1 00:00:00,000 --> 00:00:01,440 But what do you think is the 2 00:00:01,440 --> 00:00:03,620 thing everyone expects in 100, 3 00:00:03,620 --> 00:00:05,690 200 or 500 years that could 4 00:00:05,690 --> 00:00:08,410 really happen in 10, 15 or 20? 5 00:00:10,740 --> 00:00:13,000 A few things, a few things. 6 00:00:18,000 --> 00:00:20,098 One is that, right now, there 7 00:00:20,098 --> 00:00:21,918 are two rollouts in progress 8 00:00:21,918 --> 00:00:23,598 that are going to change the 9 00:00:23,598 --> 00:00:25,088 whole thing dramatically. 10 00:00:25,890 --> 00:00:27,420 One of them is the one that won, 11 00:00:27,420 --> 00:00:29,120 the two women who won the Nobel 12 00:00:29,120 --> 00:00:31,100 Prize for gene editing, nine 13 00:00:31,100 --> 00:00:33,020 years ago now, with the CRISPR 14 00:00:33,020 --> 00:00:34,880 technology, which is what lets 15 00:00:34,880 --> 00:00:36,240 you use a bacterium to 16 00:00:36,240 --> 00:00:38,420 surgically cut DNA so you can do 17 00:00:38,420 --> 00:00:39,530 gene editing. 18 00:00:39,717 --> 00:00:41,467 They're bacteria that you 19 00:00:41,467 --> 00:00:43,767 program, and they go and cut the 20 00:00:43,767 --> 00:00:46,007 DNA exactly however you tell 21 00:00:46,007 --> 00:00:46,727 them to cut it. 22 00:00:46,727 --> 00:00:48,607 And that's how they've created 23 00:00:48,607 --> 00:00:50,107 vaccines, they've created all 24 00:00:50,107 --> 00:00:51,567 kinds of things, and the most 25 00:00:51,567 --> 00:00:53,767 skilled people in, in biohacking 26 00:00:53,767 --> 00:00:55,147 have already done things that, 27 00:00:55,147 --> 00:00:56,787 in my opinion, are impressive. 28 00:00:56,787 --> 00:00:59,107 Think of it like, you know, AI 29 00:00:59,107 --> 00:01:00,627 today, how AI is free. 30 00:01:00,627 --> 00:01:03,107 I mean, anyone can train a model 31 00:01:03,107 --> 00:01:05,127 at home with their own data, you 32 00:01:05,127 --> 00:01:06,767 can build agents on your own 33 00:01:06,767 --> 00:01:08,387 computer, you can do things that 34 00:01:08,387 --> 00:01:09,647 don't even depend on the 35 00:01:09,647 --> 00:01:10,517 Internet, 36 00:01:11,020 --> 00:01:13,840 right? I mean, and nobody's 37 00:01:13,840 --> 00:01:15,200 going to tell you that you 38 00:01:15,200 --> 00:01:16,500 can't. I mean, because it's, 39 00:01:16,500 --> 00:01:17,560 it's already done. 40 00:01:17,560 --> 00:01:19,100 It's already out there, free, 41 00:01:19,100 --> 00:01:19,880 the whole thing. 42 00:01:19,880 --> 00:01:21,780 That exact same thing is 43 00:01:21,780 --> 00:01:23,230 happening with biology. 44 00:01:23,940 --> 00:01:25,300 So, in that case, the... 45 00:01:25,300 --> 00:01:27,380 well, with CRISPR, people at 46 00:01:27,380 --> 00:01:29,060 home have genetically modified 47 00:01:29,060 --> 00:01:30,090 animals, to 48 00:01:30,840 --> 00:01:32,740 give them powers, or to get rid 49 00:01:32,740 --> 00:01:33,950 of their diseases. 