1 00:00:00,200 --> 00:00:02,759 S1: Does your app get fake sign ups, throw away emails 2 00:00:02,759 --> 00:00:06,880 S1: or users abusing your free tier? Or worse, bot attacks 3 00:00:06,880 --> 00:00:10,560 S1: and brute force attempts work OS radar can block all 4 00:00:10,560 --> 00:00:14,080 S1: of this and more. Simple API gives you advanced device 5 00:00:14,080 --> 00:00:18,800 S1: fingerprinting that can detect bad actors, bots, and suspicious behavior. 6 00:00:19,320 --> 00:00:23,040 S1: Your users trust you. Let's keep it that way. Check 7 00:00:23,040 --> 00:00:32,000 S1: out OS Radar at work OS radar. That's work. OS radar. 8 00:00:32,040 --> 00:00:36,200 S1: Unsupervised learning is a podcast about trends and ideas in cybersecurity, 9 00:00:36,240 --> 00:00:41,159 S1: national security, AI, technology, and society, and how best to 10 00:00:41,200 --> 00:00:48,599 S1: upgrade ourselves to be ready for what's coming. All right. 11 00:00:48,600 --> 00:00:51,199 S1: Welcome to Unsupervised Learning. This is Daniel Miessler, and I'm 12 00:00:51,200 --> 00:00:55,279 S1: building AI to upgrade humans. Episode 466. Hope your week 13 00:00:55,280 --> 00:00:58,200 S1: is starting off better than Nvidia's did, which is probably 14 00:00:58,200 --> 00:01:01,700 S1: pretty easy to do. Went to a phenomenal offensive security 15 00:01:01,700 --> 00:01:06,940 S1: and AI conference sponsored by Rob Ragan. Not really sponsored 16 00:01:06,940 --> 00:01:09,100 S1: by him, but put on by him. Had lots of 17 00:01:09,420 --> 00:01:12,140 S1: cool sponsors for different things. But anyway, it was a 18 00:01:12,140 --> 00:01:15,060 S1: whole day of hacking. Basically, it was a hackathon fundamentally, 19 00:01:15,580 --> 00:01:19,620 S1: but it was also just a great networking event for anybody, 20 00:01:19,660 --> 00:01:22,100 S1: kind of at the intersection of AI and security in 21 00:01:22,100 --> 00:01:25,500 S1: San Francisco. And it was just just great. Rob did 22 00:01:25,500 --> 00:01:28,220 S1: a great job. Kind of a nerd observation. Far too 23 00:01:28,220 --> 00:01:32,020 S1: many people don't know how to use name drop. If 24 00:01:32,020 --> 00:01:34,780 S1: you take your iPhone and set it next to another 25 00:01:34,780 --> 00:01:38,180 S1: one's iPhone, someone else's iPhone at the top. Like if 26 00:01:38,180 --> 00:01:40,700 S1: you just touch, you don't even have to touch, but 27 00:01:40,700 --> 00:01:44,700 S1: you get close enough. It's NFC based and you could 28 00:01:44,700 --> 00:01:47,420 S1: transfer your contact. Every time I do this, people are like, 29 00:01:47,420 --> 00:01:50,300 S1: what have you done? Like, they can't believe it. It's 30 00:01:50,300 --> 00:01:53,980 S1: got like this little wobbly like liquid effect. It's like 31 00:01:53,980 --> 00:01:56,900 S1: the coolest thing ever. And it's how I like to 32 00:01:56,940 --> 00:02:01,480 S1: exchange contacts with people who have iPhones. And yeah, I 33 00:02:01,480 --> 00:02:04,440 S1: don't know why more people aren't doing it or don't 34 00:02:04,440 --> 00:02:07,240 S1: even know the feature exists. And this is in San Francisco, 35 00:02:07,240 --> 00:02:10,440 S1: by the way. This is like the center mass of 36 00:02:10,440 --> 00:02:12,960 S1: Apple in the center of mass of like the tech scene. 37 00:02:13,720 --> 00:02:16,280 S1: And people in San Francisco don't know how to do 38 00:02:16,280 --> 00:02:19,160 S1: this thing. They don't know that it exists. Really glad 39 00:02:19,160 --> 00:02:21,960 S1: I bought a bunch of TSMC last week because it 40 00:02:21,960 --> 00:02:26,560 S1: just massively crashed. Uh, yeah, that's that's awesome. All good though. 41 00:02:26,600 --> 00:02:29,400 S1: Playing the long game. I think it's fine. We'll actually 42 00:02:29,400 --> 00:02:31,800 S1: talk about that a little bit later. I just finished 43 00:02:31,800 --> 00:02:35,920 S1: reading a book for UL. Book club. Picture of Dorian gray. 44 00:02:36,160 --> 00:02:38,920 S1: Lots to say about that. That's a whole nother talk show. 45 00:02:38,919 --> 00:02:43,040 S1: But one thing I will say is that read classics. 46 00:02:43,160 --> 00:02:47,000 S1: You need to read classics. They are more dense with 47 00:02:47,000 --> 00:02:51,519 S1: value than almost any other type of book, probably any 48 00:02:51,520 --> 00:02:54,079 S1: other type of book. And every time I read one, 49 00:02:54,080 --> 00:02:56,850 S1: I'm like, how many of these are just sitting there 50 00:02:57,690 --> 00:03:01,170 S1: full of this much knowledge, and I'm not reading them. Infuriating. 51 00:03:01,450 --> 00:03:05,010 S1: So luckily we have UL Book Club to prod us 52 00:03:05,010 --> 00:03:08,690 S1: into doing that. Here's what we do. In UL Book 53 00:03:08,690 --> 00:03:17,690 S1: Club we oscillate between nonfiction, fiction, nonfiction classic. Nonfiction fiction, nonfiction. Classic. 54 00:03:18,169 --> 00:03:21,609 S1: So what we're doing is half nonfiction and half or 55 00:03:21,650 --> 00:03:26,450 S1: half nonfiction, one quarter classic and one quarter fiction. And 56 00:03:26,450 --> 00:03:28,769 S1: that has been a good mix for us. And we've 57 00:03:28,770 --> 00:03:33,169 S1: been doing this since like 2016 or something. So a 58 00:03:33,169 --> 00:03:35,690 S1: long time this book club has been going. This week's 59 00:03:35,690 --> 00:03:40,010 S1: discovery section is really good. I make some tweaks to 60 00:03:40,050 --> 00:03:42,850 S1: make it even better. So this is kind of the 61 00:03:42,850 --> 00:03:47,650 S1: first instance of that. Had a great conversation with Faisal Khan, 62 00:03:48,050 --> 00:03:51,850 S1: a GRC guy over at Vanta, and that was a 63 00:03:51,890 --> 00:03:56,910 S1: sponsored interview And fantastic. Last couple of sponsored interviews have 64 00:03:56,910 --> 00:04:00,270 S1: just been really, really good. So that was fun. I'm 65 00:04:00,270 --> 00:04:04,750 S1: going to jump into security Sonicwall vulnerability being actively exploited. 