1 00:00:01,000 --> 00:00:05,080 S1: Unsupervised Learning is a podcast about trends and ideas in cybersecurity, 2 00:00:05,200 --> 00:00:10,040 S1: national security, AI, technology and society, and how best to 3 00:00:10,080 --> 00:00:17,480 S1: upgrade ourselves to be ready for what's coming. All right, 4 00:00:17,480 --> 00:00:20,120 S1: welcome to Unsupervised Learning. This is Daniel Miessler, and I'm 5 00:00:20,120 --> 00:00:24,200 S1: building AI to upgrade humans. Episode 466. Hope your week 6 00:00:24,200 --> 00:00:27,080 S1: is starting off better than Nvidia's did, which is probably 7 00:00:27,080 --> 00:00:30,400 S1: pretty easy to do. Went to a phenomenal offensive security 8 00:00:30,400 --> 00:00:35,760 S1: and AI conference sponsored by Rob Ragan. Not really sponsored 9 00:00:35,760 --> 00:00:38,280 S1: by him, but put on by him. Had lots of 10 00:00:38,320 --> 00:00:41,040 S1: cool sponsors for different things. But anyway, it was a 11 00:00:41,040 --> 00:00:43,919 S1: whole day of hacking. Basically, it was a hackathon fundamentally, 12 00:00:44,479 --> 00:00:48,479 S1: but it was also just a great networking event for anybody, 13 00:00:48,520 --> 00:00:50,960 S1: kind of at the intersection of AI and security in 14 00:00:50,960 --> 00:00:54,360 S1: San Francisco. And it was just just great. Rob did 15 00:00:54,360 --> 00:00:57,080 S1: a great job. Kind of a nerd observation. Far too 16 00:00:57,080 --> 00:01:01,040 S1: many people don't know how to use name drop. If 17 00:01:01,040 --> 00:01:03,760 S1: you take your iPhone and set it next to another 18 00:01:03,760 --> 00:01:07,080 S1: one's iPhone, someone else's iPhone at the top, like if 19 00:01:07,080 --> 00:01:09,760 S1: you just touch, don't even have to touch, but you 20 00:01:09,760 --> 00:01:13,960 S1: get close enough. It's NFC based and you can transfer 21 00:01:13,959 --> 00:01:16,280 S1: your contact. Every time I do this, people are like, 22 00:01:16,280 --> 00:01:19,160 S1: what have you done? Like they can't believe it. It's 23 00:01:19,160 --> 00:01:22,880 S1: got like this little wobbly like liquid effect. It's like 24 00:01:22,880 --> 00:01:25,760 S1: the coolest thing ever. And it's how I like to 25 00:01:25,800 --> 00:01:30,280 S1: exchange contacts with people who have iPhones. And yeah, I 26 00:01:30,280 --> 00:01:33,280 S1: don't know why more people aren't doing it or don't 27 00:01:33,280 --> 00:01:36,120 S1: even know the feature exists. And this is in San Francisco, 28 00:01:36,120 --> 00:01:39,319 S1: by the way. This is like the center mass of 29 00:01:39,319 --> 00:01:41,840 S1: Apple and the center mass of like the tech scene. 30 00:01:42,600 --> 00:01:45,160 S1: And people in San Francisco don't know how to do 31 00:01:45,160 --> 00:01:48,040 S1: this thing. They don't know that it exists. Really glad 32 00:01:48,040 --> 00:01:50,840 S1: I bought a bunch of TSMC last week because it 33 00:01:50,840 --> 00:01:55,440 S1: just massively crashed. Uh, yeah, that's that's awesome. All good though. 34 00:01:55,480 --> 00:01:58,280 S1: Playing the long game, I think it's fine. We'll actually 35 00:01:58,280 --> 00:02:00,480 S1: talk about that a little a little bit later. Just 36 00:02:00,480 --> 00:02:04,000 S1: finished reading a book for UL. Book club. Picture of 37 00:02:04,000 --> 00:02:06,760 S1: Dorian gray. Lots to say about that. That's a whole 38 00:02:06,760 --> 00:02:09,519 S1: nother talk show. But one thing I will say is 39 00:02:09,520 --> 00:02:14,480 S1: that read classics. You need to read classics. They are 40 00:02:14,600 --> 00:02:19,640 S1: more dense with value than almost any other type of book, 41 00:02:19,639 --> 00:02:22,560 S1: probably any other type of book. And every time I 42 00:02:22,560 --> 00:02:24,919 S1: read one, I'm like, how many of these are just 43 00:02:24,919 --> 00:02:28,280 S1: sitting there full of this much knowledge and I'm not 44 00:02:28,280 --> 00:02:32,320 S1: reading them. Infuriating. So luckily we have UL Book Club 45 00:02:32,320 --> 00:02:36,880 S1: to prod us into doing that. Here's what we do 46 00:02:36,919 --> 00:02:46,640 S1: in UL Book Club we oscillate between non-fiction. Fiction, non-fiction. Classic. Non-fiction. Fiction. Nonfiction. Classic. 47 00:02:47,080 --> 00:02:50,480 S1: So what we're doing is half non-fiction and half or 48 00:02:50,520 --> 00:02:55,320 S1: half non-fiction, one quarter classic and one quarter fiction. And 49 00:02:55,320 --> 00:02:57,639 S1: that has been a good mix for us. And we've 50 00:02:57,639 --> 00:03:02,150 S1: been doing this since like 2016 or something. So a 51 00:03:02,150 --> 00:03:04,630 S1: long time this book club has been going. This week's 52 00:03:04,630 --> 00:03:08,910 S1: discovery section is really good. I'm making some tweaks to 53 00:03:08,950 --> 00:03:11,710 S1: make it even better. So this is kind of the 54 00:03:11,710 --> 00:03:16,510 S1: first instance of that. Had a great conversation with Faisal Khan, 55 00:03:16,910 --> 00:03:20,670 S1: a GRC. Guy over at Vanta, and that was a 56 00:03:20,710 --> 00:03:25,430 S1: sponsored interview. And a fantastic last couple of sponsored interviews 57 00:03:25,430 --> 00:03:28,670 S1: have just been really, really good. So that was fun. 58 00:03:28,990 --> 00:03:33,790 S1: I'm going to jump into security Sonicwall vulnerability being actively exploited. 