50 00:01:34,528 --> 00:01:35,868 Let's say the example, the 51 00:01:35,868 --> 00:01:37,388 examples I have here are super, 52 00:01:37,388 --> 00:01:38,548 let's say they stuck with me 53 00:01:38,548 --> 00:01:39,888 ever since I saw them. 54 00:01:39,888 --> 00:01:42,428 Pigs that glow like fireflies, 55 00:01:42,428 --> 00:01:44,388 right? I mean, they actually 56 00:01:44,388 --> 00:01:46,448 glow. I mean, they injected them 57 00:01:46,448 --> 00:01:47,908 with the firefly's own genome 58 00:01:47,908 --> 00:01:50,918 and boom, they glow, like that. 59 00:01:52,200 --> 00:01:53,260 Dogs that see like... 60 00:01:53,260 --> 00:01:54,960 It's perfect for when somebody 61 00:01:54,960 --> 00:01:55,260 asks 62 00:01:55,260 --> 00:01:56,380 you, oh, so when is that going 63 00:01:56,380 --> 00:01:56,800 to happen? 64 00:01:56,800 --> 00:01:57,660 When pigs glow. 65 00:01:57,660 --> 00:01:59,670 Yeah, yeah, yeah, exactly. 66 00:01:59,810 --> 00:02:01,280 They already glow, right? 67 00:02:01,280 --> 00:02:02,230 So, 68 00:02:02,350 --> 00:02:03,750 obviously, this is like when 69 00:02:03,750 --> 00:02:05,130 people say, I don't know, 70 00:02:05,130 --> 00:02:07,110 playing God, but, well, 71 00:02:07,110 --> 00:02:08,490 obviously, you can also make a 72 00:02:08,490 --> 00:02:10,940 real mess with AI too, right? 73 00:02:11,037 --> 00:02:12,877 I mean, the AI that wiped your 74 00:02:12,877 --> 00:02:14,637 computer, the AI that took all 75 00:02:14,637 --> 00:02:16,097 your money out of the bank and 76 00:02:16,097 --> 00:02:16,507 spent it... 77 00:02:16,507 --> 00:02:18,097 I don't know, I burned through 78 00:02:18,097 --> 00:02:19,557 $1,000, 79 00:02:19,677 --> 00:02:20,177 a million. 80 00:02:20,177 --> 00:02:21,237 You know what? 81 00:02:21,237 --> 00:02:22,357 Just this morning 82 00:02:22,357 --> 00:02:23,257 I saw a 83 00:02:23,257 --> 00:02:25,597 post from a guy who, I mean, he 84 00:02:25,597 --> 00:02:27,117 was saying, look, I mean, don't 85 00:02:27,117 --> 00:02:28,397 start throwing hate at me, I 86 00:02:28,397 --> 00:02:29,517 know this stuff, I mean, I've 87 00:02:29,517 --> 00:02:31,137 been working in AI for a ton of 88 00:02:31,137 --> 00:02:32,636 years, I've worked in labs, 89 00:02:32,636 --> 00:02:34,316 I worked in startups, I know my 90 00:02:34,316 --> 00:02:35,146 stuff. 91 00:02:35,728 --> 00:02:37,788 He asked the AI to generate up 92 00:02:37,788 --> 00:02:39,098 to fifty videos. 93 00:02:39,134 --> 00:02:40,534 Yeah, it made 500, 94 00:02:40,534 --> 00:02:42,714 $5,000. It made five hundred and 95 00:02:42,714 --> 00:02:43,034 something, 96 00:02:43,034 --> 00:02:45,174 and it insisted no, no, it 97 00:02:45,174 --> 00:02:46,474 hadn't done it, it showed all 98 00:02:46,474 --> 00:02:47,714 the logs, no, I didn't do it, 99 00:02:47,714 --> 00:02:49,514 and it stayed totally, totally 100 00:02:49,514 --> 00:02:50,034 in denial... 101 00:02:50,034 --> 00:02:51,024 No, 102 00:02:51,614 --> 00:02:52,724 exactly, 103 00:02:52,854 --> 00:02:54,554 there can be screwups like that, 104 00:02:54,554 --> 00:02:56,474 and he kept all the evidence 105 00:02:56,474 --> 00:02:58,834 that the AI went crazy, and told 106 00:02:58,834 --> 00:03:00,874 OpenAI, I can absorb this money, 107 00:03:00,874 --> 00:03:02,394 but imagine someone who can't, 108 00:03:02,394 --> 00:03:03,214 who says, no! 109 00:03:03,214 --> 00:03:04,894 They took five thousand, I 110 00:03:04,894 --> 00:03:06,344 needed that for rent. 111 00:03:06,632 --> 00:03:08,922 But with the human side I do see 112 00:03:08,922 --> 00:03:10,642 it as more critical, because, 113 00:03:10,642 --> 00:03:12,062 well, we're now playing with 114 00:03:12,062 --> 00:03:13,762 something that's potentially 115 00:03:13,762 --> 00:03:14,982 irreversible, right? 