66 00:04:04,990 --> 00:04:09,350 S1: Really nasty vulnerability in their SMA 1000 appliances. And it 67 00:04:09,350 --> 00:04:11,470 S1: is being used in the wild and it is a 68 00:04:11,470 --> 00:04:14,950 S1: 9.8 on the Richter scale. My buddy Sam Curry and 69 00:04:14,990 --> 00:04:18,349 S1: Shubham Shah found that they could remotely unlock, start and 70 00:04:18,350 --> 00:04:24,030 S1: track Subarus. So a couple of the goats of Bounty 71 00:04:24,070 --> 00:04:27,790 S1: and yeah, found a really nasty one. And it looks 72 00:04:27,790 --> 00:04:32,469 S1: like the network that they're using actually involves other manufacturers 73 00:04:32,470 --> 00:04:34,950 S1: as well. So it's not just Subaru, but I think 74 00:04:34,950 --> 00:04:37,990 S1: the main effect was on the Subaru stuff and got 75 00:04:37,990 --> 00:04:40,110 S1: a thread here on the downsides of everyone getting a 76 00:04:40,110 --> 00:04:42,750 S1: coding assistant, where I say one of the biggest impacts 77 00:04:42,750 --> 00:04:45,590 S1: of AI that goes kind of unnoticed is that we're 78 00:04:45,589 --> 00:04:49,390 S1: about to see an explosion of poorly built applications, specifically 79 00:04:49,390 --> 00:04:54,930 S1: applications built completely By eye with no thought of security whatsoever. 80 00:04:55,690 --> 00:04:59,610 S1: And the reason I a little bit of like opening 81 00:04:59,610 --> 00:05:03,610 S1: the kimono here. The reason I noticed this, or just 82 00:05:03,610 --> 00:05:07,450 S1: wanted to think about it, is because I'm doing it. 83 00:05:07,529 --> 00:05:11,729 S1: I see myself being sloppy. Now I'm going and cleaning up. 84 00:05:11,970 --> 00:05:15,890 S1: In most cases. But but there are certain cases where like, 85 00:05:15,930 --> 00:05:19,210 S1: it's not really publicly exposed. I'm not too worried about it. 86 00:05:19,330 --> 00:05:23,090 S1: Like it's only running local. I am moving so fast 87 00:05:23,810 --> 00:05:27,529 S1: that I can see myself being sloppy. There's another thing 88 00:05:27,610 --> 00:05:33,370 S1: that is really seductive about AI coding that you really 89 00:05:33,370 --> 00:05:36,290 S1: have to watch out for, which is the better it gets, 90 00:05:36,290 --> 00:05:39,969 S1: the more it encourages you to just kind of trust 91 00:05:39,970 --> 00:05:42,849 S1: it and just kind of go with it. And this 92 00:05:42,850 --> 00:05:47,210 S1: is bad because it's bad for a number of reasons. One, 93 00:05:47,210 --> 00:05:51,150 S1: it's bad for the security reason because what you don't understand. 94 00:05:51,510 --> 00:05:56,710 S1: You can't secure. Okay. But another reason is it just 95 00:05:56,710 --> 00:06:00,790 S1: stops you from actually getting under the covers and actually 96 00:06:00,950 --> 00:06:04,349 S1: doing the mental work of figuring out how the components 97 00:06:04,350 --> 00:06:07,430 S1: should be laid out. So the better the A's get, 98 00:06:07,550 --> 00:06:10,429 S1: what you'll do is you'll just drop a a stupid 99 00:06:10,430 --> 00:06:16,790 S1: paragraph into your AI agent in a coding tool like Klein, 100 00:06:16,790 --> 00:06:20,830 S1: or like cursor, or like windsurf or something. This is 101 00:06:20,830 --> 00:06:24,030 S1: a starting point that so many people are doing right now. 102 00:06:24,550 --> 00:06:26,870 S1: It's not my recommended way to start, and it's not 103 00:06:26,870 --> 00:06:29,550 S1: the way I start when I really want to build 104 00:06:29,550 --> 00:06:31,990 S1: a good application. If I want to build a really 105 00:06:31,990 --> 00:06:35,789 S1: good application, I actually start with a structured prompt, which 106 00:06:35,790 --> 00:06:39,870 S1: is a little bit like a PRD document, and you 107 00:06:39,870 --> 00:06:42,230 S1: have that in the directory, and then you tell the 108 00:06:42,230 --> 00:06:44,750 S1: agent to go look at that thing, or you put 109 00:06:44,750 --> 00:06:47,029 S1: it in the place where it automatically parses it, like 110 00:06:47,029 --> 00:06:49,360 S1: in the cursor rules file, for file, for example, and 111 00:06:49,360 --> 00:06:52,360 S1: that's the way you do it properly. But what most 112 00:06:52,360 --> 00:06:55,640 S1: people are doing, I would guess, because I even find 113 00:06:55,640 --> 00:07:01,320 S1: myself doing this as well, is they're just babbling, typing, 114 00:07:01,680 --> 00:07:04,240 S1: or even dictating, oh, I want an app that does 115 00:07:04,240 --> 00:07:06,080 S1: this and oh, it's going to be like this. Oh, 116 00:07:06,080 --> 00:07:08,440 S1: and make sure it looks cool. And oh, and by 117 00:07:08,440 --> 00:07:10,480 S1: the way, like it should have this thing, but I 118 00:07:10,480 --> 00:07:12,400 S1: want the button on the left instead of the right. 119 00:07:12,400 --> 00:07:16,760 S1: It's just this long paragraph of babble garbage. And then 120 00:07:16,760 --> 00:07:20,040 S1: they press enter and then the thing starts making stuff. 121 00:07:20,480 --> 00:07:24,840 S1: Now it could be making the most horrendously insecure things. 122 00:07:25,560 --> 00:07:29,000 S1: It could be making things that are not really scrutable 123 00:07:29,000 --> 00:07:33,880 S1: or understandable, like they split the logic across multiple pieces. 124 00:07:33,880 --> 00:07:36,920 S1: So it's not like it's not readable, it's not understandable. 125 00:07:37,840 --> 00:07:41,560 S1: And there's a tendency for people because they want to 126 00:07:41,560 --> 00:07:44,760 S1: move fast. And again, when I say people, I mean me. 127 00:07:44,920 --> 00:07:47,780 S1: And I think the situation is what? Much worse than 128 00:07:47,780 --> 00:07:51,380 S1: me because I'm actually guarding against this actively. So I 129 00:07:51,380 --> 00:07:54,620 S1: think it's way worse for just the general public. And 130 00:07:54,620 --> 00:07:58,020 S1: it's already bad for me. But it's like what you 131 00:07:58,020 --> 00:08:00,220 S1: end up with is a thing that kind of works. 