59 00:03:33,830 --> 00:03:38,230 S1: Really nasty vulnerability in their SMA 1000 appliances. And it 60 00:03:38,230 --> 00:03:40,350 S1: is being used in the wild and it is a 61 00:03:40,350 --> 00:03:43,790 S1: 9.8 on the Richter scale. My buddy Sam Curry and 62 00:03:43,830 --> 00:03:47,190 S1: Shubham Shah found that they could remotely unlock, start and 63 00:03:47,190 --> 00:03:54,590 S1: track Subarus. So a couple of the goats of Bounty and, uh, yeah, 64 00:03:54,630 --> 00:03:57,550 S1: found a really nasty one. And it looks like the 65 00:03:57,550 --> 00:04:01,830 S1: network that they're using actually involves other manufacturers as well. 66 00:04:01,830 --> 00:04:04,550 S1: So it's not just Subaru, but I think the main 67 00:04:04,670 --> 00:04:07,190 S1: effect was on the Subaru stuff and got a thread 68 00:04:07,190 --> 00:04:09,990 S1: here on the downsides of everyone getting a coding assistant, 69 00:04:09,990 --> 00:04:11,990 S1: where I say one of the biggest impacts of AI 70 00:04:11,990 --> 00:04:14,750 S1: that goes kind of unnoticed is that we're about to 71 00:04:14,750 --> 00:04:19,310 S1: see an explosion of poorly built applications, specifically applications built 72 00:04:19,310 --> 00:04:24,830 S1: completely by AI with no thought of security whatsoever. And 73 00:04:24,830 --> 00:04:28,590 S1: the reason I a little bit of like opening the 74 00:04:28,589 --> 00:04:32,790 S1: kimono here, the reason I noticed this, or just wanted 75 00:04:32,790 --> 00:04:36,750 S1: to think about it, is because I'm doing it. I 76 00:04:36,750 --> 00:04:40,630 S1: see myself being sloppy now I'm going and cleaning up 77 00:04:40,870 --> 00:04:44,270 S1: in most cases. But but there are certain cases where 78 00:04:44,310 --> 00:04:47,230 S1: like it's not really publicly exposed. I'm not too worried 79 00:04:47,230 --> 00:04:50,950 S1: about it, like it's only running local. I am moving 80 00:04:50,950 --> 00:04:55,790 S1: so fast that I can see myself being sloppy. There's 81 00:04:55,790 --> 00:05:01,630 S1: another thing that is really seductive about AI coding that 82 00:05:01,630 --> 00:05:04,230 S1: you really have to watch out for, which is the 83 00:05:04,230 --> 00:05:08,030 S1: better it gets, the more it encourages you to just 84 00:05:08,029 --> 00:05:10,589 S1: kind of trust it and just kind of go with it. 85 00:05:11,230 --> 00:05:15,030 S1: And this is bad because it's bad for a number 86 00:05:15,029 --> 00:05:17,830 S1: of reasons. One, it's bad for the security reason because 87 00:05:17,830 --> 00:05:23,310 S1: what you don't understand, you can't secure. Okay. But another 88 00:05:23,310 --> 00:05:28,270 S1: reason is it just stops you from actually getting under 89 00:05:28,270 --> 00:05:32,270 S1: the covers and actually doing the mental work of figuring 90 00:05:32,270 --> 00:05:34,630 S1: out how the components should be laid out. So the 91 00:05:34,630 --> 00:05:37,710 S1: better the AI's get, what you'll do is you'll just 92 00:05:37,710 --> 00:05:43,230 S1: drop a a stupid paragraph into your AI agent in 93 00:05:43,230 --> 00:05:47,550 S1: a coding tool like Klein, or like cursor, or like 94 00:05:47,550 --> 00:05:51,510 S1: windsurf or something. This is a starting point that so 95 00:05:51,510 --> 00:05:54,670 S1: many people are doing right now. It's not my recommended 96 00:05:54,710 --> 00:05:56,150 S1: way to way to start. And it's not the way 97 00:05:56,150 --> 00:05:59,550 S1: I start when I really want to build a good application. 98 00:05:59,870 --> 00:06:01,750 S1: If I want to build a really good application, I 99 00:06:01,750 --> 00:06:05,110 S1: actually start with a structured prompt, which is a little 100 00:06:05,110 --> 00:06:09,350 S1: bit like a PRD document, and you have that in 101 00:06:09,350 --> 00:06:11,790 S1: the directory, and then you tell the agent to go 102 00:06:11,830 --> 00:06:13,950 S1: look at that thing, or you put it in the 103 00:06:13,950 --> 00:06:16,469 S1: place where it automatically parses it, like in the cursor 104 00:06:16,470 --> 00:06:18,750 S1: rules file, for example. And that's the way you do 105 00:06:18,750 --> 00:06:23,270 S1: it properly. But what most people are doing, I would guess, 106 00:06:23,710 --> 00:06:27,750 S1: because I even find myself doing this as well, is 107 00:06:27,750 --> 00:06:32,510 S1: they're just babbling, typing, or even dictating, oh, I want 108 00:06:32,510 --> 00:06:34,310 S1: an app that does this and oh, it's going to 109 00:06:34,310 --> 00:06:36,109 S1: be like this. Oh, and make sure it looks cool. 110 00:06:36,550 --> 00:06:38,470 S1: And oh, and by the way, like it should have 111 00:06:38,470 --> 00:06:40,350 S1: this thing, but I want the button on the left 112 00:06:40,350 --> 00:06:43,910 S1: instead of the right. It's just this long paragraph of 113 00:06:43,910 --> 00:06:47,630 S1: babble garbage. And then they press enter and then the 114 00:06:47,630 --> 00:06:51,270 S1: thing starts making stuff. Now it could be making the 115 00:06:51,270 --> 00:06:55,870 S1: most horrendously insecure things. It could be making things that 116 00:06:55,870 --> 00:07:00,830 S1: are not really scrutable or. Understandable, like they split the 117 00:07:00,830 --> 00:07:04,589 S1: logic across multiple pieces. So it's not like. It's not readable, 118 00:07:04,589 --> 00:07:10,070 S1: it's not understandable. And there's a tendency for people. Because 119 00:07:10,070 --> 00:07:12,830 S1: they want to move fast. And again, when I say people, 120 00:07:12,830 --> 00:07:15,750 S1: I mean me. And I think the situation is what, 121 00:07:15,790 --> 00:07:20,030 S1: much worse than me because I'm actually. Guarding against this actively. 122 00:07:20,030 --> 00:07:22,830 S1: So I think it's way worse for just the general public. 123 00:07:23,350 --> 00:07:26,750 S1: And it's already bad for me. But it's like what 124 00:07:26,750 --> 00:07:29,110 S1: you end up with is a thing that. Kind of works. 125 00:07:29,190 --> 00:07:32,590 S1: Which is why I use this image here, um, which 126 00:07:32,590 --> 00:07:35,550 S1: my team actually came up with. Angela came up with this, 127 00:07:35,550 --> 00:07:40,110 S1: and it's like, you end up with this monstrosity. That can, like, 128 00:07:40,150 --> 00:07:45,590 S1: fall over so easily. It's just total, absolute garbage. And 129 00:07:45,590 --> 00:07:48,310 S1: then you're like, uh, no, not like that. No, no. 130 00:07:48,310 --> 00:07:51,990 S1: Change that one thing. And then after like 20, 30, 50, 131 00:07:51,990 --> 00:07:55,060 S1: 100 iterations of this thing over the course of like 132 00:07:55,060 --> 00:07:57,660 S1: an hour and a half. Like if you're doing something 133 00:07:57,660 --> 00:07:59,380 S1: else on the side, but you come back and check 134 00:07:59,380 --> 00:08:03,340 S1: on its progress. Okay, maybe you made something, but one 135 00:08:03,340 --> 00:08:06,140 S1: can you edit it? Can you? Do you understand it? 136 00:08:06,580 --> 00:08:09,020 S1: Do you understand all the technologies that it used to 137 00:08:09,060 --> 00:08:13,860 S1: build that thing? Probably not. Did it use proper technique 138 00:08:14,140 --> 00:08:18,380 S1: and best practices to build this stuff? Probably not. It depends. 