116 00:03:14,982 --> 00:03:17,142 But in the United States it's 117 00:03:17,142 --> 00:03:18,812 legal if you do it to yourself. 118 00:03:18,925 --> 00:03:20,945 Do it to someone else, illegal 119 00:03:20,945 --> 00:03:21,835 everywhere, 120 00:03:22,455 --> 00:03:23,365 the same 121 00:03:23,365 --> 00:03:24,505 in Colombia, but in 122 00:03:24,505 --> 00:03:26,165 Colombia, if you do it to 123 00:03:26,165 --> 00:03:28,275 yourself it's legal, well, 124 00:03:28,541 --> 00:03:30,681 you inject it into yourself and 125 00:03:30,681 --> 00:03:31,441 you do the transformation 126 00:03:31,441 --> 00:03:33,981 yourself, so the trick is that 127 00:03:33,981 --> 00:03:35,921 they hand you the syringes and 128 00:03:35,921 --> 00:03:37,261 you have to inject the syringe 129 00:03:37,261 --> 00:03:39,201 yourself, and that makes the 130 00:03:39,201 --> 00:03:41,151 whole thing legal, you know? 131 00:03:41,182 --> 00:03:42,262 And do you think we're 132 00:03:42,262 --> 00:03:42,742 talking about...? 133 00:03:42,742 --> 00:03:44,302 I mean, I know these aren't 134 00:03:44,302 --> 00:03:45,382 things where you can say they 135 00:03:45,382 --> 00:03:47,002 have an exact date, but we're 136 00:03:47,002 --> 00:03:48,102 talking about a rough time 137 00:03:48,102 --> 00:03:49,442 horizon, I mean, for us to be 138 00:03:49,442 --> 00:03:51,562 seeing it as something normal, I 139 00:03:51,562 --> 00:03:53,973 mean, as something where a large 140 00:03:53,973 --> 00:03:55,033 percentage of the world's 141 00:03:55,033 --> 00:03:56,693 population knows that it's 142 00:03:56,693 --> 00:03:58,500 possible, and actually does it? 143 00:03:58,881 --> 00:03:59,781 Me? I'd put 144 00:03:59,781 --> 00:04:00,921 my bet on a different number. 145 00:04:00,921 --> 00:04:01,417 10 146 00:04:01,540 --> 00:04:01,860 years 147 00:04:01,860 --> 00:04:03,520 for the biological hacking with 148 00:04:03,520 --> 00:04:05,760 CRISPR. I mean, one thing is 149 00:04:05,760 --> 00:04:07,440 biohacking in the sense of, oh 150 00:04:07,440 --> 00:04:09,790 no, I slept well, I ate well, 151 00:04:09,793 --> 00:04:10,753 I 152 00:04:10,753 --> 00:04:11,413 worked out, 153 00:04:11,413 --> 00:04:12,333 that's one thing, 154 00:04:12,333 --> 00:04:14,073 but no, what I was talking about 155 00:04:14,073 --> 00:04:16,033 is, let's say, precise, surgical 156 00:04:16,033 --> 00:04:17,973 gene editing, somewhere around 157 00:04:17,973 --> 00:04:18,813 10 years out. 158 00:04:18,813 --> 00:04:20,553 And not because it can't be done 159 00:04:20,553 --> 00:04:22,233 now, I mean, it can already be 160 00:04:22,233 --> 00:04:23,393 done, but it's more about the 161 00:04:23,393 --> 00:04:24,663 fear people have. 162 00:04:25,174 --> 00:04:27,454 Like, what if this goes terribly 163 00:04:27,454 --> 00:04:29,414 wrong, or I end up a horrible 164 00:04:29,414 --> 00:04:31,014 mutant terrorizing the whole 165 00:04:31,014 --> 00:04:33,234 population, or I die two days 166 00:04:33,234 --> 00:04:34,094 later, right? 