132 00:08:00,340 --> 00:08:03,700 S1: Which is why I use this image here, um, which 133 00:08:03,700 --> 00:08:07,740 S1: my team actually came up with. Angela came up with this, and, uh, 134 00:08:07,740 --> 00:08:11,220 S1: it's like you end up with this monstrosity that can, like, 135 00:08:11,260 --> 00:08:16,740 S1: fall over so easily. It's just total, absolute garbage. And 136 00:08:16,740 --> 00:08:19,460 S1: then you're like, uh, no, not like that. No, no. 137 00:08:19,460 --> 00:08:23,100 S1: Change that one thing. And then after like 20, 30, 50, 138 00:08:23,140 --> 00:08:26,060 S1: 100 iterations of this thing over the course of like 139 00:08:26,060 --> 00:08:28,660 S1: an hour and a half, like if you're doing something 140 00:08:28,660 --> 00:08:30,460 S1: else on the side, but you come back and check 141 00:08:30,460 --> 00:08:34,420 S1: on its progress. Okay, maybe you made something, but one 142 00:08:34,420 --> 00:08:37,260 S1: can you edit it? Can you? Do you understand it? 143 00:08:37,700 --> 00:08:40,179 S1: Do you understand all the technologies that you use to 144 00:08:40,179 --> 00:08:45,120 S1: build that thing? Probably not. Did it use proper Technique 145 00:08:45,360 --> 00:08:49,800 S1: and best practices to build the stuff? Probably not. It depends. There's. 146 00:08:49,840 --> 00:08:54,800 S1: Wide variation there, but probably not. And finally, like okay, 147 00:08:54,840 --> 00:08:57,880 S1: for all the configs, did you use cloud services? Okay cool. 148 00:08:57,880 --> 00:09:01,640 S1: Where are the API keys? Are the configs readable? Like 149 00:09:02,280 --> 00:09:05,600 S1: oh did you use cloud storage? Are those buckets locked down? 150 00:09:06,000 --> 00:09:09,120 S1: All these things are like the faster we get at 151 00:09:09,120 --> 00:09:12,040 S1: using these AI tools, and the more people get in 152 00:09:12,080 --> 00:09:16,400 S1: who aren't actual coders who never coded anything beforehand, this 153 00:09:16,400 --> 00:09:20,960 S1: problem is 100 times worse for that. So bottom line is, 154 00:09:20,960 --> 00:09:22,760 S1: there is so much good that's going to come out 155 00:09:22,760 --> 00:09:26,800 S1: of AI. And I would say it's still more good 156 00:09:27,000 --> 00:09:29,760 S1: is going to come out of AI. Because I would 157 00:09:29,760 --> 00:09:33,800 S1: say creation is more important than security. That would be 158 00:09:33,800 --> 00:09:37,400 S1: my my thing. And which might be strange to hear 159 00:09:37,400 --> 00:09:40,600 S1: from a security person, but I honestly believe that's true. 160 00:09:41,080 --> 00:09:45,620 S1: I would, I want secure creation. That's what I want. But, um, 161 00:09:45,620 --> 00:09:49,380 S1: if I had to choose. Everything is secure and nothing happens. 162 00:09:49,500 --> 00:09:53,100 S1: I would rather have insecure creation. So I think that's 163 00:09:53,100 --> 00:09:56,900 S1: more important for humans. But bottom line is there are 164 00:09:57,020 --> 00:10:03,380 S1: externalities to. These AI tools, severe externalities to these AI tools. 165 00:10:04,020 --> 00:10:09,179 S1: The better they get, the more they coax you into. 166 00:10:09,220 --> 00:10:13,300 S1: Not looking under the covers. And it's nasty under the 167 00:10:13,300 --> 00:10:16,140 S1: covers with a lot of these AI tools. Crater of 168 00:10:16,140 --> 00:10:19,980 S1: curl just announced they're completely abandoning Cvss scoring because it's 169 00:10:19,980 --> 00:10:26,540 S1: fundamentally broken for wide, widely used open source projects. And 170 00:10:26,540 --> 00:10:31,059 S1: this is specifically Daniel Stenberg explaining how Cisa recently marked 171 00:10:31,660 --> 00:10:36,660 S1: a low severity vulnerability as a critical with A91. And 172 00:10:36,660 --> 00:10:38,940 S1: he's basically saying it should not have been A91. It 173 00:10:38,940 --> 00:10:41,910 S1: should have been a low. So all this talk has 174 00:10:41,910 --> 00:10:44,630 S1: really been flying around for a long time, and they 175 00:10:44,670 --> 00:10:48,390 S1: kind of made some updates with the latest version of CSS. 176 00:10:48,830 --> 00:10:52,550 S1: But I think ultimately the problem is that these static 177 00:10:52,550 --> 00:10:55,350 S1: things are not made to be looked at constantly. They're 178 00:10:55,350 --> 00:10:58,030 S1: not made to look at the actual risk of your 179 00:10:58,030 --> 00:11:01,390 S1: org and your codebase and all the context around that, 180 00:11:01,550 --> 00:11:04,150 S1: combined with the risk of the vulnerability. And then to 181 00:11:04,190 --> 00:11:07,750 S1: compare those two. And this is probably going to surprise you, 182 00:11:07,750 --> 00:11:10,910 S1: but I think I might be a good application here. 183 00:11:11,110 --> 00:11:13,589 S1: You've never heard that from me that I might be useful. 184 00:11:14,110 --> 00:11:19,910 S1: But the idea is that what you what CVS, CVS 185 00:11:19,950 --> 00:11:22,829 S1: and other tools like this are bad at is they 186 00:11:22,870 --> 00:11:27,670 S1: are they're static. Right. You put some dumb information there 187 00:11:27,670 --> 00:11:30,829 S1: about like, oh how exposed am I or whatever. That's 188 00:11:30,830 --> 00:11:33,790 S1: not really the way to find out how exposed you are. 189 00:11:34,070 --> 00:11:36,710 S1: The real way to do that is to pull in 190 00:11:36,710 --> 00:11:40,650 S1: real time context. So the future of all of this 191 00:11:40,650 --> 00:11:46,050 S1: is like super obvious. It's basically context plus intelligence of 192 00:11:46,050 --> 00:11:50,810 S1: combining a vulnerability with your actual like risk posture. And 193 00:11:50,809 --> 00:11:55,770 S1: that's what's going to determine like what your current risk is. 194 00:11:56,050 --> 00:12:00,370 S1: That's that's just the reality of what's coming. And so 195 00:12:00,370 --> 00:12:03,250 S1: it could be that, you know, there's nothing particularly wrong 196 00:12:03,250 --> 00:12:07,610 S1: with Cvss. It's it's always been like this. The previous 197 00:12:07,610 --> 00:12:11,809 S1: versions were, you know, worse. And then there's other systems 198 00:12:11,809 --> 00:12:14,370 S1: and they also have their problems. But ultimately it comes 199 00:12:14,370 --> 00:12:17,010 S1: down to this one thing, which is what is the 200 00:12:17,010 --> 00:12:19,969 S1: state of the world? What exactly is the vuln and 201 00:12:19,970 --> 00:12:22,250 S1: how does it relate to me? And if you want 202 