139 00:08:18,380 --> 00:08:23,660 S1: There's wide variation there, but probably not. And finally, like okay, 140 00:08:23,700 --> 00:08:26,780 S1: for all the configs, did you use cloud services? Okay cool. 141 00:08:26,780 --> 00:08:30,500 S1: Where are the API keys? Are the configs readable? Like 142 00:08:31,180 --> 00:08:34,459 S1: oh did you use cloud storage? Are those buckets locked down? 143 00:08:34,900 --> 00:08:37,980 S1: All these things are like the faster we get at 144 00:08:37,980 --> 00:08:40,900 S1: using these AI tools, and the more people get in 145 00:08:40,940 --> 00:08:45,260 S1: who aren't actual coders who never coded anything beforehand, this 146 00:08:45,260 --> 00:08:49,819 S1: problem is 100 times worse for that. So bottom line is, 147 00:08:49,820 --> 00:08:51,620 S1: there is so much good that's going to come out 148 00:08:51,620 --> 00:08:55,740 S1: of AI, and I would say it's still more good 149 00:08:55,980 --> 00:08:58,699 S1: is going to come out of AI, because I would 150 00:08:58,700 --> 00:09:02,699 S1: say creation is more important than security. That would be 151 00:09:02,700 --> 00:09:06,300 S1: my my thing. And which might be strange to hear 152 00:09:06,300 --> 00:09:09,939 S1: from a security person, but I honestly believe that's true. Um, 153 00:09:09,940 --> 00:09:14,380 S1: I would I want secure creation. That's what I want. But, um, 154 00:09:14,380 --> 00:09:18,179 S1: if I had to choose. Everything is secure and nothing happens. 155 00:09:18,340 --> 00:09:21,939 S1: I would rather have insecure creation. So I think that's 156 00:09:21,940 --> 00:09:25,780 S1: more important for humans. But bottom line is there are 157 00:09:25,900 --> 00:09:32,300 S1: externalities to these AI tools, severe externalities to these AI tools. 158 00:09:32,900 --> 00:09:38,059 S1: The better they get, the more they coax you into 159 00:09:38,100 --> 00:09:42,179 S1: not looking under the covers. And it's nasty under the 160 00:09:42,179 --> 00:09:45,020 S1: covers with a lot of these AI tools. Crater of 161 00:09:45,020 --> 00:09:48,900 S1: curl just announced they're completely abandoning Cvss scoring because it's 162 00:09:48,900 --> 00:09:55,459 S1: fundamentally broken for wide widely used open source projects. And 163 00:09:55,460 --> 00:09:59,979 S1: this is specifically Daniel Stenberg explaining how CSA recently marked 164 00:10:00,540 --> 00:10:05,540 S1: a low severity vulnerability as a critical with A91. And 165 00:10:05,540 --> 00:10:07,820 S1: he's basically saying it should not have been A91. It 166 00:10:07,820 --> 00:10:10,620 S1: should have been a low. So all this talk has 167 00:10:10,620 --> 00:10:13,380 S1: really been flying around for a long time. And they 168 00:10:13,420 --> 00:10:16,500 S1: kind of made some updates with, uh, the latest version 169 00:10:16,500 --> 00:10:20,060 S1: of CSS. But I think ultimately the problem is that 170 00:10:20,780 --> 00:10:23,900 S1: these static things are not made to be looked at constantly. 171 00:10:24,059 --> 00:10:26,540 S1: They're not made to look at the actual risk of 172 00:10:26,540 --> 00:10:29,500 S1: your org and your code base and all the context 173 00:10:29,500 --> 00:10:32,780 S1: around that, combined with the risk of the vulnerability. And 174 00:10:32,780 --> 00:10:36,100 S1: then to compare those two. And this is probably going 175 00:10:36,100 --> 00:10:38,220 S1: to surprise you, but I think I might be a 176 00:10:38,220 --> 00:10:41,380 S1: good application here. You've never heard that from me that 177 00:10:41,380 --> 00:10:46,140 S1: I might be useful. But the idea is that what 178 00:10:46,140 --> 00:10:50,180 S1: you what CVS, CVS and other tools like this are 179 00:10:50,300 --> 00:10:54,660 S1: bad at is they are they're static. Right. You put 180 00:10:54,660 --> 00:10:58,339 S1: some dumb information there about like, oh, how exposed am 181 00:10:58,340 --> 00:11:01,460 S1: I or whatever. That's not really the way to find 182 00:11:01,460 --> 00:11:03,780 S1: out how exposed you are. The real way to do 183 00:11:03,780 --> 00:11:08,340 S1: that is to pull in real time context. So the 184 00:11:08,340 --> 00:11:10,780 S1: future of all of this is like super obvious. It's 185 00:11:10,780 --> 00:11:16,699 S1: basically context plus intelligence of combining a vulnerability with your 186 00:11:17,179 --> 00:11:20,980 S1: actual like risk posture. And that's what's going to determine 187 00:11:21,460 --> 00:11:26,140 S1: like what your current risk is. That's that's just the 188 00:11:26,140 --> 00:11:30,140 S1: reality of what's coming. And so it could be that, 189 00:11:30,179 --> 00:11:34,620 S1: you know, there's nothing particularly wrong with Cvss. It's it's 190 00:11:34,620 --> 00:11:39,300 S1: always been like this. The previous versions were, you know, worse. 191 00:11:39,420 --> 00:11:42,260 S1: And then there's other systems and they also have their problems. 192 00:11:42,260 --> 00:11:44,900 S1: But ultimately it comes down to this one thing, which 193 00:11:44,900 --> 00:11:47,740 S1: is what is the state of the world? What exactly 194 00:11:47,740 --> 00:11:49,689 S1: is the vuln and how does it does it relate 195 00:11:49,690 --> 00:11:52,290 S1: to me? And if you want to do that properly 196 00:11:52,730 --> 00:11:57,410 S1: in the old days, that's that's like sending security engineers 197 00:11:57,410 --> 00:12:01,490 S1: to go do research, to study to, you know, look 198 00:12:01,490 --> 00:12:03,530 S1: at a whole bunch of docs, figure out where we 199 00:12:03,530 --> 00:12:06,250 S1: have this thing installed. We're talking about asset management. We're 200 00:12:06,250 --> 00:12:10,329 S1: talking about nightmares. I mean, you could spend weeks looking 201 00:12:10,330 --> 00:12:13,330 S1: at one alone and seeing how vulnerable you are to it. 202 00:12:13,330 --> 00:12:15,850 S1: If you want to have a good assessment. Right. So 203 00:12:15,850 --> 00:12:17,970 S1: the whole point of this is that you're going to 204 00:12:17,970 --> 00:12:19,730 S1: be able to do that much faster now. And I 205 00:12:19,730 --> 00:12:24,729 S1: think that's ultimately what replaces systems like Cvss and the change. 