WEBVTT - UL NO. 435: Making New Things is Post-AI Safety 0:00:00.200 --> 0:00:03.320 All right. Welcome to unsupervised learning. It's Daniel Meisler. All right. 0:00:03.320 --> 0:00:05.750 What do we got here? All right. So got a 0:00:05.750 --> 0:00:09.500 new definition of prompt engineering that I'm basically using. Now 0:00:09.500 --> 0:00:12.110 when I talk to people. And I can share that 0:00:12.110 --> 0:00:15.380 a little bit later. AI's impact on the job market. 0:00:15.380 --> 0:00:17.960 Working on a number of new talks. I didn't think 0:00:17.960 --> 0:00:19.369 I was going to do this. I thought I was 0:00:19.370 --> 0:00:21.020 actually going to be like 2 or 3 for the 0:00:21.020 --> 0:00:24.080 whole year, but it's like turning into a lot more. 0:00:24.079 --> 0:00:26.840 So curious to see where that's going to end up. 0:00:26.840 --> 0:00:29.030 It's going to end up being like 4 or 5 talks. 0:00:29.030 --> 0:00:32.480 What I like about this system that I'm using, though, 0:00:32.479 --> 0:00:35.270 is that I'm only speaking about stuff that I normally 0:00:35.270 --> 0:00:38.870 talk about anyway, and that's what essentially what the newsletter 0:00:38.870 --> 0:00:41.390 is and what the podcast is. It's like stuff I'm 0:00:41.390 --> 0:00:44.480 already talking about and it's the same for talks, it's 0:00:44.479 --> 0:00:48.320 the same for presentations, it's the same for blogs. It's 0:00:48.320 --> 0:00:51.440 the same for tweets. Like I'm not doing a strategy 0:00:51.440 --> 0:00:57.380 that relates to particular media or syndication destinations. I'm not 0:00:57.380 --> 0:01:00.290 thinking at all about that, actually. And I've talked about 0:01:00.290 --> 0:01:04.190 this multiple times before, but people keep bringing it up, like, 0:01:04.190 --> 0:01:06.289 how do you come up with ideas or whatever? First 0:01:06.290 --> 0:01:08.060 of all, the way I come up with ideas is 0:01:08.060 --> 0:01:11.000 I read a crap ton of stuff, like all the time. 0:01:11.000 --> 0:01:14.330 I'm always reading mostly books, but also articles and stuff 0:01:14.330 --> 0:01:18.589 like that, and social media and other builders and creators 0:01:18.590 --> 0:01:21.589 in AI and security and stuff like that. So that 0:01:21.590 --> 0:01:24.020 is the main way that I get ideas is by 0:01:24.020 --> 0:01:27.200 having tons and tons of inputs. And the metaphor that 0:01:27.200 --> 0:01:29.720 I use for this before was like a particle accelerator, 0:01:29.720 --> 0:01:32.750 like I'm just like the atom, the tiny little brain 0:01:32.750 --> 0:01:35.990 in the middle. But I'm being bombarded by the coolest 0:01:35.990 --> 0:01:41.780 ideas from like Greek classics to like ancient philosophy productivity 0:01:41.780 --> 0:01:46.220 books to AI stuff to whatever. And when that stuff 0:01:46.220 --> 0:01:49.070 hits my brain. If you've ever seen those outputs from 0:01:49.070 --> 0:01:54.290 one of these particle accelerators, it's like this super colorful, amazing, 0:01:54.290 --> 0:01:57.980 like explosion of things that come out of it. And 0:01:57.980 --> 0:02:00.890 those are all the subatomic particles. And what's interesting about 0:02:00.890 --> 0:02:04.309 this analogy is those only last for this very small 0:02:04.310 --> 0:02:07.220 amount of time. And that's why I'm so obsessed with capturing, 0:02:07.220 --> 0:02:10.370 because if I'm having a conversation or I'm out for 0:02:10.370 --> 0:02:12.680 a walk and it's funny because going out for a 0:02:12.680 --> 0:02:15.770 walk or taking a shower with no tech is very 0:02:15.770 --> 0:02:20.180 similar to a particle accelerator, where you're actually capturing ideas 0:02:20.180 --> 0:02:22.820 from really smart things, because what happens is your brain 0:02:22.820 --> 0:02:25.250 will start freaking out and it's like, hey, let's think 0:02:25.250 --> 0:02:27.320 about some stuff. Hey, what about that one time where 0:02:27.320 --> 0:02:30.440 you read that one thing and it starts just churning 0:02:30.440 --> 0:02:32.690 and coming up with cool ideas, which is why you 0:02:32.690 --> 0:02:35.450 have ideas in the shower, because your brain is freaking 0:02:35.450 --> 0:02:37.700 out because it doesn't