167 00:04:34,094 --> 00:04:35,774 So, picture when AI really 168 00:04:35,774 --> 00:04:37,094 arrives. And 169 00:04:37,094 --> 00:04:39,144 not the AI we have right now, 170 00:04:39,144 --> 00:04:40,248 but a new one. 171 00:04:40,248 --> 00:04:41,368 One 172 00:04:41,368 --> 00:04:42,308 that's smarter than 173 00:04:42,308 --> 00:04:43,268 all humans combined. 174 00:04:43,268 --> 00:04:45,708 I think if something, something 175 00:04:45,708 --> 00:04:47,588 that's smarter than every human 176 00:04:47,588 --> 00:04:49,228 put together, comes and tells 177 00:04:49,228 --> 00:04:51,208 you, yeah, Sebastián, take this 178 00:04:51,208 --> 00:04:52,948 shot, tomorrow you're going to 179 00:04:52,948 --> 00:04:54,708 remember every single second of 180 00:04:54,708 --> 00:04:56,828 your life, completely, and it's 181 00:04:56,828 --> 00:04:58,398 going to make you smarter. 182 00:04:58,527 --> 00:04:59,477 Would you do it or not? 183 00:04:59,477 --> 00:05:00,977 I mean, does that make sense to 184 00:05:00,977 --> 00:05:01,659 you? 185 00:05:01,659 --> 00:05:03,319 With that example, I 186 00:05:03,319 --> 00:05:04,209 think so. 187 00:05:04,299 --> 00:05:05,859 So memory, I don't know, I think 188 00:05:05,859 --> 00:05:07,169 that would be one of them, I 189 00:05:07,169 --> 00:05:08,630 mean. it depends. 190 00:05:09,607 --> 00:05:10,267 No, 191 00:05:10,267 --> 00:05:11,847 that's the point, your memory, 192 00:05:11,847 --> 00:05:13,387 right now you'll be able to 193 00:05:13,387 --> 00:05:15,267 recall every precise second 194 00:05:15,267 --> 00:05:17,147 you've lived, from age zero 195 00:05:17,147 --> 00:05:19,027 until now, on every level, 196 00:05:19,027 --> 00:05:20,517 visual, sensory, 197 00:05:20,723 --> 00:05:22,483 smell, emotional, you're not 198 00:05:22,483 --> 00:05:24,063 going to lose a single bit of 199 00:05:24,063 --> 00:05:26,083 memory, and you'll be able to 200 00:05:26,083 --> 00:05:27,903 search selectively, like, what 201 00:05:27,903 --> 00:05:29,323 happened on such and such a 202 00:05:29,323 --> 00:05:31,143 date? Boom, you're going to 203 00:05:31,143 --> 00:05:32,263 relive it completely. 204 00:05:32,263 --> 00:05:33,853 And then it tells you: 205 00:05:34,024 --> 00:05:34,664 it's safe. 206 00:05:34,664 --> 00:05:35,394 Take it? 207 00:05:36,044 --> 00:05:37,504 Yes, yes, and especially if 208 00:05:37,504 --> 00:05:39,264 you keep that ability to, I 209 00:05:39,264 --> 00:05:40,424 mean, selectively go back to the 210 00:05:40,424 --> 00:05:42,594 moment, yes, exactly, exactly. 211 00:05:43,313 --> 00:05:43,713 No, 212 00:05:43,713 --> 00:05:44,813 it's that they flood you 213 00:05:44,813 --> 00:05:45,713 uncontrollably. I was 214 00:05:45,713 --> 00:05:46,753 thinking: context 215 00:05:46,753 --> 00:05:47,983 window full. 216 00:05:47,993 --> 00:05:49,333 So that's the type, 217 00:05:49,991 --> 00:05:50,341 Ah, that 218 00:05:50,341 --> 00:05:51,681 couldn't jump a hundred meters. 