206 00:12:25,170 --> 00:12:29,850 S1: Healthcare ransomware attack is now officially largest health care breach 207 00:12:29,850 --> 00:12:35,690 S1: in the US 190 people affected. Now Swedish authorities just 208 00:12:35,690 --> 00:12:38,929 S1: grabbed a ship they think could an underwater internet cable 209 00:12:39,130 --> 00:12:45,130 S1: between Sweden and Latvia. This is after multiple similar incidents 210 00:12:45,130 --> 00:12:47,450 S1: and a lot of them are tying back to Russia. 211 00:12:48,050 --> 00:12:52,130 S1: All right. The big story. Deep seek in AI and tech. 212 00:12:52,610 --> 00:12:57,729 S1: So Nvidia loses $600 billion after deep sea AI breakthrough. 213 00:12:58,330 --> 00:13:01,890 S1: And the stock market overall on Monday lost $1 trillion. 214 00:13:02,650 --> 00:13:05,450 S1: So it was the single biggest day of market loss 215 00:13:05,450 --> 00:13:10,170 S1: in history. And it beat the record previously from also 216 00:13:10,170 --> 00:13:14,410 S1: from Nvidia. But um, basically here's what happened. This is 217 00:13:14,410 --> 00:13:17,570 S1: what it came down to. Like Nvidia had been a 218 00:13:17,570 --> 00:13:22,410 S1: darling of AI hype because they're the GPU leaders. Much 219 00:13:22,410 --> 00:13:24,810 S1: of the future hope of making money from AI has 220 00:13:24,809 --> 00:13:29,930 S1: been embodied by Nvidia. The idea is that GPUs rule 221 00:13:29,929 --> 00:13:34,250 S1: the AI world and Nvidia rules the GPU world. And 222 00:13:34,250 --> 00:13:38,650 S1: implicit in that assumption is that Nvidia chips are scarce 223 00:13:38,650 --> 00:13:42,450 S1: and necessary and expensive. That means anyone who wants to 224 00:13:42,450 --> 00:13:45,170 S1: be a leader will have to have lots of Nvidia chips. 225 00:13:45,410 --> 00:13:47,610 S1: So deep tech just blew that out of out of 226 00:13:47,610 --> 00:13:50,090 S1: the water because they produce something that should have cost 227 00:13:50,090 --> 00:13:53,330 S1: them like tens of billions, but they did it for 228 00:13:53,330 --> 00:13:57,650 S1: $5.6 million. They basically found workarounds that allowed them to 229 00:13:57,690 --> 00:14:03,209 S1: get a lot more performance for less resources. And it 230 00:14:03,210 --> 00:14:06,370 S1: freaked everyone out. Mostly investors, but it like it messed 231 00:14:06,370 --> 00:14:10,010 S1: with the entire stock market because the stock market is 232 00:14:10,010 --> 00:14:15,770 S1: largely pushed by tech. So basically less necessary equals less valuable. 233 00:14:15,850 --> 00:14:21,410 S1: My analysis is so what. Right. If anything, deep seek 234 00:14:21,450 --> 00:14:25,570 S1: is nothing but exciting because we're getting a more I 235 00:14:25,610 --> 00:14:30,450 S1: for less resources and we need way more AI. And 236 00:14:30,450 --> 00:14:33,210 S1: I've got a whole video which I'll link up here, 237 00:14:33,850 --> 00:14:36,610 S1: but it's like how much do we need? Way, way 238 00:14:36,610 --> 00:14:39,290 S1: more than we have. And I keep doing this thing 239 00:14:39,290 --> 00:14:42,890 S1: where I'm like, oh, we only have 0.000 whatever, ten 240 00:14:42,890 --> 00:14:46,970 S1: zeros and A19. We're only at that percentage of how 241 00:14:46,970 --> 00:14:49,810 S1: much I that we need. But I don't know that percentage. 242 00:14:49,810 --> 00:14:51,610 S1: I'm making that up. I'm trying to show lots of 243 00:14:51,610 --> 00:14:54,770 S1: zeros to prove it's like a tiny fraction of 1%. 244 00:14:54,770 --> 00:14:58,330 S1: It's a made up number. It's everyone should obviously know that. 245 00:14:58,330 --> 00:15:01,410 S1: But the point is we're just getting started. And in 246 00:15:01,450 --> 00:15:04,010 S1: that video, I lay this all out. I lay out 247 00:15:04,170 --> 00:15:08,210 S1: what types of things we as a society are going 248 00:15:08,210 --> 00:15:12,010 S1: to want to do with AI, and why that requires 249 00:15:12,450 --> 00:15:16,970 S1: so much processing, so much inference, so much intelligence to 250 00:15:16,970 --> 00:15:20,690 S1: be able to do that for basically every human problem. 251 00:15:20,690 --> 00:15:23,010 S1: And and when I say human problem, I mean that 252 00:15:23,010 --> 00:15:29,450 S1: means also business problems and stuff like that, personal problems, everything. So, uh, 253 00:15:29,450 --> 00:15:33,650 S1: I think we're very confused about this whole thing. Um, 254 00:15:33,650 --> 00:15:35,970 S1: and the advantage that Deep Seek found is an example 255 00:15:35,970 --> 00:15:38,290 S1: of what I've been calling slack in the rope. And 256 00:15:38,290 --> 00:15:41,290 S1: I've been talking about this with my friend Jai Patel 257 00:15:41,970 --> 00:15:45,570 S1: for a very long time, and he's, like, one of 258 00:15:45,570 --> 00:15:48,010 S1: one of the smartest AI people that I know. And 259 00:15:48,090 --> 00:15:51,610 S1: I basically said back in like 23, I'm like, look there. 