have any input. So it starts 0:02:37.700 --> 0:02:40.760 like just generating things. I wonder if we're doing the 0:02:40.760 --> 0:02:43.430 same things in dreams or whatever. I'm not sure. Maybe 0:02:43.430 --> 0:02:46.760 a similar mechanism, but the point is, I have ideas 0:02:46.760 --> 0:02:50.150 because I consume a lot and the ideas don't even 0:02:50.150 --> 0:02:53.240 necessarily have anything to do with what I'm consuming. They 0:02:53.240 --> 0:02:55.820 tend to be peripheral to what I'm consuming, and that's 0:02:55.820 --> 0:02:59.690 why I read deeply but broadly. Right? Because I'll switch 0:02:59.690 --> 0:03:02.780 to fiction and there'll be a cool idea in the fiction, 0:03:02.780 --> 0:03:05.570 and then suddenly the fiction is the peanut butter. And 0:03:05.570 --> 0:03:09.200 this nonfiction AI thing that I'm working on or coding 0:03:09.200 --> 0:03:12.470 on is like the chocolate, and they mix together and 0:03:12.470 --> 0:03:14.299 all of a sudden I have a cool idea. And 0:03:14.300 --> 0:03:16.940 this is why I really believe that this whole idea 0:03:16.940 --> 0:03:19.910 of like, oh, people are super smart or people are 0:03:19.910 --> 0:03:22.370 super genius or whatever, look at this person who has 0:03:22.370 --> 0:03:26.300 all the ideas. Nope. The ideas come from inputs. If 0:03:26.300 --> 0:03:28.490 you stop reading and I know this for a fact 0:03:28.490 --> 0:03:31.910 for myself, if I stop reading within a few months, 0:03:31.910 --> 0:03:35.210 even a couple of weeks, like my brain will just 0:03:35.210 --> 0:03:38.420 slow down. Like I will not have the same number 0:03:38.420 --> 0:03:41.480 or quality of ideas coming in. And this is why 0:03:41.510 --> 0:03:44.600 people like Buffett and Charlie Munger and all these people, 0:03:44.600 --> 0:03:47.029 they're like, look, if you want to be smart, you 0:03:47.030 --> 0:03:49.850 need to read. You need to be consuming really high 0:03:49.850 --> 0:03:52.970 quality stuff at a very high rate of speed. And 0:03:52.970 --> 0:03:55.160 that's not always the case with like the rate you 0:03:55.160 --> 0:03:58.340 can actually get really high quality ideas. Riva has talked 0:03:58.340 --> 0:04:02.390 about this where like slow reading, a deep book like 0:04:02.390 --> 0:04:07.340 the Bible, or like a classic from the Greeks or something, whatever. 0:04:07.340 --> 0:04:10.310 You could slow read and try to memorize and like 0:04:10.310 --> 0:04:13.460 go super deep with the thing, try to fully extract 0:04:13.460 --> 0:04:16.219 all the wisdom from it and do that on a 0:04:16.220 --> 0:04:19.400 regular basis. And you still get lots of ideas. But 0:04:19.400 --> 0:04:22.790 what you can't do is do nothing and think nothing, 0:04:22.790 --> 0:04:26.239 and consume nothing and just consume. You don't want to 0:04:26.240 --> 0:04:29.600 just be like watching Netflix or something. Of course, there 0:04:29.600 --> 0:04:31.580 could be some good stuff on Netflix where you actually 0:04:31.580 --> 0:04:34.460 get some ideas, but in general, that is a thing 0:04:34.460 --> 0:04:37.850 that calms your mind and like sedates it right? You're 0:04:37.850 --> 0:04:41.870 not getting high quality content coming in. It's more like relaxation. 0:04:41.870 --> 0:04:46.550 So all right, cool giant diatribe. And we've got through 0:04:46.550 --> 0:04:49.159 one bullet in the damn newsletter. All right. Publish a 0:04:49.160 --> 0:04:53.930 new official pattern template in fabric. So official Pattern Template 0:04:53.930 --> 0:04:57.020 is the name of it. And it's basically all my 0:04:57.020 --> 0:05:01.200 latest formats and instructions. And like the structure of the 0:05:01.230 --> 0:05:05.190 of the pattern or a prompt really an I prompt 0:05:05.190 --> 0:05:07.830 that works best for me. And I did a few 0:05:07.830 --> 0:05:11.010 updates here. I've got a lot more, I would say 0:05:11.010 --> 0:05:15.210 comments or descriptions of the sections. I broke out more sections. 