00:12:22,250 --> 00:12:26,690 S1: to do that properly in the old days, that's that's 203 00:12:26,690 --> 00:12:32,010 S1: like sending security engineers to go do research, to study to, 204 00:12:32,050 --> 00:12:34,010 S1: you know, look at a whole bunch of docs, figure 205 00:12:34,010 --> 00:12:36,330 S1: out where we have this thing installed. We're talking about 206 00:12:36,330 --> 00:12:40,469 S1: asset management. We're talking about nightmares. I mean, you can 207 00:12:40,470 --> 00:12:43,190 S1: spend weeks looking at one phone and seeing how vulnerable 208 00:12:43,190 --> 00:12:45,270 S1: you are to it. If you want to have a 209 00:12:45,270 --> 00:12:48,270 S1: good assessment, right. So the whole point of this is 210 00:12:48,270 --> 00:12:49,790 S1: that you're going to be able to do that much 211 00:12:49,790 --> 00:12:53,110 S1: faster now. And I think that's ultimately what replaces systems 212 00:12:53,110 --> 00:12:58,429 S1: like Cvss and the change. Healthcare ransomware attack is now 213 00:12:58,429 --> 00:13:04,510 S1: officially largest healthcare breach in the US 190 people affected. 214 00:13:04,510 --> 00:13:08,069 S1: Now Swedish authorities just grabbed a ship they think could 215 00:13:08,070 --> 00:13:13,429 S1: an underwater internet cable between Sweden and Latvia. This is 216 00:13:13,429 --> 00:13:17,510 S1: after multiple similar incidents and a lot of them are 217 00:13:17,550 --> 00:13:21,230 S1: tying back to Russia. All right, the big story deep 218 00:13:22,030 --> 00:13:26,470 S1: in AI and tech. So Nvidia loses $600 billion after 219 00:13:26,470 --> 00:13:30,950 S1: deep sea AI breakthrough. And the stock market overall on 220 00:13:30,950 --> 00:13:35,790 S1: Monday lost $1 trillion. So it was the single biggest 221 00:13:35,790 --> 00:13:39,010 S1: day of a market loss in history, and it beat 222 00:13:39,010 --> 00:13:44,410 S1: the record previously from also from Nvidia. But um, basically 223 00:13:44,450 --> 00:13:46,890 S1: here's what happened. This is what it came down to like. 224 00:13:47,050 --> 00:13:51,730 S1: Nvidia had been a darling of AI hype because they're 225 00:13:51,730 --> 00:13:54,770 S1: the GPU leaders. Much of the future hope of making 226 00:13:54,770 --> 00:13:59,650 S1: money from AI has been embodied by Nvidia. The idea 227 00:13:59,650 --> 00:14:03,089 S1: is that GPUs rule the AI world and Nvidia rules 228 00:14:03,090 --> 00:14:07,850 S1: the GPU world. And implicit in that assumption is that 229 00:14:07,890 --> 00:14:12,770 S1: Nvidia chips are scarce and necessary and expensive. That means 230 00:14:12,770 --> 00:14:14,530 S1: anyone who wants to be a leader will have to 231 00:14:14,530 --> 00:14:18,250 S1: have lots of Nvidia chips. So deep tech just blew 232 00:14:18,250 --> 00:14:20,330 S1: that out of the water because they produced something that 233 00:14:20,330 --> 00:14:23,730 S1: should have cost them like tens of billions, but they 234 00:14:23,730 --> 00:14:28,130 S1: did it for $5.6 million. They basically found workarounds that 235 00:14:28,130 --> 00:14:32,730 S1: allowed them to get a lot more performance for less resources. 236 00:14:33,610 --> 00:14:37,220 S1: And it freaked everyone out. Mostly investors, but it like 237 00:14:37,260 --> 00:14:40,540 S1: it messed with the entire stock market because the stock 238 00:14:40,540 --> 00:14:45,660 S1: market is largely pushed by tech. So basically less necessary 239 00:14:45,700 --> 00:14:52,020 S1: equals less valuable. My analysis is so what. Right. If anything, 240 00:14:52,020 --> 00:14:56,180 S1: deep seek is nothing but exciting because we're getting more 241 00:14:56,300 --> 00:15:01,260 S1: AI for less resources and we need way more AI. 242 00:15:01,300 --> 00:15:04,980 S1: And I've got a whole video which I'll link up here, 243 00:15:04,980 --> 00:15:07,740 S1: but it's like, how much do we need? Way, way 244 00:15:07,740 --> 00:15:10,420 S1: more than we have. And I keep doing this thing 245 00:15:10,420 --> 00:15:13,980 S1: where I'm like, oh, we only have 0.000 whatever, ten 246 00:15:13,980 --> 00:15:17,940 S1: zeros and A19. We're only at that percentage of how 247 00:15:17,940 --> 00:15:20,820 S1: much AI that we need. But I don't know that percentage. 248 00:15:20,820 --> 00:15:22,660 S1: I'm making that up. I'm trying to show lots of 249 00:15:22,660 --> 00:15:25,780 S1: zeros to prove it's like a tiny fraction of 1%. 250 00:15:25,860 --> 00:15:29,460 S1: It's a made up number. It's everyone should obviously know that. 251 00:15:29,460 --> 00:15:32,720 S1: But the point is we're just getting started. And in 252 00:15:32,760 --> 00:15:35,320 S1: that video, I lay this all out. I lay out 253 00:15:35,360 --> 00:15:39,360 S1: what types of things we as a society are going 254 00:15:39,360 --> 00:15:43,600 S1: to want to do with AI, and why that requires 255 00:15:43,600 --> 00:15:48,080 S1: so much processing, so much inference, so much intelligence to 256 00:15:48,120 --> 00:15:51,320 S1: be able to do that for basically every human problem. 257 00:15:51,800 --> 00:15:54,320 S1: And when I say human problem, I mean that means 258 00:15:54,320 --> 00:16:00,560 S1: also business problems and stuff like that. Personal problems, everything. So, uh, 259 00:16:00,560 --> 00:16:04,760 S1: I think we're very confused about this whole thing. Um, 260 00:16:04,760 --> 00:16:07,080 S1: and the advantage that Deep Seek found is an example 261 00:16:07,080 --> 00:16:09,400 S1: of what I've been calling slack in the rope. And 262 00:16:09,400 --> 00:16:12,400 S1: I've been talking about this with my friend Jai Patel 263 00:16:13,120 --> 00:16:16,720 S1: for a very long time, and he's, like, one of 264 00:16:16,720 --> 00:16:19,760 S1: the smartest AI people that I know. And I basically 265 00:16:19,760 --> 00:16:22,960 S1: said back in like 23, I'm like, look there. There 266 00:16:22,960 --> 00:16:25,680 S1: is so much slack here that's going to allow people 267 00:16:25,680 --> 00:16:29,800 S1: to jump way ahead, and it's going to jump them 268 00:16:29,840 --> 00:16:32,860 S1: ahead of the of the line, even above people who've 269 00:16:32,860 --> 00:16:36,620 S1: been grinding. So, for example, you spend billions of dollars, 270 00:16:36,660 --> 00:16:40,420 S1: let's say we're at parity between like two competitors and 271 00:16:40,420 --> 00:16:43,340 S1: somebody spends a billions of dollars and they jump up 272 00:16:43,340 --> 00:16:47,580 S1: by 20%, but somebody figures out, hey, wait a minute. 