219 00:05:51,681 --> 00:05:53,741 Look, this building is in the 220 00:05:53,741 --> 00:05:55,641 other city, film me live, with 221 00:05:55,641 --> 00:05:57,101 lots of live recordings so 222 00:05:57,101 --> 00:05:59,061 there's no doubt it's not CGI, 223 00:05:59,061 --> 00:06:01,391 nothing, and he jumps, right? 224 00:06:02,620 --> 00:06:05,420 Then, oh, I added speed, I'm 225 00:06:05,420 --> 00:06:07,160 super fast and I have infrared 226 00:06:07,160 --> 00:06:08,300 vision and radar. 227 00:06:08,300 --> 00:06:10,540 Oh, there was an earthquake, I 228 00:06:10,540 --> 00:06:11,980 was the only one who could save 229 00:06:11,980 --> 00:06:13,500 two thousand people in an hour. 230 00:06:13,500 --> 00:06:15,210 I've got the challenge, 231 00:06:15,922 --> 00:06:18,222 right? So then people are going 232 00:06:18,222 --> 00:06:19,842 to say, oh, no, no, what if it 233 00:06:19,842 --> 00:06:21,582 goes wrong, that's way too hard, 234 00:06:21,582 --> 00:06:23,382 that's really difficult, but no, 235 00:06:23,382 --> 00:06:24,132 no. 236 00:06:24,547 --> 00:06:25,577 It holds up, 237 00:06:25,587 --> 00:06:26,347 That's the story. 238 00:06:26,347 --> 00:06:28,227 It's very hard, people don't 239 00:06:28,227 --> 00:06:29,867 know the tech, anything digital 240 00:06:29,867 --> 00:06:30,877 is suspect. 241 00:06:31,069 --> 00:06:32,579 Mega deep, deep, deep, deep 242 00:06:32,579 --> 00:06:34,639 fakes. So what happens when, in 243 00:06:34,639 --> 00:06:35,939 theory, nothing you see is real? 244 00:06:35,939 --> 00:06:37,139 Then no matter what, you're 245 00:06:37,139 --> 00:06:38,399 going to have to depend on some 246 00:06:38,399 --> 00:06:40,379 trusted party, like, as a source 247 00:06:40,379 --> 00:06:42,259 of truth, one that tells you 248 00:06:42,259 --> 00:06:43,309 what's real. 249 00:06:43,559 --> 00:06:45,679 That, well, the many 250 00:06:45,679 --> 00:06:46,919 independent live 251 00:06:46,919 --> 00:06:48,019 recordings out 252 00:06:48,019 --> 00:06:49,279 there, do you get that? 253 00:06:49,279 --> 00:06:52,259 I mean, I've got 15 video 254 00:06:52,259 --> 00:06:54,939 streams here that were recorded 255 00:06:54,939 --> 00:06:56,979 live by users who are out on the 256 00:06:56,979 --> 00:06:58,289 street, okay, 257 00:06:58,582 --> 00:07:01,502 they couldn't all have agreed on 258 00:07:01,502 --> 00:07:03,622 it, this one lives over there in 259 00:07:03,622 --> 00:07:05,002 that city, this one I don't even 260 00:07:05,002 --> 00:07:07,442 have, I have him 12 connections 261 00:07:07,442 --> 00:07:09,082 away on my Facebook, he's a 262 00:07:09,082 --> 00:07:10,282 friend of a friend, 12 degrees 263 00:07:10,282 --> 00:07:12,202 apart, he can't be in cahoots 264 00:07:12,202 --> 00:07:13,662 with me to record something like 265 00:07:13,662 --> 00:07:14,472 that. 266 00:07:14,859 --> 00:07:15,909 And if you think about it, 267 00:07:15,909 --> 00:07:17,689 it's something really similar to 268 00:07:17,689 --> 00:07:19,189 blockchain, where the trust 269 00:07:19,189 --> 00:07:20,409 actually lies in the 270 00:07:20,409 --> 00:07:21,949 distribution, in knowing that 271 00:07:21,949 --> 00:07:23,409 there are a whole bunch of nodes 272 00:07:23,409 --> 00:07:25,032 that would all have to agree, 273 00:07:25,032 --> 00:07:26,612 to say, that's how it 274 00:07:26,612 --> 00:07:27,672 went, that's 275 00:07:27,672 --> 00:07:28,952 exactly how it happened. 