0:05:15.210 --> 0:05:19.830 I added a examples and negative examples section, and I 0:05:19.830 --> 0:05:22.290 also broke out the identity and the goal. So a 0:05:22.290 --> 0:05:25.560 number of upgrades in there and people have been asking like, okay, 0:05:25.560 --> 0:05:28.170 how do you make a perfect prompt? So I made this, 0:05:28.170 --> 0:05:30.780 that basically shows that. And of course that's going to 0:05:30.779 --> 0:05:33.839 be updated constantly. Right. Because this is like there's no 0:05:33.839 --> 0:05:37.170 testing for these things. So everything I'm doing is like 0:05:37.170 --> 0:05:40.890 it's like it's like placebo. If it works for me, 0:05:40.890 --> 0:05:42.810 I'm like, yeah, this is really cool. I like it, 0:05:42.810 --> 0:05:45.270 and I continue to do it because it works for me. 0:05:45.270 --> 0:05:47.729 And when I test it against other things, I am 0:05:47.730 --> 0:05:52.380 doing a B testing. But first of all, LMS are non-deterministic, 0:05:52.380 --> 0:05:55.440 so I can get different answers for whatever it's like. 0:05:55.440 --> 0:05:57.270 In order to do this properly, you need to run 0:05:57.270 --> 0:06:00.870 it like tens of times, dozens of times, hundreds of times. 0:06:00.870 --> 0:06:02.700 And of course you have to worry about cost here, 0:06:02.700 --> 0:06:05.130 but then you have to have a really good objective 0:06:05.130 --> 0:06:08.520 rating system that's looking at those and actually finding the 0:06:08.520 --> 0:06:11.790 differences between these prompt techniques. And a bunch of papers 0:06:11.790 --> 0:06:14.370 have tried to do this. I have not necessarily agreed 0:06:14.370 --> 0:06:18.090 with the results of those papers. And again, that could 0:06:18.089 --> 0:06:20.250 be because the whole thing is a moving target in 0:06:20.250 --> 0:06:23.280 the first place, right? Bottom line is this what works 0:06:23.279 --> 0:06:26.490 for me? It is the official pattern template in fabric 0:06:26.490 --> 0:06:29.910 right now. Did my buddy Jason Haddix red blue, purple 0:06:29.910 --> 0:06:33.030 I class last week. It was fantastic. It was Monday, 0:06:33.240 --> 0:06:36.450 no Thursday and Friday of last week. It was so good. 0:06:36.450 --> 0:06:39.630 Jason is definitely the best teacher that I know and 0:06:39.630 --> 0:06:42.330 it was like, you're learning stuff. But he's also just 0:06:42.330 --> 0:06:46.560 hanging out. He's talking or we're talking all his friends. 0:06:46.560 --> 0:06:49.919 Different people like myself were also contributing to the class 0:06:49.920 --> 0:06:52.440 a little bit. He would ask a question, we would 0:06:52.440 --> 0:06:55.500 give an answer. Jason would chime in on that. And 0:06:55.500 --> 0:06:58.080 the people taking the class were also super smart. So 0:06:58.080 --> 0:07:00.419 they were contributing lots of stuff, and it was just 0:07:00.420 --> 0:07:03.599 like it was just great. It was absolutely great. Both 0:07:03.600 --> 0:07:06.089 days of content really enjoyed, and the next time it 0:07:06.089 --> 0:07:08.669 comes out, you should definitely sign up. So I got 0:07:08.670 --> 0:07:12.870 a new sponsor conversation here with Abhishek from Material Security. 0:07:12.900 --> 0:07:15.450 He's the CEO over there and we talked about a 0:07:15.450 --> 0:07:18.720 bunch of cool stuff. It was really a great conversation. 0:07:18.720 --> 0:07:21.420 One of my favorite topics was like, why are product 0:07:21.420 --> 0:07:24.210 managers so good at being CEOs? And so he gave 0:07:24.210 --> 0:07:27.210 a great answer to that. I got another piece up 0:07:27.210 --> 0:07:30.540 here on getting a perfect sound from your microphone. And 0:07:30.540 --> 0:07:33.600 this sound that you're currently listening to is a good 0:07:33.600 --> 0:07:36.780 version of that, although I don't do post-production for this 0:07:36.780 --> 0:07:40.560 particular thing because this is an OBS video going directly 0:07:40.560 --> 0:07:44.070 into YouTube. So I'm not doing post-production on here, but 0:07:44.070 --> 0:07:47.580 it should be a pretty decent initial sound. I got 0:07:47.580 --> 0:07:50.490 a political one here. It's called The left's. It's political, 0:07:50.490 --> 0:07:52.770 but it's centrist, so I don't think it should bother 0:07:52.770 --> 0:07:55.740 you unless you're crazy left or crazy right? But you 0:07:55.740 --> 0:07:58.350 should check that out. Lots of people from the left 0:07:58.350 --> 0:08:00.630 and the right have reached out and said that they 0:08:00.630 --> 0:08:04.140 loved it. So that should be a good metric for you. Yeah, 0:08:04.140 --> 0:08:08.580 it's called Left's Brexit security got a bad checkpoint. Von 0:08:08.580 --> 0:08:14.340 massive botnet was busted up. Ukraine is using AI against jamming. 