273 00:16:47,580 --> 00:16:50,620 S1: Like if you just double the size of the alphabet 274 00:16:51,020 --> 00:16:54,580 S1: and you add some special characters to the mix or whatever, 275 00:16:54,580 --> 00:16:58,780 S1: it actually jumps up by 600%. Isn't that weird? And 276 00:16:58,780 --> 00:17:03,940 S1: then everyone starts doing that, and then everyone collectively jumps 600%. Well, 277 00:17:03,940 --> 00:17:07,379 S1: what happened to the last six months where competitor B 278 00:17:07,619 --> 00:17:12,700 S1: just spent $4 billion trying to go up by 20%? 279 00:17:12,859 --> 00:17:17,780 S1: My argument to GI was basically, there are so many 280 00:17:17,780 --> 00:17:21,419 S1: of those big jumps completely hidden from us because we 281 00:17:21,420 --> 00:17:25,300 S1: have no idea how any of this stuff works. So 282 00:17:25,300 --> 00:17:28,620 S1: this China team just found one of these things they 283 00:17:28,619 --> 00:17:32,080 S1: just found through a number of things with like. Mixture 284 00:17:32,080 --> 00:17:35,400 S1: of experts. And they've got papers detailing everything they did. 285 00:17:36,080 --> 00:17:38,679 S1: But it's like, hey, look, we just did this. And 286 00:17:38,680 --> 00:17:41,880 S1: it's like, now, now it's obvious to everyone, everyone's going 287 00:17:41,880 --> 00:17:45,240 S1: to go copy that. That's another big jump. I think 288 00:17:45,640 --> 00:17:50,719 S1: there is hundreds of times the amount of performance, like 289 00:17:50,760 --> 00:17:54,639 S1: laying on the floor, laying, sitting on the table ready 290 00:17:54,640 --> 00:17:58,000 S1: for the taking. They just they have to be discovered. 291 00:17:58,720 --> 00:18:03,280 S1: And I think as the technology changes, there's a new algorithm, 292 00:18:03,280 --> 00:18:07,879 S1: there's new transformers that we use, maybe a completely different 293 00:18:07,880 --> 00:18:11,560 S1: type of AI. Every time we go into this like unknown. 294 00:18:12,400 --> 00:18:15,040 S1: It's a new set of things, a new set of 295 00:18:15,040 --> 00:18:17,639 S1: tricks or slack in the rope just sitting around to 296 00:18:17,640 --> 00:18:21,960 S1: be discovered. And my point is that those pieces of slack, 297 00:18:22,400 --> 00:18:26,600 S1: those tricks that allow you to jump forward, and I 298 00:18:26,600 --> 00:18:29,220 S1: don't want to want to demean with the word trick. 299 00:18:29,220 --> 00:18:32,300 S1: Someone mentioned that. Like when I say trick, I just 300 00:18:32,300 --> 00:18:36,700 S1: mean in retrospect it'll appear obvious. It doesn't mean it 301 00:18:36,700 --> 00:18:39,939 S1: doesn't take extreme genius and hard work and discipline to 302 00:18:39,980 --> 00:18:44,139 S1: actually discover it. So I just want to make that clear. 303 00:18:44,220 --> 00:18:47,220 S1: But the point is, we're going to be able to 304 00:18:47,220 --> 00:18:52,020 S1: just keep jumping massively. And I think a very large 305 00:18:52,020 --> 00:18:54,820 S1: percent of those jumps are going to come from these 306 00:18:54,820 --> 00:18:59,060 S1: sorts of tricks or slack in the rope type effects. 307 00:18:59,859 --> 00:19:02,900 S1: And that's what I think we had with, uh, with 308 00:19:02,900 --> 00:19:06,780 S1: deep Post-training is perhaps the most powerful category of these tricks. 309 00:19:06,780 --> 00:19:09,100 S1: This is what I said back in August of last year. 310 00:19:09,100 --> 00:19:13,139 S1: The most powerful category of these tricks is like teaching 311 00:19:13,140 --> 00:19:16,300 S1: a giant alien brain how to be smart when it 312 00:19:16,300 --> 00:19:19,739 S1: had tremendous potential before, but no direction. And again, I'm 313 00:19:19,740 --> 00:19:22,500 S1: saying there's lots of these different ones, but kind of 314 00:19:22,540 --> 00:19:25,669 S1: playing off this thing from 2024, I'm going to say 315 00:19:25,670 --> 00:19:27,629 S1: this in a different way, and this is kind of 316 00:19:27,670 --> 00:19:31,070 S1: a casual way of me thinking about this, which, uh, 317 00:19:31,070 --> 00:19:33,469 S1: I feel like GI doesn't really like when I do 318 00:19:33,470 --> 00:19:37,630 S1: this because it's not perfectly accurate. But one way I'm 319 00:19:37,630 --> 00:19:40,550 S1: casually thinking about this is that there are now two 320 00:19:40,550 --> 00:19:43,790 S1: steps here the intelligence, which is the model, and then 321 00:19:43,790 --> 00:19:47,070 S1: the wisdom, which is the reinforcement learning. So this is 322 00:19:47,070 --> 00:19:50,150 S1: kind of where a lot of this slack has come from, 323 00:19:50,150 --> 00:19:52,750 S1: is the fact that the model itself is not super 324 00:19:52,750 --> 00:19:56,870 S1: intelligence by itself. You actually have to tell it what 325 00:19:56,869 --> 00:20:03,190 S1: constitutes intelligence or wisdom or smarts or usefulness. And that's 326 00:20:03,190 --> 00:20:07,550 S1: after the fact with the. And more generally, the reinforcement learning. 327 00:20:07,670 --> 00:20:11,030 S1: So it's almost like intelligence is the size of the brain. 328 00:20:11,030 --> 00:20:14,430 S1: And reinforcement learning is the life experience. And like I said, 329 00:20:14,430 --> 00:20:17,510 S1: it's not technically true. But I think it's it's pretty powerful. 330 00:20:18,150 --> 00:20:21,150 S1: So bottom line is I think the the market reaction 331 00:20:21,150 --> 00:20:24,290 S1: to all of this is very wrong. Well, I've got 332 00:20:24,290 --> 00:20:28,570 S1: another piece in here about the total tam of I. 333 00:20:28,570 --> 00:20:30,170 S1: In fact, I'm just going to jump to that right 334 00:20:30,170 --> 00:20:32,649 S1: now because I don't think I mentioned it anywhere else. 335 00:20:32,810 --> 00:20:36,850 S1: Let's see here. Oh, this is really cool from Andre. 