260 00:15:51,650 --> 00:15:54,250 S1: There is so much slack here that's going to allow 261 00:15:54,250 --> 00:15:58,290 S1: people to jump way ahead, and it's going to jump 262 00:15:58,290 --> 00:16:02,490 S1: them ahead of the line, even above people who've been grinding. So, 263 00:16:02,490 --> 00:16:06,170 S1: for example, you spend billions of dollars, let's say we're 264 00:16:06,170 --> 00:16:10,690 S1: at parity between like two competitors and somebody spends billions 265 00:16:10,690 --> 00:16:14,570 S1: of dollars and they jump up by 20%, but somebody 266 00:16:14,570 --> 00:16:17,690 S1: figures out, hey, wait a minute. Like if you just 267 00:16:17,690 --> 00:16:20,570 S1: double the size of the alphabet and you add some 268 00:16:20,570 --> 00:16:24,370 S1: special characters to the mix or whatever, it actually jumps 269 00:16:24,370 --> 00:16:28,490 S1: up by 600%. Isn't that weird? And then everyone starts 270 00:16:28,490 --> 00:16:33,010 S1: doing that, and then everyone collectively jumps 600%. Well, what 271 00:16:33,010 --> 00:16:36,730 S1: happened to the last six months where competitor B just 272 00:16:36,730 --> 00:16:41,970 S1: spent $4 billion trying to go up by 20%? My 273 00:16:41,970 --> 00:16:46,920 S1: argument to Jai was basically there are so many of 274 00:16:46,920 --> 00:16:50,600 S1: those big jumps completely hidden from us because we have 275 00:16:50,600 --> 00:16:54,440 S1: no idea how any of this stuff works. So this 276 00:16:54,440 --> 00:16:57,760 S1: China team just found one of these things. They just 277 00:16:57,760 --> 00:17:01,600 S1: found through a number of things with like mixture of experts, 278 00:17:01,600 --> 00:17:06,040 S1: and they've got papers detailing everything they did. But it's like, hey, look, 279 00:17:06,040 --> 00:17:09,280 S1: we just did this. And it's like now, now it's 280 00:17:09,320 --> 00:17:11,760 S1: obvious to everyone, everyone's going to go copy that. That's 281 00:17:11,760 --> 00:17:17,360 S1: another big jump. I think there is hundreds of times 282 00:17:17,480 --> 00:17:21,600 S1: the amount of performance, like laying on the floor, laying, 283 00:17:21,640 --> 00:17:25,560 S1: sitting on the table, ready for the taking. They just 284 00:17:25,560 --> 00:17:28,639 S1: they have to be discovered. And I think as the 285 00:17:28,640 --> 00:17:34,800 S1: technology changes, there's a new algorithm, there's new transformers that 286 00:17:34,800 --> 00:17:38,240 S1: we use, maybe a completely different type of AI every 287 00:17:38,280 --> 00:17:41,800 S1: time we go into this like unknown. It's a new 288 00:17:41,800 --> 00:17:45,159 S1: set of things, a new set of tricks or slack 289 00:17:45,160 --> 00:17:47,879 S1: in the rope just sitting around to be discovered. And 290 00:17:47,880 --> 00:17:52,200 S1: my point is that those pieces of slack, those tricks 291 00:17:52,920 --> 00:17:55,800 S1: that allow you to jump forward, and I don't want 292 00:17:55,800 --> 00:17:59,240 S1: to demean with the word trick. Someone mentioned that, like 293 00:18:00,000 --> 00:18:03,040 S1: when I say trick, I just mean in retrospect, it'll 294 00:18:03,040 --> 00:18:07,360 S1: appear obvious, doesn't mean it doesn't take extreme genius and 295 00:18:07,359 --> 00:18:11,879 S1: hard work and discipline to actually discover it. So I 296 00:18:11,880 --> 00:18:14,679 S1: just want to make that clear. But the point is, 297 00:18:14,920 --> 00:18:18,639 S1: we're going to be able to just keep jumping massively. 298 00:18:19,200 --> 00:18:22,560 S1: And I think a very large percent of those jumps 299 00:18:22,560 --> 00:18:25,960 S1: are going to come from these sorts of tricks or 300 00:18:25,960 --> 00:18:29,800 S1: slack in the rope type effects. And that's what I 301 00:18:29,800 --> 00:18:33,080 S1: think we had with, uh, with deep post training is 302 00:18:33,080 --> 00:18:35,919 S1: perhaps the most powerful category of these tricks. This is 303 00:18:35,920 --> 00:18:39,200 S1: what I said back in August of last year. Most 304 00:18:39,200 --> 00:18:42,480 S1: powerful category of these tricks is like teaching a giant 305 00:18:42,640 --> 00:18:46,000 S1: alien brain how to be smart when it had tremendous 306 00:18:46,000 --> 00:18:49,200 S1: potential before, but no direction. And again, I'm saying there's 307 00:18:49,200 --> 00:18:51,880 S1: lots of these different ones, but kind of playing off 308 00:18:51,880 --> 00:18:54,639 S1: this thing from 2024, I'm going to say this in 309 00:18:54,640 --> 00:18:56,920 S1: a different way, and this is kind of a casual 310 00:18:56,920 --> 00:19:00,480 S1: way of me thinking about this, which I feel like 311 00:19:00,760 --> 00:19:03,640 S1: GI doesn't really like when I do this, because it's 312 00:19:03,640 --> 00:19:07,639 S1: not perfectly accurate. But one way I'm casually thinking about 313 00:19:07,640 --> 00:19:11,200 S1: this is that there are now two steps here the intelligence, 314 00:19:11,200 --> 00:19:13,640 S1: which is the model, and then the wisdom, which is 315 00:19:13,640 --> 00:19:16,639 S1: the reinforcement learning. So this is kind of where a 316 00:19:16,640 --> 00:19:19,560 S1: lot of this slack has come from, is the fact 317 00:19:19,560 --> 00:19:23,000 S1: that the model itself is not super intelligence by itself. 318 00:19:23,320 --> 00:19:28,440 S1: You actually have to tell it what constitutes intelligence or 319 00:19:28,480 --> 00:19:32,960 S1: wisdom or smarts or usefulness. And that's after the fact 320 00:19:33,359 --> 00:19:36,840 S1: with the and more generally, the reinforcement learning. So it's 321 00:19:36,840 --> 00:19:40,120 S1: almost like intelligence is the size of the brain. And 322 00:19:40,119 --> 00:19:43,440 S1: reinforcement learning is the life experience. And like I said, 323 00:19:43,440 --> 00:19:46,480 S1: it's not technically true, but I think it's it's pretty powerful. 