0:08:14.340 --> 0:08:19.800 So yeah, it's basically like kill decision from Daniel Suarez where, um, 0:08:19.800 --> 0:08:24.720 the drones were getting jammed, GPS jammed, and also just 0:08:24.720 --> 0:08:28.800 any RF was getting jammed. So they made them autonomous 0:08:28.800 --> 0:08:33.600 so they could just navigate by vision and looking at landmarks. 0:08:33.600 --> 0:08:36.540 And then the idea was you have a drone swarm. 0:08:36.540 --> 0:08:39.780 And this is from the book and probably also from reality, 0:08:39.780 --> 0:08:42.660 also brought to you by the Pentagon, a drone swarm 0:08:42.660 --> 0:08:45.870 with tiny little explosives on them. And they would just 0:08:45.870 --> 0:08:48.750 all fly and navigate. They would have a target that 0:08:48.750 --> 0:08:51.570 they were trying to kill, and they would navigate by 0:08:51.570 --> 0:08:54.420 the landmarks. And when they find the person, they could 0:08:54.420 --> 0:08:57.150 do facial recognition on board, don't need to call out 0:08:57.150 --> 0:09:01.980 anywhere and then just crash into them, kamikaze and kill 0:09:01.980 --> 0:09:05.550 the person. And jammers don't matter because they're not using 0:09:05.550 --> 0:09:10.890 navigation based on RF smart home tech in warfare. So 0:09:10.890 --> 0:09:14.550 Home Assistant is being used in Ukraine to warn people 0:09:14.550 --> 0:09:19.260 of alerts of attacks incoming. Yeah. Incogni is another data 0:09:19.260 --> 0:09:22.560 deletion service similar to delete me moves your info from 0:09:22.559 --> 0:09:26.339 over 170 data brokers. They are not a sponsor. Otherwise 0:09:26.340 --> 0:09:29.370 you would say sponsor on the top of it Ticketmaster. 0:09:29.370 --> 0:09:32.250 So they got hit by a group called Shiny Hunters, 0:09:32.250 --> 0:09:36.300 and it looks like around 560 million people's data has 0:09:36.300 --> 0:09:39.840 been stolen. 560 million. I'm so glad they were getting 0:09:39.840 --> 0:09:42.270 broken up, or at least we're trying to break them up. 0:09:42.270 --> 0:09:45.510 And so glad that Taylor made that happen. I think 0:09:45.510 --> 0:09:49.320 she did. A few people spread the most information about 0:09:49.320 --> 0:09:52.470 fake news about vaccines and stuff like that, going, this 0:09:52.470 --> 0:09:57.449 is back to like 2020 and 2021, but, uh, older 0:09:57.450 --> 0:10:03.210 Republican women. 80% of the misinformation. Yeah, I wrote this myself. Small, 0:10:03.210 --> 0:10:07.230 persistent groups can have a significant impact on propaganda. You 0:10:07.230 --> 0:10:09.959 wouldn't think that. You would think that small groups like 0:10:09.960 --> 0:10:12.360 that would just be like a glass of water in 0:10:12.360 --> 0:10:15.900 the ocean, but turns out if you're persistent enough, it'll 0:10:15.900 --> 0:10:18.090 get out there. NSA released a guide on how to 0:10:18.090 --> 0:10:20.460 keep your phone from getting hacked, and they basically said 0:10:20.460 --> 0:10:23.640 you should reboot your phone once a week. Frightening. What 0:10:23.640 --> 0:10:25.949 do they know that I don't know? Or they're like, yeah, 0:10:25.950 --> 0:10:28.830 you should reboot your phone every week. And I'm like, why? 0:10:28.860 --> 0:10:32.730 Never mind. Don't want to know technology. All right. So grok, 0:10:32.730 --> 0:10:35.699 I talked about this. How fast it is wickedly fast 0:10:35.700 --> 0:10:39.420 this thing okay. So it can now do be lama 0:10:39.420 --> 0:10:43.950 3 to 1200 tokens per second. That's essentially like it 0:10:43.950 --> 0:10:46.920 just fills out the page as if it just instant. 0:10:46.920 --> 0:10:49.860 And so here's my question. Why doesn't Google or OpenAI 0:10:49.860 --> 0:10:53.130 or Microsoft or someone just like, walk up like, you know, 0:10:53.130 --> 0:10:56.760 a clearinghouse sweepstakes or something? And just like, here's a 0:10:56.760 --> 0:11:02.760 box of $1 billion, please give us your technology and disappear. Right. 