336 00:20:36,890 --> 00:20:39,930 S1: Move 37 is the word of the day. It's when 337 00:20:39,930 --> 00:20:43,290 S1: an AI trained via the trial and error process of 338 00:20:43,290 --> 00:20:48,130 S1: reinforcement learning discovers actions that are new, surprising and secretly brilliant. 339 00:20:48,609 --> 00:20:52,770 S1: So this is in reference to Lee Soto losing in 340 00:20:52,770 --> 00:20:57,010 S1: go to an AI and it was, uh, it was 341 00:20:57,010 --> 00:20:59,609 S1: moved 37 that made everyone freak out. So I thought 342 00:20:59,609 --> 00:21:02,649 S1: that was a cool concept, but, um. Yeah, if I 343 00:21:02,650 --> 00:21:06,530 S1: go here, here we go. The total Tam for AI 344 00:21:06,570 --> 00:21:10,130 S1: is a combination of two primary components. The total cost 345 00:21:10,130 --> 00:21:13,690 S1: of human workforces, and the amount of money that current 346 00:21:13,690 --> 00:21:17,210 S1: and future companies will pay to start ten x or 347 00:21:17,210 --> 00:21:21,210 S1: 1000 X their business. So we're talking about hundreds of 348 00:21:21,210 --> 00:21:24,630 S1: trillions of dollars. So if you go back to this, 349 00:21:25,350 --> 00:21:30,230 S1: this is why I think, uh, the whole Nvidia deep 350 00:21:30,270 --> 00:21:34,030 S1: sea thing is very mistaken. The Tam and I did 351 00:21:34,030 --> 00:21:38,830 S1: a deep sea analysis, actually, of this very tentative, very non-scientific, 352 00:21:38,830 --> 00:21:44,230 S1: but it came up with like 80 to $120 trillion 353 00:21:44,230 --> 00:21:48,270 S1: as being the Tam for AI over the next ten years. 354 00:21:48,510 --> 00:21:51,750 S1: So I don't know. Again, it's better if you do 355 00:21:51,790 --> 00:21:55,350 S1: like one of the, uh, the new Google Deep research 356 00:21:55,350 --> 00:21:58,350 S1: projects to actually feed it that thing and then give 357 00:21:58,350 --> 00:22:01,790 S1: that to deep Seac or O1 or O3 or whenever 358 00:22:01,790 --> 00:22:04,669 S1: it comes out. Um, that would be a better way 359 00:22:04,670 --> 00:22:07,510 S1: to get those numbers and have it fully explained the numbers. 360 00:22:08,030 --> 00:22:13,030 S1: But either way, it's trillions upon trillions of dollars. So 361 00:22:13,230 --> 00:22:16,390 S1: that's why I think this whole thing is overblown and mistaken. 362 00:22:16,390 --> 00:22:18,990 S1: And that's why my attitude is very much so. What, 363 00:22:19,850 --> 00:22:23,290 S1: Because the market has gone from being foolish to overvalue 364 00:22:23,490 --> 00:22:28,290 S1: Nvidia to being foolish to undervalue it. Right. It's because 365 00:22:28,450 --> 00:22:32,890 S1: there is just so much to go, so much room 366 00:22:32,890 --> 00:22:35,850 S1: before we get to anything like a ceiling. And here's 367 00:22:35,850 --> 00:22:38,570 S1: where I talk about that. Right. So what happens to 368 00:22:38,570 --> 00:22:41,570 S1: Nvidia or any other part of the stack doesn't matter 369 00:22:41,570 --> 00:22:44,850 S1: much at all, because we're still at whatever percentage of 370 00:22:44,850 --> 00:22:47,370 S1: the amount of AI that we need in the world. 371 00:22:47,410 --> 00:22:49,650 S1: Doesn't matter how we get there, it's not predictable. Could 372 00:22:49,650 --> 00:22:53,570 S1: be ARM processors, GPUs, some brand new thing. Doesn't matter. 373 00:22:53,690 --> 00:22:58,010 S1: I think Nvidia and TSMC and all these people are 374 00:22:58,010 --> 00:23:01,810 S1: going to continue to rise because again, we're at the 375 00:23:01,850 --> 00:23:05,690 S1: tiniest little sliver. The question is who is going to 376 00:23:05,690 --> 00:23:08,689 S1: move into the next space. Can Nvidia pivot? I believe 377 00:23:08,690 --> 00:23:10,890 S1: they can. If they had to I'm not sure they 378 00:23:10,890 --> 00:23:14,409 S1: have to. First of all, Deep Seek was built on Nvidia. 379 00:23:14,770 --> 00:23:17,050 S1: If they had better Nvidia deep seek would be better. 380 00:23:17,090 --> 00:23:20,940 S1: Like it's like more Nvidia still makes everything better, right? 381 00:23:21,220 --> 00:23:24,500 S1: And I'm not trying to just go pro Nvidia because 382 00:23:24,500 --> 00:23:29,420 S1: I'm invested. I actually got a decent amount out of Nvidia. 383 00:23:29,540 --> 00:23:33,180 S1: I want to say, um, about two weeks before this, 384 00:23:33,180 --> 00:23:37,900 S1: because I made the choice to move more money into 385 00:23:37,940 --> 00:23:41,939 S1: places that would benefit from AI, and not necessarily just 386 00:23:41,940 --> 00:23:45,980 S1: the roads and bridges. Happened to get lucky there because 387 00:23:45,980 --> 00:23:49,859 S1: I pulled some money out before it went down. But, um, 388 00:23:49,859 --> 00:23:53,460 S1: I very much I did buy the dip a little bit, 389 00:23:53,700 --> 00:23:55,980 S1: and I very much would have liked to buy even 390 00:23:55,980 --> 00:23:58,420 S1: more at the dip of Nvidia, because I think it's 391 00:23:58,420 --> 00:24:02,460 S1: still massively going up along with everything else. It's like, 392 00:24:02,500 --> 00:24:07,180 S1: I don't think they're particularly perfectly, uh, set to be 393 00:24:07,180 --> 00:24:10,340 S1: the ones. Everyone's going to be the one if you're 394 00:24:10,340 --> 00:24:12,340 S1: smart about it and you have a good leader, which 395 00:24:12,380 --> 00:24:15,939 S1: Jensen is. All right. Enough about that. So OpenAI launched 396 00:24:15,940 --> 00:24:19,480 S1: a preview of operator which can navigate web browsers just 397 00:24:19,480 --> 00:24:22,440 S1: like a human. I think we need more generalized agents. 398 00:24:22,520 --> 00:24:25,919 S1: I didn't like the whole App Store vibe to it. 399 00:24:25,960 --> 00:24:29,040 S1: Google just dropped a massive update to Gemini. This is 400 00:24:29,040 --> 00:24:31,879 S1: like one of the sleepers. Honestly, Gemini is one of 401 00:24:31,880 --> 00:24:36,600 S1: the sleepers of this whole competition. Google has been just 402 00:24:36,600 --> 00:24:39,880 S1: doing insanely good for the last six months 12 months. 