324 00:19:47,080 --> 00:19:50,040 S1: So bottom line is I think the the market reaction 325 00:19:50,040 --> 00:19:53,399 S1: to all of this is very wrong. I've got another 326 00:19:53,400 --> 00:19:57,680 S1: piece in here about the total Tam of AI. In fact, 327 00:19:57,680 --> 00:19:59,840 S1: I'm just going to jump to that right now because 328 00:19:59,840 --> 00:20:03,239 S1: I don't think I mentioned it anywhere else. Let's see here. Oh, 329 00:20:03,240 --> 00:20:06,840 S1: this is really cool. From, uh, Andre, move 37 is 330 00:20:06,840 --> 00:20:09,920 S1: the word of the day. It's when an AI trained 331 00:20:10,520 --> 00:20:14,040 S1: via the trial and error process of reinforcement learning discovers 332 00:20:14,080 --> 00:20:17,959 S1: actions that are new, surprising, and secretly brilliant. So this 333 00:20:17,960 --> 00:20:22,760 S1: is in reference to Lee Soto losing in go to 334 00:20:22,800 --> 00:20:26,639 S1: an AI. And it was, uh, it was moved 37 335 00:20:26,640 --> 00:20:28,760 S1: that made everyone freak out. So I thought that was 336 00:20:28,760 --> 00:20:32,120 S1: a cool concept, but, um. Yeah, if I go here, 337 00:20:32,359 --> 00:20:35,639 S1: here we go. The total Tam for AI is a 338 00:20:35,640 --> 00:20:39,520 S1: combination of two primary components. The total cost of human 339 00:20:39,520 --> 00:20:42,920 S1: workforces Forces and the amount of money that current and 340 00:20:42,920 --> 00:20:46,600 S1: future companies will pay to start ten x or 1000 341 00:20:46,640 --> 00:20:50,600 S1: X their business. So we're talking about hundreds of trillions 342 00:20:50,600 --> 00:20:54,439 S1: of dollars. So if you go back to this, this 343 00:20:54,440 --> 00:20:59,280 S1: is why I think, uh, the whole Nvidia deep tech 344 00:20:59,320 --> 00:21:03,000 S1: thing is very mistaken. The Tam and I did a 345 00:21:03,000 --> 00:21:07,720 S1: deep tech analysis actually, of this very tentative, very non-scientific, 346 00:21:07,720 --> 00:21:13,119 S1: but it came up with like 80 to $120 trillion 347 00:21:13,119 --> 00:21:17,119 S1: as being the Tam for AI over the next ten years. 348 00:21:17,400 --> 00:21:20,600 S1: So I don't know. Again, it's better if you do 349 00:21:20,640 --> 00:21:24,240 S1: like one of the, uh, the new Google Deep research 350 00:21:24,240 --> 00:21:27,200 S1: projects to actually feed it that thing and then give 351 00:21:27,200 --> 00:21:30,679 S1: that to deep Seac or O1 or O3 or whenever 352 00:21:30,680 --> 00:21:33,560 S1: it comes out. Um, that would be a better way 353 00:21:33,560 --> 00:21:36,399 S1: to get those numbers and have it fully explain the numbers. 354 00:21:36,920 --> 00:21:41,950 S1: But either way, it's trillions upon trillions of dollars. So 355 00:21:42,230 --> 00:21:45,310 S1: that's why I think this whole thing is overblown and mistaken. 356 00:21:45,310 --> 00:21:47,910 S1: And that's why my attitude is very much so. What? 357 00:21:48,630 --> 00:21:52,070 S1: Because the market has gone from being foolish to overvalue 358 00:21:52,230 --> 00:21:57,109 S1: Nvidia to being foolish to undervalue it. Right. It's because 359 00:21:57,310 --> 00:22:01,750 S1: there is just so much to go, so much room 360 00:22:01,750 --> 00:22:04,750 S1: before we get to anything like a ceiling. And here's 361 00:22:04,750 --> 00:22:07,469 S1: where I talk about that. Right. So what happens to 362 00:22:07,470 --> 00:22:10,430 S1: Nvidia or any other part of the stack doesn't matter 363 00:22:10,430 --> 00:22:13,710 S1: much at all, because we're still at whatever percentage of 364 00:22:13,710 --> 00:22:15,830 S1: the amount of AI that we need in the world. 365 00:22:16,270 --> 00:22:18,510 S1: Doesn't matter how we get there. It's not predictable. Could 366 00:22:18,510 --> 00:22:22,470 S1: be ARM processors, GPUs, some brand new thing. Doesn't matter. 367 00:22:22,550 --> 00:22:26,909 S1: I think Nvidia and TSMC and all these people are 368 00:22:26,910 --> 00:22:30,710 S1: going to continue to rise because again, we're at the 369 00:22:30,710 --> 00:22:34,510 S1: tiniest little sliver. The question is who is going to 370 00:22:34,550 --> 00:22:37,550 S1: move into the next space. Can Nvidia pivot? I believe 371 00:22:37,550 --> 00:22:39,910 S1: they can. If they had to. I'm not sure they 372 00:22:39,910 --> 00:22:43,350 S1: have to. First of all, Deep Seek was built on Nvidia. 373 00:22:43,670 --> 00:22:45,950 S1: If they had better Nvidia, deep seek would be better. 374 00:22:45,990 --> 00:22:49,750 S1: Like it's like more Nvidia still makes everything better right? 375 00:22:50,030 --> 00:22:53,310 S1: And I'm not trying to just go pro Nvidia because 376 00:22:53,310 --> 00:22:58,270 S1: I'm invested. I actually got a decent amount out of Nvidia. 377 00:22:58,430 --> 00:23:02,070 S1: I want to say um, about two weeks before this, 378 00:23:02,070 --> 00:23:06,830 S1: because I made the choice to move more money into 379 00:23:06,869 --> 00:23:10,830 S1: places that would benefit from AI, and not necessarily just 380 00:23:10,830 --> 00:23:14,869 S1: the roads and bridges. Happened to get lucky there because 381 00:23:14,869 --> 00:23:18,750 S1: I pulled some money out before it went down. But, um, 382 00:23:18,750 --> 00:23:22,350 S1: I very much I did buy the dip a little bit, 383 00:23:22,590 --> 00:23:24,869 S1: and I very much would have liked to buy even 384 00:23:24,869 --> 00:23:27,310 S1: more at the dip of Nvidia, because I think it's 385 00:23:27,310 --> 00:23:31,350 S1: still massively going up along with everything else. It's like, 386 00:23:31,390 --> 00:23:36,110 S1: I don't think they're particularly perfectly, uh, set to be 387 00:23:36,109 --> 00:23:39,190 S1: the ones But everyone's going to be the one if 388 00:23:39,190 --> 00:23:40,950 S1: you're smart about it and you have a good leader, 389 00:23:40,950 --> 00:23:44,430 S1: which Jensen is. All right, enough about that. So OpenAI 390 00:23:44,470 --> 00:23:48,030 S1: launched a preview of operator which can navigate web browsers 391 00:23:48,030 --> 00:23:51,350 S1: just like a human. I think we need more generalized agents. 392 00:23:51,390 --> 00:23:54,750 S1: I didn't like the whole App Store vibe to it. 393 00:23:54,790 --> 00:23:57,910 S1: Google just dropped a massive update to Gemini. This is 394 00:23:57,910 --> 00:24:00,750 S1: like one of the sleepers, honestly. Gemini is one of 395 00:24:00,750 --> 00:24:05,470 S1: the sleepers of this whole competition. Google has been just 396 00:24:05,470 --> 00:24:08,750 S1: doing insanely good for the last six months 12 months. 