0:11:02.820 --> 0:11:07.050 This is like Magic Beans proprietary chip tech that lets 0:11:07.050 --> 0:11:11.820 him basically run AI. What, tens of times faster than 0:11:11.820 --> 0:11:15.780 anyone else? Ridiculous. Looks like Satya Nadella is freaking out 0:11:15.780 --> 0:11:19.560 about a potential partnership between OpenAI and Apple. I think 0:11:19.559 --> 0:11:21.420 he's right to freak out, and he should make a 0:11:21.420 --> 0:11:25.050 phone before it's too late. Or a personal assistant on 0:11:25.050 --> 0:11:28.949 some kind of device. He needs a personal device. Anyone 0:11:28.950 --> 0:11:32.190 who wants to play going forward need a personal device 0:11:32.190 --> 0:11:34.050 and an OS, and it needs to be able to 0:11:34.050 --> 0:11:37.860 see and hear and basically be integrated with all your context. 0:11:37.860 --> 0:11:40.650 And anyone who's not doing this is screwed. And people 0:11:40.650 --> 0:11:43.709 have a massive advantage. Are people who already have access 0:11:43.710 --> 0:11:50.010 into a robust operating system, namely Android and Apple. Researchers 0:11:50.010 --> 0:11:55.949 have new attention mechanisms that actually outperform standard multi-head attention. Cool. Well, 0:11:55.950 --> 0:11:59.490 let me know when it fixes haystack performance. Need I 0:11:59.490 --> 0:12:02.070 need haystack performance. I need to be able to give 0:12:02.070 --> 0:12:05.969 you giant text documents and you don't miss a single thing. 0:12:05.970 --> 0:12:09.000 So wake me up when that happens, all right? Recall 0:12:09.000 --> 0:12:12.810 a privacy disaster. This is the Microsoft thing that records everything. 0:12:12.809 --> 0:12:15.780 Stealing everything. This is what I wrote. Stealing everything you've 0:12:15.780 --> 0:12:17.940 ever typed or viewed on your windows PC is now 0:12:17.940 --> 0:12:21.720 a reality, thanks to a feature in Copilot plus recall 0:12:21.720 --> 0:12:25.110 that is a privacy nightmare. Yeah, disagree with that take. 0:12:25.350 --> 0:12:28.110 I think in ten years almost everyone will think computers 0:12:28.110 --> 0:12:31.080 who don't have this are basically worthless. And in fact, 0:12:31.080 --> 0:12:35.040 ten years most people won't be talking, will only be talking, 0:12:35.040 --> 0:12:38.579 will mostly be talking and gesturing to their computers. Computer 0:12:38.580 --> 0:12:41.880 will mostly be like monitors and HUD displays or whatever, 0:12:41.880 --> 0:12:45.300 and your computers are basically doing the work. So the 0:12:45.300 --> 0:12:48.059 idea that it's going to forget something you worked on 0:12:48.059 --> 0:12:50.190 or forget something that was said to you, or that 0:12:50.190 --> 0:12:53.819 you said to somebody else is completely asinine. Of course 0:12:53.820 --> 0:12:56.520 your AI is going to remember everything. What else would 0:12:56.520 --> 0:13:00.480 it do? All right, Llms aren't just internet simulators anymore. Yeah, 0:13:00.480 --> 0:13:05.010 so increasingly llms are being trained on custom non-internet data 0:13:05.370 --> 0:13:08.070 that's there's still a data wall there as well. But 0:13:08.070 --> 0:13:09.689 I don't know, I think a lot of people like 0:13:09.690 --> 0:13:13.650 social media companies. Meta is in great position for this, 0:13:13.650 --> 0:13:16.319 for example. I mean, because they have Instagram, they have 0:13:16.320 --> 0:13:19.080 all this thing. So basically whatever people are posting and 0:13:19.080 --> 0:13:22.980 doing and talking about, that becomes a massive source of data. 0:13:22.980 --> 0:13:25.980 Anyone who's like really close to the OS and especially 0:13:25.980 --> 0:13:29.460 in messaging apps, social media apps, that is a massive 0:13:29.460 --> 0:13:32.309 source of data that they can then sell to other 0:13:32.309 --> 0:13:35.250 AI companies or keep it for themselves to have an advantage. 0:13:35.250 --> 0:13:38.219 In the case of like Lama3, Nvidia is getting ready 0:13:38.220 --> 0:13:43.080 to overtake Apple in market cap. That's insane. Completely insane. 