403 00:24:40,440 --> 00:24:44,719 S1: I use the the flash 2.0 Pro. I think it 404 00:24:44,720 --> 00:24:48,439 S1: is for a lot of analysis where I need really 405 00:24:48,440 --> 00:24:52,560 S1: good haystack performance over like 2 million tokens. So when 406 00:24:52,560 --> 00:24:56,920 S1: I do rag searches across like all my content, for example, 407 00:24:56,920 --> 00:24:59,760 S1: and I get back like 50 documents that hit on 408 00:24:59,760 --> 00:25:05,159 S1: the rag and this is over like 3000 posts, 3200 409 00:25:05,160 --> 00:25:10,080 S1: posts since going like back to 1999. So if I 410 00:25:10,080 --> 00:25:12,280 S1: have a big giant piece of content like that and 411 00:25:12,280 --> 00:25:15,500 S1: I don't want to miss like the The nuggets or 412 00:25:15,500 --> 00:25:18,460 S1: the or the, you know, the needles inside of this 413 00:25:18,460 --> 00:25:21,179 S1: massive haystack. I send it over to Google and it 414 00:25:21,180 --> 00:25:24,940 S1: does a better job of of finding the haystack. Plus, 415 00:25:24,980 --> 00:25:27,740 S1: I mean, just brute force. It has 2 million tokens 416 00:25:27,740 --> 00:25:32,140 S1: as opposed to 200,000 for like sonic anthropic build citations 417 00:25:33,140 --> 00:25:37,740 S1: API to combat hallucinations. That's really cool. Uh, Google just 418 00:25:37,740 --> 00:25:41,820 S1: dropped another billion dollars into anthropic. Very interesting. So they're 419 00:25:41,820 --> 00:25:45,340 S1: not only doing their own play. They're also betting against 420 00:25:45,380 --> 00:25:49,340 S1: OpenAI with anthropic. I don't know if they're in OpenAI 421 00:25:49,380 --> 00:25:52,620 S1: as well. But anyway, they're investing in anthropic, which I 422 00:25:52,619 --> 00:25:56,780 S1: thought was really interesting. Leaked memo from Apple says they're 423 00:25:56,780 --> 00:26:03,260 S1: focusing completely on rebuilding series infrastructure and improving their existing 424 00:26:03,380 --> 00:26:07,100 S1: AI models this year, which, yeah, it's the AI chief 425 00:26:07,140 --> 00:26:10,139 S1: talking about it. So obviously they're talking about AI. Well, 426 00:26:10,180 --> 00:26:14,879 S1: overall startup funding has dropped significantly since 2021, seed rounds 427 00:26:14,880 --> 00:26:19,639 S1: are actually getting bigger. This is interesting. My friend, uh, Mike, um, 428 00:26:19,720 --> 00:26:23,240 S1: has a really, really cool podcast on this, uh, which 429 00:26:23,240 --> 00:26:25,840 S1: you could check in the show notes. His name is 430 00:26:25,840 --> 00:26:30,400 S1: Mike Privett, and you should absolutely go check out his stuff. Uh, 431 00:26:30,400 --> 00:26:34,320 S1: Colorado police are now giving away free AirTags and the 432 00:26:34,320 --> 00:26:37,520 S1: tile trackers to help prevent vehicle theft in their community. 433 00:26:37,720 --> 00:26:42,200 S1: A ring camera in Canada. This is insane. Caught the 434 00:26:42,200 --> 00:26:46,520 S1: exact moment of a meteorite smashing into a basically a 435 00:26:46,520 --> 00:26:48,960 S1: walkway in front of their house. So they just came 436 00:26:48,960 --> 00:26:52,600 S1: out and just like passed that that location and you 437 00:26:52,640 --> 00:26:57,200 S1: hear like this really loud crash and it hit tiles 438 00:26:57,560 --> 00:27:01,000 S1: on the walkway. And I, I don't understand how it didn't, like, 439 00:27:01,000 --> 00:27:03,720 S1: blow out the tile and create a little hole in 440 00:27:03,720 --> 00:27:05,840 S1: the ground. But anyway, they picked it up. It was 441 00:27:05,840 --> 00:27:09,320 S1: an actual meteor and it just like blew up against 442 00:27:09,320 --> 00:27:11,889 S1: the tile. I don't know, That had to be going 443 00:27:11,930 --> 00:27:15,370 S1: thousands of miles per hour, right? It's terminal velocity, I guess. 444 00:27:15,410 --> 00:27:17,890 S1: Is it hundreds of miles an hour? I don't know, 445 00:27:17,890 --> 00:27:20,330 S1: but had to be going fast. It definitely would have 446 00:27:20,330 --> 00:27:22,970 S1: killed someone if it hit him, I think. Anyway, I 447 00:27:22,970 --> 00:27:25,730 S1: don't know why it didn't crack the, uh, the stone 448 00:27:25,730 --> 00:27:28,010 S1: on the ground. A new Harvard study shows we should 449 00:27:28,010 --> 00:27:31,250 S1: be taking blood pressure readings while lying down instead of sitting. 450 00:27:31,450 --> 00:27:34,170 S1: And Hans Zimmer is apparently in talks with Saudi Arabia 451 00:27:34,170 --> 00:27:38,570 S1: to remake their national anthem. And a pre-mortem is basically 452 00:27:38,570 --> 00:27:41,129 S1: where you imagine your product has already failed, and you 453 00:27:41,170 --> 00:27:44,050 S1: work backwards to figure out why, and you do this 454 00:27:44,050 --> 00:27:48,369 S1: before you even start the project. That's brilliant. Reminds me 455 00:27:48,369 --> 00:27:52,450 S1: of the PR system inside of Amazon, which I learned 456 00:27:52,450 --> 00:27:56,090 S1: a lot because we use that at Apple. And uh, yeah, 457 00:27:56,130 --> 00:27:59,810 S1: four components of top AI model ecosystems. It's in the 458 00:27:59,810 --> 00:28:04,850 S1: ideas section related heavily to this deep tech stuff. And 459 00:28:04,850 --> 00:28:09,290 S1: for discovery, Klein is my new go to for doing 460 00:28:09,350 --> 00:28:13,510 S1: AI stuff. It is an extension built into standard VS 461 00:28:13,510 --> 00:28:18,710 S1: code and actually like it more than using cursor, which 462 00:28:18,710 --> 00:28:24,270 S1: is like more integrated and everything. Klein is more conversational. 