:12.570 a collar or like a neck piece or like a 0:29:12.570 --> 0:29:16.860 heavy earphone or headphone because you need to power this, uh, 0:29:16.860 --> 0:29:19.260 this camera that's running all the time. So you're going 0:29:19.260 --> 0:29:22.490 to have battery issues, plus lenses and stuff like that. 0:29:22.490 --> 0:29:25.430 But what I really want is I want the AirPods 0:29:25.430 --> 0:29:28.190 or this headset or whatever it is, but it's got 0:29:28.190 --> 0:29:29.750 to be every day. But I want it to see 0:29:29.750 --> 0:29:32.750 behind me. That's super critical. If you look at my 0:29:32.750 --> 0:29:34.940 big eye piece, I want it to see in front 0:29:34.940 --> 0:29:37.310 of me. I want it to see behind me. It'd 0:29:37.340 --> 0:29:39.380 be nice if it saw to the sides and it 0:29:39.380 --> 0:29:42.740 could connect them for 360. That would be nice. But 0:29:42.740 --> 0:29:46.390 most importantly, in front and behind And then that all 0:29:46.390 --> 0:29:48.910 being processed and it's like, hey, there's a dangerous thing 0:29:48.910 --> 0:29:50.800 in front of you. Hey, watch out for this. This 0:29:50.800 --> 0:29:53.860 car is coming in your direction. Whatever. Someone's sneaking up 0:29:53.860 --> 0:29:57.610 behind you. Be great for women. Like walking at night. 0:29:57.730 --> 0:30:00.730 It's like this person is following you. Yeah, but no, 0:30:00.730 --> 0:30:04.570 the thing that actually is getting shipped, it's for gesture controls. 0:30:04.570 --> 0:30:07.000 So you get like, you know, go like this and 0:30:07.000 --> 0:30:10.080 you could do things with your AirPods. That's the rumor. 0:30:10.080 --> 0:30:12.840 David Brooks sat down with Steve Bannon to understand his 0:30:12.840 --> 0:30:16.680 vision for the global populist movement. Um, number of U.S. 0:30:16.680 --> 0:30:20.100 high school graduates is expected to peak in 25 and 0:30:20.100 --> 0:30:23.670 then decline for years. I don't understand this. Schools and 0:30:23.670 --> 0:30:26.700 colleges are closing. Faculty members are being laid off. I, 0:30:26.910 --> 0:30:30.270 I what I don't understand why there's going to be 0:30:30.270 --> 0:30:33.830 fewer people graduating. Is there a few people going into 0:30:33.830 --> 0:30:37.400 the system? Is the population declining? I'm not sure exactly 0:30:37.400 --> 0:30:39.860 what's going on here. US dollar just hit a 38 0:30:39.860 --> 0:30:43.550 year peak against the yen. Peoples whose eyes dilated more 0:30:43.550 --> 0:30:47.060 performed better on tests of working memory, suggesting that pupil 0:30:47.060 --> 0:30:49.010 size is linked to how well we can process and 0:30:49.010 --> 0:30:54.080 remember information. And this is from Scientific American. The FDA 0:30:54.080 --> 0:30:58.900 approved Eli Lilly's Alzheimer's drug Kazuna. Why was that hard 0:30:58.900 --> 0:31:04.090 to pronounce? Kazuna. Donanemab is the real name which shows 0:31:04.090 --> 0:31:08.320 disease progression, slows disease progression by about a third. These 0:31:08.320 --> 0:31:11.920 drug companies are killing it with Ozempic and now Alzheimer's. 0:31:11.920 --> 0:31:15.370 High work in progress is killing your business, innovation and morale. 0:31:15.400 --> 0:31:18.700 The more tasks you juggle, the slower everything moves. Yeah, 0:31:18.700 --> 0:31:22.620 tell me about it. I'm struggling with this right now. Ideas. 0:31:22.620 --> 0:31:26.310 The I class gap. There's a separation that is really 0:31:26.310 --> 0:31:30.090 freaking me out about AI right now, which is dramatically 0:31:30.090 --> 0:31:37.470 increased separation between social groups or classes, socioeconomic classes because 0:31:37.470 --> 0:31:40.500 of AI. So basically one small group at the top 0:31:40.500 --> 0:31:43.500 of like 1% or whatever, how you want to call 0:31:43.500 --> 0:31:46.040 it 5%, we'll use AI to have a staff of 0:31:46.040 --> 0:31:51.140 10,000 executive assistants, tutors and analysts working constantly to run 0:31:51.140 --> 0:31:55.670 their businesses, optimize their entire lives, accountants doing their paperwork, 0:31:55.670 --> 0:31:59.870 filing their taxes, optimizing everything. Just think about that. A 0:31:59.870 --> 0:32:02.600 lot of people, this top 1%, are going to have 0:32:02.600 --> 0:32:06.780 like a team of 10 or 50 or 100 or 0:32:06.780 --> 0:32:10.959 10,000 executive assistants, which are basically independent eyes, all working 0:32:10.960 --> 0:32:14.890 towards the