0:13:43.080 --> 0:13:45.210 But here's a way to make sense of this. Imagine 0:13:45.210 --> 0:13:48.180 that there's a giant filter on how many people are 0:13:48.179 --> 0:13:51.270 capable of creating a hit movie, writing a book, starting 0:13:51.270 --> 0:13:53.880 a business, sharing their art with billions of people. In 0:13:53.880 --> 0:13:56.790 other words, imagine the current number of major builders or 0:13:56.790 --> 0:14:06.210 creators is something like 0.00000000000 17 of our 8 billion people. 0:14:06.210 --> 0:14:08.400 I just made up that number. It's not a real number, 0:14:08.400 --> 0:14:11.130 despite how precise it is. So just work with me. 0:14:11.130 --> 0:14:14.040 So the reason Nvidia is rising so fast is because 0:14:14.040 --> 0:14:18.240 one of the primary pieces of tech is the GPU. Okay? 0:14:18.240 --> 0:14:21.870 It is like center mass of this entire thing. And 0:14:21.870 --> 0:14:25.260 what we're talking about is removing 3 to 5 zeros 0:14:25.260 --> 0:14:29.100 from this big zero number. We're talking about multiplying humanity's 0:14:29.100 --> 0:14:32.790 creativity output by thousands in the next decade. And that 0:14:32.790 --> 0:14:36.000 requires chips. And thank you for coming to my Ted 0:14:36.030 --> 0:14:39.990 talk about Nvidia. That's essentially why they are skyrocketing. Google 0:14:39.990 --> 0:14:44.580 released an SRE handbook. Basically saying simplicity is a core 0:14:44.580 --> 0:14:48.060 principle for reliability makes total sense to me. They should 0:14:48.060 --> 0:14:50.850 make one for product management. That's mean. But it's funny. 0:14:50.850 --> 0:14:56.970 All right. Karpathy's GPT two 124 million parameters looks like 0:14:57.000 --> 0:15:02.280 LMK and 90 minutes for. $20. Karpathy is an absolute beast. 0:15:02.310 --> 0:15:06.960 Tiny Grad's latest update. This is from Geohot. Slashes Python 0:15:06.960 --> 0:15:11.970 time and ditches external dependencies. Terminal animations. These look really cool. 0:15:11.970 --> 0:15:13.290 I'm not sure I'm going to use them, but they 0:15:13.290 --> 0:15:16.920 look really cool. And humans little known intelligence bureau that 0:15:16.920 --> 0:15:22.140 nailed the outcomes in Vietnam, Iraq and Ukraine. I want 0:15:22.140 --> 0:15:26.220 to build something that basically programmatically goes and gets these 0:15:26.220 --> 0:15:30.090 predictions and incorporates them into my life model and how 0:15:30.090 --> 0:15:33.330 I see the world, because it's essentially like these prediction 0:15:33.330 --> 0:15:37.680 markets are the smartest people doing the best kind of predictions, 0:15:37.680 --> 0:15:42.510 and they're doing it in a very transparent and logical way. 0:15:42.510 --> 0:15:45.690 And I learned about this from the book called Superforecasters. 0:15:45.690 --> 0:15:48.060 Is that Tetlock? Bill Tetlock, I think, might be the 0:15:48.060 --> 0:15:51.750 name of the author or one of the authors. Unbelievable stuff. Really, 0:15:51.750 --> 0:15:55.170 really good stuff. New study says sleep inequality in the 0:15:55.170 --> 0:15:58.350 US is tied to economic stress. Makes sense to me. 0:15:58.350 --> 0:16:02.310 Mice with PTSD like behavior showed significant improvement after having 0:16:02.310 --> 0:16:07.500 access to a running wheel. Basically, exercise induced neurogenesis like 0:16:07.500 --> 0:16:10.590 healing trauma with exercise. That was a bit of a stretch, 0:16:10.590 --> 0:16:13.650 but that's where they're going with this Ozempic and wegovy 0:16:13.650 --> 0:16:17.940 changing eating habits, fine tuning your taste buds, making sweets 0:16:17.940 --> 0:16:20.790 taste sweeter. Mm. Yeah. I mean, it's definitely having an 0:16:20.790 --> 0:16:23.850 effect on me. Like I had pizza. I've been having 0:16:23.850 --> 0:16:27.990 pizza because of watching this damn pizza influencer who samples 0:16:27.990 --> 0:16:30.540 pizza and is just making me want to eat pizza. Anyway, 0:16:30.540 --> 0:16:32.850 I get a pizza, I eat, like three slices and 0:16:32.850 --> 0:16:34.950 I'm just like, yeah, it was good, I guess, but 0:16:34.950 --> 0:16:38.250 I'm full. Whereas before before I took Wegovy, I would 0:16:38.250 --> 0:16:40.500 be like, I would eat three quarters of the pizza 0:16:40.500 --> 