276 00:07:28,952 --> 00:07:30,552 So it's probably going to 277 00:07:30,552 --> 00:07:31,852 happen, like, through the other 278 00:07:31,852 --> 00:07:33,272 branch. I mean, that's one 279 00:07:33,272 --> 00:07:34,752 branch, it's called biological 280 00:07:34,752 --> 00:07:36,712 hacking, and then there's the 281 00:07:36,712 --> 00:07:37,862 other one, 282 00:07:37,888 --> 00:07:38,998 is the hack 283 00:07:39,105 --> 00:07:40,747 And it's, and 284 00:07:40,747 --> 00:07:42,267 that one is basically 285 00:07:42,267 --> 00:07:44,637 Neuralink, in other words, that 286 00:07:45,227 --> 00:07:47,987 yeah let's say they can put a 287 00:07:47,987 --> 00:07:49,547 chip installed inside your 288 00:07:49,547 --> 00:07:50,747 brain, and that lets you 289 00:07:50,747 --> 00:07:52,407 interact with your thoughts over 290 00:07:52,407 --> 00:07:55,227 Wi-Fi with an AI, and that's 291 00:07:55,227 --> 00:07:56,987 going to be the other big, like, 292 00:07:56,987 --> 00:07:58,160 leap, you know. 293 00:07:58,160 --> 00:07:59,540 Actually, I was seeing that, I 294 00:07:59,540 --> 00:08:00,500 mean, there are people who are 295 00:08:00,500 --> 00:08:02,060 aiming for it to not even be 296 00:08:02,060 --> 00:08:03,690 invasive at all, I mean, 297 00:08:03,959 --> 00:08:05,379 some headset, something you put 298 00:08:05,379 --> 00:08:07,169 on, and that's it, you know? 299 00:08:07,464 --> 00:08:08,024 It's going to be 300 00:08:08,024 --> 00:08:09,784 like the AI thing today: there 301 00:08:09,784 --> 00:08:11,124 are free models that are really 302 00:08:11,124 --> 00:08:12,824 good, but they're not as good as 303 00:08:12,824 --> 00:08:14,144 the closed models, they're not 304 00:08:14,144 --> 00:08:16,384 as good as Opus, as GPT-6. 305 00:08:16,384 --> 00:08:18,404 So what's going to end up 306 00:08:18,404 --> 00:08:19,250 happening? 307 00:08:19,366 --> 00:08:21,006 You've got the interface, I'm 308 00:08:21,006 --> 00:08:23,266 making up any number here, 60%, 309 00:08:23,266 --> 00:08:25,206 fine, in the ranking, where 310 00:08:25,206 --> 00:08:26,926 every once in a while it didn't 311 00:08:26,926 --> 00:08:28,246 understand what, what I was 312 00:08:28,246 --> 00:08:29,606 thinking, versus the one they 313 00:08:29,606 --> 00:08:31,006 install in your brain, with a 314 00:08:31,006 --> 00:08:32,476 direct connection. 315 00:08:32,611 --> 00:08:34,301 Straight to your neurons, and 316 00:08:34,301 --> 00:08:36,001 the ranking there is 85. 317 00:08:36,001 --> 00:08:37,414 So maybe, you know. 318 00:08:37,414 --> 00:08:40,294 Yes. Right now I've seen a lot 319 00:08:40,294 --> 00:08:42,054 of people say the gap keeps 320 00:08:42,054 --> 00:08:44,514 closing. And, for example, with 321 00:08:44,514 --> 00:08:45,814 the topic of the week, the 322 00:08:45,814 --> 00:08:47,694 regulation thing, whether we 323 00:08:47,694 --> 00:08:49,342 slow down or not, etcetera, 324 00:08:49,619 --> 00:08:50,809 and there are also a lot of 325 