same goals, optimizing everything in your life. That's 0:32:14.890 --> 0:32:17.590 the top 1%. And a lot of other people either 0:32:17.590 --> 0:32:19.900 won't be using AI at all, or they'll only be 0:32:19.900 --> 0:32:25.150 using it for gaming, watching media porn entertainment, watching videos, 0:32:25.150 --> 0:32:31.270 watching Netflix, just wasting time like slightly more efficiently using AI. 0:32:31.270 --> 0:32:34.150 And this is largely the same split as between like 0:32:34.150 --> 0:32:37.660 voracious readers and everyone else today. But it's going to 0:32:37.660 --> 0:32:40.780 be way worse with AI, because AI is going to 0:32:40.780 --> 0:32:44.680 lift the top 1% way more. Even more than being 0:32:44.680 --> 0:32:48.970 a heavy reader. Discovery self publishing a tech book mma AI. 0:32:48.970 --> 0:32:52.960 This guy is predicting MMA fight outcomes with very high 0:32:52.960 --> 0:32:57.110 accuracy just using AI, and all he's doing is he's 0:32:57.110 --> 0:32:59.540 looking at the, um, the people that they fought in 0:32:59.540 --> 0:33:03.560 the past. Sam Parr launched Sam's List, a database of CPAs, 0:33:03.560 --> 0:33:07.670 accountants and tax strategists, and he's already made $32,000 in 0:33:07.670 --> 0:33:11.510 just two months. And illustrated transformer visual intuitive guide to 0:33:11.510 --> 0:33:15.770 understanding how Transformers work. 11 labs has a new voice isolator. 0:33:15.770 --> 0:33:21.110 This thing's scary. Rachel Toback, my buddy, talked about this, uh, talking. 0:33:21.110 --> 0:33:25.060 I forgot how to pronounce this. Um, talking. Yeah, I 0:33:25.060 --> 0:33:27.190 feel like I should know how to pronounce that. Anyway, 0:33:27.520 --> 0:33:31.420 Ursula K Le Guin did a translation of this, which 0:33:31.420 --> 0:33:34.300 I've not read this thing yet. I really need to 0:33:34.300 --> 0:33:36.520 read this thing in full. And the recommendation of the 0:33:36.520 --> 0:33:41.080 week is ground news. Ground news. This is like the 0:33:41.080 --> 0:33:45.660 best website right now for news, especially going into this, uh, 0:33:45.660 --> 0:33:52.470 political election series, uh, situation in the US. So it 0:33:52.470 --> 0:33:57.060 shows you bias in how stories are covered. In fact, 0:33:57.060 --> 0:33:59.430 let me just pull this up. It's the whole purpose 0:33:59.430 --> 0:34:02.820 of being on, uh, on video. Look at this thing. Okay. 0:34:02.820 --> 0:34:07.530 I'm logging in, maybe and doxing myself while I log in. 0:34:07.530 --> 0:34:11.089 That's cool. Okay, look at these. You see these outlines? 0:34:11.090 --> 0:34:14.480 It's got an outline for the type of coverage that 0:34:14.480 --> 0:34:19.940 it's mostly getting. Look at this bar. Left 17% center 16%. 0:34:19.940 --> 0:34:24.140 Right 67% okay. And you have these bars all over 0:34:24.140 --> 0:34:28.640 the place. And what's really cool blind spot. Look at 0:34:28.640 --> 0:34:32.360 this blind spot thing for the left. This is exposing 0:34:32.360 --> 0:34:34.960 you to things that the right is talking about, that 0:34:34.960 --> 0:34:37.509 the left is not. And same for over here, for 0:34:37.510 --> 0:34:40.299 the right exposing you to things that the left is 0:34:40.300 --> 0:34:43.270 talking about, that the right is not. So this this 0:34:43.270 --> 0:34:46.900 helps you basically balance yourself, or at least have more 0:34:46.900 --> 0:34:51.009 of an option of balancing yourself. Highly recommended. Especially if 0:34:51.010 --> 0:34:53.740 you know somebody who claims to be wanting to be 0:34:53.739 --> 0:34:56.200 less biased. And they're like, well, I'm just looking for 0:34:56.200 --> 0:34:59.759 a good source or whatever. Send them this. Okay. And 0:34:59.760 --> 0:35:02.730 the aphorism of the week, clear writing and clear speaking 0:35:02.730 --> 0:35:07.320 are simply outputs of clear thinking. Naval clear writing and 0:35:07.320 --> 0:35:13.650 clear speaking are simply outputs of clear thinking. Naval Unsupervised 0:35:13.650 --> 0:35:16.170 Learning is produced and edited by Daniel Miller on a 0:35:16.170 --> 0:35:21.029 Neumann U87 AI microphone using Hindenburg. Intro and outro music 0:35:21.030 --> 0:35:24.110 is by Zomby with the Y, and to get the 0:35:24.110 --> 0:35:26.180 text and links from this episode, sign up for the 0:35:26.180 --> 0:35:31.820 newsletter version of the show at Daniel miessler.com/newsletter. We'll see 0:35:31.820 --> 0:35:32.569 you next time.