0:16:42.810 and be like, I should not eat any more. And 0:16:42.810 --> 0:16:44.910 then I would eat the rest and be like, what's 0:16:44.910 --> 0:16:49.380 up with ice cream? John Carmack dives into bullshit jobs. Yeah, 0:16:49.380 --> 0:16:52.230 really good book by David Graeber. Got to read that book. 0:16:52.230 --> 0:16:54.360 I need to read it again. No, I don't I 0:16:54.360 --> 0:16:57.090 should just summarize it. I have a fabric pattern for that. 0:16:57.090 --> 0:17:00.720 Imagine navigating high school social maze without a smartphone, missing 0:17:00.720 --> 0:17:03.600 out on chats and memes, beginning a unique perspective on 0:17:03.600 --> 0:17:06.630 life and friendships, the social lives of teens who don't 0:17:06.630 --> 0:17:08.940 have phones. I don't have to imagine this. I grew 0:17:08.940 --> 0:17:11.969 up in the 80s. It was glorious. Yeah, very much 0:17:12.000 --> 0:17:17.010 on the tip of Jonathan Haidt basically saying getting phones 0:17:17.010 --> 0:17:22.050 out of schools highly support that effort. Comprehensive review finds 0:17:22.050 --> 0:17:26.880 I exercises might actually help in preventing and controlling myopia. 0:17:26.880 --> 0:17:30.449 This is actually an incorrect read. This was my fault. 0:17:30.450 --> 0:17:34.260 This is an incorrect read. The study. There are a 0:17:34.260 --> 0:17:36.030 whole bunch of studies saying it helps and a whole 0:17:36.030 --> 0:17:38.639 bunch of studies saying it's not. But this particular study 0:17:38.640 --> 0:17:42.960 actually says they couldn't find much evidence, strong evidence that 0:17:42.960 --> 0:17:45.959 it did help. So that's that's boo on my part. 0:17:45.960 --> 0:17:49.440 Lung cancer breakthrough unprotected. Oh yeah a lot of people 0:17:49.440 --> 0:17:51.750 are talking about this. I don't get excited until it 0:17:51.750 --> 0:17:55.320 comes to market and everyone loves it like Wegovy. But 0:17:55.320 --> 0:18:00.000 lots of people are talking about how this thing is insane. 60% 0:18:00.000 --> 0:18:04.139 of advanced stage patients progression free for five years. Marc 0:18:04.140 --> 0:18:06.959 Andreessen says to stay in school. But then he says, 0:18:06.960 --> 0:18:09.390 if you're the type of person to drop out, you 0:18:09.390 --> 0:18:12.930 also won't listen to this advice. And I love this letter. 0:18:12.930 --> 0:18:15.660 Love my Wife is dead letter from Richard Feynman to 0:18:15.660 --> 0:18:19.139 his dead wife. Very touching ideas and analysis. My favorite 0:18:19.140 --> 0:18:22.020 way to explain prompt engineering. Basically, don't think of it 0:18:22.020 --> 0:18:25.050 like I think of it in this way. Instead, prompt 0:18:25.050 --> 0:18:28.409 engineering is how to explain to a superior intelligence what 0:18:28.410 --> 0:18:30.810 your problem is and what you would like to happen. 0:18:30.810 --> 0:18:33.930 And if you can't do that well, you will lose 0:18:33.930 --> 0:18:36.480 to people who can't. That's it. How to explain to 0:18:36.480 --> 0:18:39.419 a superior intelligence what your problem is and what you 0:18:39.420 --> 0:18:42.690 would like to happen. That's prompt engineering. All right. My 0:18:42.690 --> 0:18:45.270 X thread on why it's so hard to find tech 0:18:45.270 --> 0:18:47.760 jobs right now. All right. So this this is the 0:18:47.760 --> 0:18:50.369 vibe here problem for the job market isn't that AI 0:18:50.369 --> 0:18:52.800 is happening. The problem is that it's happening at the 0:18:52.800 --> 0:18:56.969 exact same moment that most companies are figuring out that 80% 0:18:56.970 --> 0:19:00.030 of their employees are worthless. We need to stop expecting 0:19:00.030 --> 0:19:02.190 things to go back to the way they were. The 0:19:02.190 --> 0:19:07.050 new reality is companies mostly hiring super ambitious, exceptional, proven 0:19:07.050 --> 0:19:10.710 people who are total gods. With AI, this means most 0:19:10.710 --> 0:19:13.290 formal education becomes a waste of time and money, because 0:19:13.290 --> 0:19:16.890 a degree doesn't certify that you are super ambitious or 0:19:16.890 --> 0:19:20.460 exceptional or proven, it doesn't certify any of those things. 