397 00:24:09,310 --> 00:24:13,590 S1: I use the the flash 2.0 Pro. I think it 398 00:24:13,590 --> 00:24:17,310 S1: is for a lot of analysis where I need really 399 00:24:17,310 --> 00:24:21,430 S1: good haystack performance over like 2 million tokens. So when 400 00:24:21,430 --> 00:24:25,790 S1: I do rag searches across like all my content, for example, 401 00:24:25,790 --> 00:24:28,670 S1: and I get back like 50 documents that hit on 402 00:24:28,670 --> 00:24:34,030 S1: the rag and this is over like 3000 posts, 3200 403 00:24:34,030 --> 00:24:38,790 S1: posts since going like back to Back to 1999. So 404 00:24:38,790 --> 00:24:40,830 S1: if I have a big giant piece of content like 405 00:24:40,830 --> 00:24:43,629 S1: that and I don't want to miss, like the, the 406 00:24:43,670 --> 00:24:46,869 S1: nuggets or the or the, you know, the needles inside 407 00:24:46,869 --> 00:24:49,790 S1: of this massive haystack, I send it over to Google 408 00:24:49,790 --> 00:24:53,790 S1: and it does a better job of of finding the haystack. Plus, 409 00:24:53,830 --> 00:24:56,630 S1: I mean, just brute force. It has 2 million tokens 410 00:24:56,630 --> 00:25:01,030 S1: as opposed to 200,000 for like sonic anthropic build citations 411 00:25:02,030 --> 00:25:06,630 S1: API to combat hallucinations. That's really cool. Uh, Google just 412 00:25:06,630 --> 00:25:10,710 S1: dropped another billion dollars into anthropic. Very interesting. So they're 413 00:25:10,710 --> 00:25:14,230 S1: not only doing their own play. They're also betting against 414 00:25:14,230 --> 00:25:18,230 S1: OpenAI with anthropic. I don't know if they're in OpenAI 415 00:25:18,270 --> 00:25:21,510 S1: as well. But anyway, they're investing in anthropic, which I 416 00:25:21,510 --> 00:25:25,630 S1: thought was really interesting. Leaked memo from Apple says they're 417 00:25:25,630 --> 00:25:32,230 S1: focusing completely on rebuilding series infrastructure and improving their existing 418 00:25:32,270 --> 00:25:35,710 S1: AI models this year. Uh, which, yeah, it's the AI 419 00:25:35,710 --> 00:25:38,820 S1: eye chief talking about it. So obviously they're talking about AI. Well, 420 00:25:39,140 --> 00:25:43,620 S1: overall startup funding has dropped significantly since 2021. Seed rounds 421 00:25:43,619 --> 00:25:48,540 S1: are actually getting bigger. This is interesting. My friend. Uh, Mike, um, 422 00:25:48,540 --> 00:25:52,100 S1: has a really, really cool podcast on this, uh, which 423 00:25:52,100 --> 00:25:54,700 S1: you could check in the show notes. His name is 424 00:25:54,700 --> 00:25:59,260 S1: Mike Privett, and you should absolutely go check out his stuff. Uh, 425 00:25:59,300 --> 00:26:03,220 S1: Colorado police are now giving away free AirTags and the 426 00:26:03,220 --> 00:26:06,540 S1: tile trackers to help prevent vehicle theft in their community. 427 00:26:06,580 --> 00:26:11,060 S1: A ring camera in Canada. This is insane. Caught the 428 00:26:11,060 --> 00:26:15,340 S1: exact moment of a meteorite smashing into a basically a 429 00:26:15,380 --> 00:26:17,859 S1: walkway in front of their house. So they just came 430 00:26:17,859 --> 00:26:21,500 S1: out and just like passed that that location and you 431 00:26:21,540 --> 00:26:26,100 S1: hear like this really loud crash and it hit tiles 432 00:26:26,420 --> 00:26:29,899 S1: on the walkway. And I'm, I don't understand how it didn't, like, 433 00:26:29,900 --> 00:26:32,580 S1: blow out the tile and create a little hole in 434 00:26:32,580 --> 00:26:34,740 S1: the ground. But anyway, they picked it up. It was 435 00:26:34,740 --> 00:26:38,340 S1: an actual meteor and it just, like, blew up against 436 00:26:38,340 --> 00:26:40,740 S1: the tile. I don't know. It had to be going 437 00:26:40,780 --> 00:26:44,260 S1: thousands of miles per hour, right? Its terminal velocity, I guess. 438 00:26:44,300 --> 00:26:46,740 S1: Is it hundreds of miles an hour? I don't know, 439 00:26:46,740 --> 00:26:49,060 S1: but it had to be going fast. It definitely would 440 00:26:49,060 --> 00:26:51,580 S1: have killed someone if it hit him, I think. Anyway, 441 00:26:51,580 --> 00:26:54,340 S1: I don't know why it didn't crack the, uh, the 442 00:26:54,340 --> 00:26:56,659 S1: stone on the ground. A new Harvard study shows we 443 00:26:56,660 --> 00:26:59,500 S1: should be taking blood pressure readings while lying down instead 444 00:26:59,500 --> 00:27:02,380 S1: of sitting. And Hans Zimmer is apparently in talks with 445 00:27:02,380 --> 00:27:06,900 S1: Saudi Arabia to remake their national anthem. And a pre-mortem 446 00:27:06,900 --> 00:27:09,780 S1: is basically where you imagine your product has already failed, 447 00:27:09,780 --> 00:27:12,460 S1: and you work backwards to figure out why, and you 448 00:27:12,460 --> 00:27:16,580 S1: do this before you even start the project. That's brilliant. 449 00:27:16,820 --> 00:27:20,980 S1: Reminds me of the PR system inside of Amazon, which 450 00:27:20,980 --> 00:27:23,700 S1: I learned a lot because we use that at Apple. 451 00:27:23,859 --> 00:27:28,140 S1: And uh, yeah, four components of top AI model ecosystems. 452 00:27:28,140 --> 00:27:31,860 S1: It's in the ideas section related heavily to this deep 453 00:27:31,900 --> 00:27:36,900 S1: tech stuff and for discovery. Klein is my new go 454 00:27:36,940 --> 00:27:40,940 S1: to for doing AI stuff. It is an extension built 455 00:27:40,940 --> 00:27:45,420 S1: into standard VS code and actually like it more than 456 00:27:46,300 --> 00:27:51,060 S1: using cursor, which is like more integrated and everything. Klein 457 00:27:51,140 --> 00:27:56,780 S1: is more conversational. It's more following my thread. It's just 458 00:27:56,780 --> 00:28:01,020 S1: seems more intelligent and more like cohesive in the way 459 00:28:01,020 --> 00:28:05,060 S1: that it answers things. Um, and again, it's, uh, I'll 460 00:28:05,060 --> 00:28:09,300 S1: show you the website. Yeah. Klein bot thoughtful AI coder. 