463 00:28:24,350 --> 00:28:29,030 S1: It's more following my thread. It's just seems more intelligent 464 00:28:29,030 --> 00:28:34,630 S1: and more like cohesive in the way that it answers things. Um, 465 00:28:34,670 --> 00:28:38,590 S1: and again, it's I'll show you the website. Yeah. Klein bot. 466 00:28:39,190 --> 00:28:43,310 S1: Thoughtful AI coder. I don't know why it's better, but honestly, 467 00:28:43,310 --> 00:28:45,790 S1: it feels better to me. I take this pain right here. 468 00:28:45,790 --> 00:28:48,270 S1: I actually move it over to the right. That's how 469 00:28:48,270 --> 00:28:50,870 S1: I have it going. Developer created a brilliant trap for 470 00:28:50,870 --> 00:28:55,950 S1: web scrapers by using specifically crafted CSS selectors. So anti 471 00:28:55,990 --> 00:29:00,710 S1: scraper trap I recommend you try out deep seek. One 472 00:29:00,710 --> 00:29:04,630 S1: thing about deep seek that is really, really cool is that, 473 00:29:04,630 --> 00:29:08,250 S1: uh you when it responds, it just starts talking out 474 00:29:08,250 --> 00:29:11,130 S1: loud and you just see it. It's very similar to 475 00:29:11,170 --> 00:29:13,610 S1: O1 in that way. You just see it. That's the 476 00:29:13,610 --> 00:29:16,770 S1: whole reason it's doing so well in the benchmarks. That's 477 00:29:16,770 --> 00:29:20,250 S1: the reason it's controversial because it's very much acting like O1. 478 00:29:20,770 --> 00:29:24,370 S1: But bottom line is you see it having a conversation 479 00:29:24,370 --> 00:29:27,490 S1: with itself and it goes back and forth and it's like, yeah, 480 00:29:27,490 --> 00:29:29,290 S1: but this is really cool. Yeah. But have you thought 481 00:29:29,290 --> 00:29:32,010 S1: of this? Actually that could be wrong. Let me think 482 00:29:32,010 --> 00:29:34,570 S1: about this. And it does that for a number of 483 00:29:34,570 --> 00:29:38,970 S1: paragraphs before it actually gives you an answer. I recommend 484 00:29:38,970 --> 00:29:41,730 S1: you try it out with Olama. It's really easy to install. 485 00:29:41,930 --> 00:29:47,530 S1: And magenta NVMe is one of the I first, uh, 486 00:29:47,530 --> 00:29:50,250 S1: vim tools that I'm trying to use. Ultimately, I'd like 487 00:29:50,250 --> 00:29:52,810 S1: to get out of VSCode and have all my AI stuff. 488 00:29:52,890 --> 00:29:58,489 S1: Something like Klein inside of vim, but it's hard to 489 00:29:58,490 --> 00:30:00,410 S1: do that in the terminal, which is why I'm like 490 00:30:00,410 --> 00:30:03,810 S1: stuck in VSCode and not happy about it. Ben Thompson 491 00:30:03,850 --> 00:30:08,310 S1: had a pretty cool deep seq FAQ, FAQ, and Lane 492 00:30:08,350 --> 00:30:10,390 S1: Changes dropped. A cool new tool that lets you do 493 00:30:10,430 --> 00:30:12,390 S1: deep web research. This is the thing I was talking 494 00:30:12,390 --> 00:30:17,510 S1: about before completely using a hosted LMS. I've not messed 495 00:30:17,510 --> 00:30:20,790 S1: with this one yet, but I cannot wait to. Um. 496 00:30:20,830 --> 00:30:24,030 S1: Somebody created a simple service that converts WordPress blogs to 497 00:30:24,070 --> 00:30:28,150 S1: Hugo static sites. Highly recommend. Anyone building a website goes 498 00:30:28,150 --> 00:30:32,790 S1: static these days and basically keep your markdown. Your markdown 499 00:30:32,790 --> 00:30:35,590 S1: is your website that is your content. You can make 500 00:30:35,590 --> 00:30:38,750 S1: a rag out of it. It is just text and 501 00:30:38,750 --> 00:30:42,150 S1: basically any place you want to bring your blog to 502 00:30:42,550 --> 00:30:45,310 S1: give them the markdown and have it turn it into 503 00:30:45,350 --> 00:30:48,550 S1: a nice website that you like. You like the aesthetic of. 504 00:30:48,550 --> 00:30:52,870 S1: So it's content aesthetic and don't mix the two and 505 00:30:52,870 --> 00:30:56,150 S1: don't have it where it's like the platform owns the 506 00:30:56,150 --> 00:31:00,150 S1: mixture because then you're screwed. Um, and I've had this 507 00:31:00,150 --> 00:31:03,150 S1: happen to me multiple times. I just had it happen again, 508 00:31:03,390 --> 00:31:07,320 S1: getting my Getting my content out of beehive, which I love. 509 00:31:07,360 --> 00:31:10,360 S1: Beehive for my newsletter, but it's not my favorite place 510 00:31:10,600 --> 00:31:14,720 S1: to run an actual website. Philips Hue bulbs getting motion sensing. 511 00:31:15,200 --> 00:31:17,000 S1: I really hope this is true and I can't wait 512 00:31:17,000 --> 00:31:20,120 S1: to get it and play with it. Be nice to 513 00:31:20,160 --> 00:31:24,200 S1: have all different hue things basically. Also be motion sensors. 514 00:31:24,840 --> 00:31:29,280 S1: How to say no as a product manager. This is great. 515 00:31:29,480 --> 00:31:32,280 S1: This is great. I think it just refreshes. Yeah it 516 00:31:32,280 --> 00:31:37,080 S1: just refreshes. Let's not do that. So I'm hitting refresh. 517 00:31:37,800 --> 00:31:40,959 S1: Can we revisit this after the next sprint refresh. We 518 00:31:40,960 --> 00:31:45,000 S1: need to ensure alignment before moving forward. Refresh. This could 519 00:31:45,000 --> 00:31:47,240 S1: be part of a larger initiative, but let's hold off 520 00:31:47,240 --> 00:31:51,320 S1: for now. This is really good. It's really good. Uh, 521 00:31:51,320 --> 00:31:55,160 S1: and recommendation of the week. Remember that AI is not 522 00:31:55,200 --> 00:31:59,120 S1: AI stocks. AI is not the survival of AI companies 523 00:31:59,120 --> 00:32:05,260 S1: that did marketing in 2023 and 2024 A's. Tam is 524 00:32:05,260 --> 00:32:08,740 S1: the replacement of human labor and the magnification of GDP 525 00:32:09,180 --> 00:32:12,020 S1: that can come from millions or billions of people becoming 526 00:32:12,020 --> 00:32:15,820 S1: a founder, builder or creator. That is the ball to 527 00:32:15,860 --> 00:32:20,060 S1: watch and everything else is noise. And the aphorism of 528 00:32:20,060 --> 00:32:23,660 S1: the week is to be completely cured of newspapers spend 529 00:32:23,660 --> 00:32:26,300 S1: a year reading the paper from the previous week to 530 00:32:26,340 --> 00:32:31,340 S1: be completely cured of newspapers. Spend a year reading the 531 00:32:31,340 --> 00:32:38,300 S1: newspaper from the previous week. Nassim Nicholas Taleb. Unsupervised learning 532 00:32:38,300 --> 00:32:42,220 S1: is produced on Hindenburg Pro using an SM seven B microphone. 533 00:32:43,060 --> 00:32:45,380 S1: A video version of the podcast is available on the 534 00:32:45,380 --> 00:32:49,020 S1: Unsupervised Learning YouTube channel, and the text version with full 535 00:32:49,020 --> 00:32:53,780 S1: links and notes is available at Daniel. Com slash newsletter. 536 00:32:54,420 --> 00:32:55,380 S1: We'll see you next time.