0:19:20.460 --> 0:19:23.250 So in this model, only elite schools will matter because 0:19:23.250 --> 0:19:26.910 the filtering for being exceptional will have happened just by 0:19:26.910 --> 0:19:29.970 being accepted into the school. That doesn't count for like 0:19:29.970 --> 0:19:33.960 people who got in because their mom already graduated, right? 0:19:33.960 --> 0:19:36.959 But for people who got in via merit to a 0:19:36.960 --> 0:19:39.570 top school, that is a good filter. So really this 0:19:39.570 --> 0:19:42.510 is two separate things. One, you're exceptional enough to be 0:19:42.510 --> 0:19:46.200 accepted into the top school and two, you finish some classes. 0:19:46.200 --> 0:19:49.590 Problem is, number two, you can get anywhere, right? The 0:19:49.590 --> 0:19:53.729 actual training in school between one like a state school 0:19:53.730 --> 0:19:56.609 and like Harvard is not that much different. Number one 0:19:56.609 --> 0:19:59.830 is the thing that employers actually care. The filter to 0:19:59.830 --> 0:20:02.740 be able to get into an elite institution. So the 0:20:02.740 --> 0:20:05.770 result will be companies hiring the top 10 to 20% 0:20:05.770 --> 0:20:09.130 of people in competence. And this will be filtered by 0:20:09.130 --> 0:20:13.120 elite school attendance and proof of competence via something you've 0:20:13.119 --> 0:20:15.970 put into the world, like on your website, YouTube as 0:20:15.970 --> 0:20:19.119 a tool or a company you built, or an AI 0:20:19.119 --> 0:20:23.110 algorithm rating you. This is going to be huge as well. 0:20:23.109 --> 0:20:26.620 So the hyperbolic form of this is if you're not 0:20:26.619 --> 0:20:29.890 19 and at Harvard, or if you don't have your 0:20:29.890 --> 0:20:32.830 own projects or companies you've built and talked about online, 0:20:32.830 --> 0:20:36.459 you are not going to be interesting to employers. You 0:20:36.460 --> 0:20:40.030 will be part of the 90% basically fighting for scraps 0:20:40.030 --> 0:20:43.090 and the recommendation of the week for everyone you know 0:20:43.090 --> 0:20:45.550 who's going to be thinking about their career in the future, 0:20:45.550 --> 0:20:49.179 which is basically anyone who's not independently wealthy try to 0:20:49.180 --> 0:20:52.150 get them to think about things in the following way. 0:20:52.180 --> 0:20:56.080 AI is going to take out most executors of knowledge work. 0:20:56.080 --> 0:21:00.040 The people who will survive are the people with ideas 0:21:00.040 --> 0:21:04.480 who are actively building that thing, actively building that thing. 0:21:04.480 --> 0:21:07.090 So it's not just ideas. So the safest thing to 0:21:07.090 --> 0:21:09.669 be is a creator or a builder. That means a 0:21:09.670 --> 0:21:14.200 founder or an entrepreneur, a programmer, somebody with ideas who 0:21:14.200 --> 0:21:18.010 is productive and ambitious, like an artist like you could 0:21:18.010 --> 0:21:21.070 be an artist. It doesn't have to be some hardcore engineer, 0:21:21.070 --> 0:21:23.709 but it's got to be some kind of category like that. 0:21:23.710 --> 0:21:25.810 And the trick is you have to be able to 0:21:25.810 --> 0:21:29.410 have the ideas, be able to make it yourself, or 0:21:29.410 --> 0:21:32.560 attract the talent and the AI and harness the AI 0:21:32.590 --> 0:21:34.330 to be able to make it. And then you've got 0:21:34.330 --> 0:21:35.770 to be able to market the hell out of it, 0:21:35.770 --> 0:21:38.229 which is a lot of writing. So a lot of 0:21:38.230 --> 0:21:42.580 like communication and persuasion. So bottom line, the winners in 0:21:42.580 --> 0:21:45.159 this new game will be the people making new things 0:21:45.160 --> 0:21:47.800 and bringing them into the world. I'm going to say 0:21:47.800 --> 0:21:50.530 that again, the winners in this new game will be 0:21:50.530 --> 0:21:54.100 the people making new things and bringing them into the world. 0:21:54.100 --> 0:21:57.220 And the aphorism of the week. The most beautiful people 0:21:57.220 --> 0:22:00.700 we have known are those who have known defeat, known suffering, 0:22:00.700 --> 0:22:04.119 known struggle, known loss, and have found their way out 0:22:04.119 --> 0:22:06.940 of the depths. Elisabeth Kubler-Ross.