461 00:28:09,740 --> 00:28:12,500 S1: I don't know why it's better, but honestly, it feels 462 00:28:12,500 --> 00:28:14,740 S1: better to me. I take this pain right here. I 463 00:28:14,740 --> 00:28:17,139 S1: actually move it over to the right. Uh, that's how 464 00:28:17,140 --> 00:28:19,739 S1: I have it going. Developer created a brilliant trap for 465 00:28:19,740 --> 00:28:24,820 S1: web scrapers by using specifically crafted CSS selectors. So anti 466 00:28:24,859 --> 00:28:29,580 S1: scraper trap, I recommend you try out deep seek. One 467 00:28:29,580 --> 00:28:34,940 S1: thing about deep seek that is really, really cool is that, uh, use. 468 00:28:34,980 --> 00:28:38,300 S1: When it responds, it just starts talking out loud and 469 00:28:38,300 --> 00:28:40,540 S1: you just see it. It's very similar to O1 in 470 00:28:40,580 --> 00:28:43,140 S1: that way. You just see it. That's the whole reason 471 00:28:43,260 --> 00:28:46,020 S1: it's doing so well in the benchmarks. That's the reason 472 00:28:46,020 --> 00:28:49,940 S1: it's controversial because it's very much acting like O1. But 473 00:28:49,940 --> 00:28:53,460 S1: bottom line is you see it having a conversation with 474 00:28:53,460 --> 00:28:56,380 S1: itself and it goes back and forth and it's like, yeah, 475 00:28:56,380 --> 00:28:58,180 S1: but this is really cool. Yeah. But have you thought 476 00:28:58,180 --> 00:29:00,860 S1: of this? Actually that could be wrong. Let me think 477 00:29:00,860 --> 00:29:03,420 S1: about this. And it does that for a number of 478 00:29:03,420 --> 00:29:07,380 S1: paragraphs before it actually gives you an answer. Um, I 479 00:29:07,380 --> 00:29:09,940 S1: recommend you try it out with Olama. It's really easy 480 00:29:09,940 --> 00:29:16,380 S1: to install and magenta NVMe is one of the I first, uh, 481 00:29:16,380 --> 00:29:19,140 S1: vim tools that I'm trying to use. Ultimately, I'd like 482 00:29:19,140 --> 00:29:21,740 S1: to get out of VSCode and have all my AI stuff, 483 00:29:21,740 --> 00:29:27,340 S1: something like Klein inside of vim, but it's hard to 484 00:29:27,340 --> 00:29:29,300 S1: do that in the terminal, which is why I'm like 485 00:29:29,300 --> 00:29:32,780 S1: stuck in VSCode and not happy about it. Ben Thompson 486 00:29:32,820 --> 00:29:37,140 S1: had a pretty cool deep sig fak fark and Lang 487 00:29:37,180 --> 00:29:39,020 S1: chain just dropped a cool new tool that lets you 488 00:29:39,020 --> 00:29:41,020 S1: do deep web research. This is the thing I was 489 00:29:41,020 --> 00:29:46,180 S1: talking about before completely using a hosted LMS. I've not 490 00:29:46,180 --> 00:29:49,460 S1: messed with this one yet, but I cannot wait to. Um. 491 00:29:49,700 --> 00:29:52,900 S1: Somebody created a simple service that converts WordPress blogs to 492 00:29:52,940 --> 00:29:57,020 S1: Hugo static sites. Highly recommend. Anyone building a website goes 493 00:29:57,020 --> 00:30:01,660 S1: static these days and basically keep your markdown. Your markdown 494 00:30:01,660 --> 00:30:04,460 S1: is your website that is your content. You can make 495 00:30:04,460 --> 00:30:07,620 S1: a rag out of it. It is just text and 496 00:30:07,620 --> 00:30:11,020 S1: basically any place you want to bring your blog to 497 00:30:11,420 --> 00:30:14,220 S1: give them the markdown and have it turn it into 498 00:30:14,260 --> 00:30:17,420 S1: a nice website that you like. You like the aesthetic of. 499 00:30:17,460 --> 00:30:21,740 S1: So it's content aesthetic and don't mix the two and 500 00:30:21,740 --> 00:30:25,020 S1: don't have it where it's like the platform owns the 501 00:30:25,020 --> 00:30:29,020 S1: mixture because then you're screwed. Um, and I've had this 502 00:30:29,020 --> 00:30:32,130 S1: happen to me multiple times. I just had it happen again. 503 00:30:32,330 --> 00:30:36,570 S1: Getting my content out of beehive, which I love. Beehive 504 00:30:36,570 --> 00:30:39,570 S1: for my newsletter, but it's not my favorite place to 505 00:30:39,610 --> 00:30:43,570 S1: run an actual website. Philips Hue bulbs getting motion sensing. 506 00:30:44,050 --> 00:30:45,850 S1: I really hope this is true and I can't wait 507 00:30:45,850 --> 00:30:49,010 S1: to get it and play with it. Be nice to 508 00:30:49,050 --> 00:30:53,050 S1: have all different hue things basically. Also be motion sensors. 509 00:30:53,690 --> 00:30:58,170 S1: How to say no as a product manager. This is great. 510 00:30:58,330 --> 00:31:01,130 S1: This is great. I think it just refreshes. Yeah it 511 00:31:01,130 --> 00:31:05,930 S1: just refreshes. Let's not do that. So I'm hitting refresh. 512 00:31:06,650 --> 00:31:09,850 S1: Can we revisit this after the next sprint refresh. We 513 00:31:09,850 --> 00:31:13,890 S1: need to ensure alignment before moving forward. Refresh. This could 514 00:31:13,890 --> 00:31:16,090 S1: be part of a larger initiative, but let's hold off 515 00:31:16,090 --> 00:31:20,130 S1: for now. This is really good. It's really good. Uh, 516 00:31:20,170 --> 00:31:24,050 S1: and recommendation of the week. Remember that AI is not 517 00:31:24,090 --> 00:31:27,970 S1: AI stocks. AI is not the survival of AI companies 518 00:31:27,970 --> 00:31:34,130 S1: that did marketing in 2023 and 2024. AI's Tam is 519 00:31:34,130 --> 00:31:37,570 S1: the replacement of human labor and the magnification of GDP 520 00:31:38,050 --> 00:31:40,810 S1: that can come from millions or billions of people becoming 521 00:31:40,810 --> 00:31:44,690 S1: a founder, builder or creator. That is the ball to 522 00:31:44,730 --> 00:31:48,930 S1: watch and everything else is noise. And the aphorism of 523 00:31:48,930 --> 00:31:52,530 S1: the week is to be completely cured of newspapers spend 524 00:31:52,530 --> 00:31:55,209 S1: a year reading the paper from the previous week to 525 00:31:55,210 --> 00:32:00,250 S1: be completely cured of newspapers, spend a year reading the 526 00:32:00,250 --> 00:32:07,130 S1: newspaper from the previous week. Nassim Nicholas Taleb. Unsupervised learning 527 00:32:07,130 --> 00:32:11,090 S1: is produced on Hindenburg Pro using an SM seven B microphone. 528 00:32:11,930 --> 00:32:14,210 S1: A video version of the podcast is available on the 529 00:32:14,210 --> 00:32:17,850 S1: Unsupervised Learning YouTube channel, and the text version with full 530 00:32:17,850 --> 00:32:23,450 S1: links and notes is available at Daniel Miessler newsletter. We'll 531 00:32:23,450 --> 00:32:24,250 S1: see you next time.