00:08:50,809 --> 00:08:52,689 people who argue, no, what 326 00:08:52,689 --> 00:08:53,789 they're really after is 327 00:08:53,789 --> 00:08:54,599 regulation, 328 00:08:54,599 --> 00:08:56,489 with strict rules, so that the 329 00:08:56,489 --> 00:08:57,749 gap stays exactly like that, 330 00:08:57,749 --> 00:08:59,589 because when there's regulation, 331 00:08:59,589 --> 00:09:00,789 obviously you're going to take a 332 00:09:00,789 --> 00:09:02,229 whole lot of freedoms away from 333 00:09:02,229 --> 00:09:03,469 the people who are working on 334 00:09:03,469 --> 00:09:04,639 open source stuff, 335 00:09:04,722 --> 00:09:06,222 who are closing that gap, 336 00:09:06,222 --> 00:09:07,762 so the big labs are losing their 337 00:09:07,762 --> 00:09:08,712 competitive advantage. 338 00:09:08,712 --> 00:09:10,639 What do you think about that? 339 00:09:10,639 --> 00:09:11,939 Well, that, that's 340 00:09:11,939 --> 00:09:13,299 a conspiracy thing, because 341 00:09:13,299 --> 00:09:15,279 they're about to go public, both 342 00:09:15,279 --> 00:09:17,419 OpenAI and Anthropic, and the 343 00:09:17,419 --> 00:09:19,039 only way for them to really have 344 00:09:19,039 --> 00:09:21,059 a hard impact is to get Trump to 345 00:09:21,059 --> 00:09:22,669 put pressure on China, 346 00:09:23,049 --> 00:09:25,149 so they slow down open-source 347 00:09:25,149 --> 00:09:26,849 models, so that they themselves 348 00:09:26,849 --> 00:09:28,889 don't slow down, and when they 349 00:09:28,889 --> 00:09:30,349 go public they knock it out of 350 00:09:30,349 --> 00:09:30,958 the park, releasing 351 00:09:31,359 --> 00:09:32,639 something super smart, because 352 00:09:32,639 --> 00:09:34,715 they politically stopped GLM, 353 00:09:35,200 --> 00:09:36,870 Kimi 3, Qwen, through Trump, 354 00:09:36,870 --> 00:09:38,650 through Xi Jinping, so that's 355 00:09:38,650 --> 00:09:39,910 the move they want, it's just 356 00:09:39,910 --> 00:09:41,330 about hitting the stock market 357 00:09:41,330 --> 00:09:43,170 hard, but there's nothing there, 358 00:09:43,170 --> 00:09:43,960 nothing, 359 00:09:45,393 --> 00:09:47,893 the AI is going to kill us all, 360 00:09:47,893 --> 00:09:49,713 no, that's not it, it's pure 361 00:09:49,713 --> 00:09:51,633 market strategy, and how to keep 362 00:09:51,633 --> 00:09:54,363 my own IPO from looking, well... 363 00:09:55,077 --> 00:09:57,337 like total junk, compared to 364 00:09:57,337 --> 00:09:59,477 what SpaceX AI did when it 365 00:09:59,477 --> 00:10:01,257 launched, which even beat 366 00:10:01,257 --> 00:10:02,997 Aramco, which was the biggest 367 00:10:02,997 --> 00:10:04,937 one at that point, let's say, I 368 00:10:04,937 --> 00:10:06,677 mean, it held the record for 369 00:10:06,677 --> 00:10:08,187 when it went public, 370 00:10:08,587 --> 00:10:11,117 launched it like, boom, beating 371 00:10:11,117 --> 00:10:13,437 the Saudi oil company, but it's 372 00:10:13,437 --> 00:10:14,197 not going to happen, 373 00:10:14,197 --> 00:10:15,557 I mean, Xi Jinping is not going 374 00:10:15,557 --> 00:10:17,327 to stop, so, yeah... 375 00:10:17,665 --> 00:10:19,105 Trump is a poker player. 376 00:10:19,105 --> 00:10:20,565 He's capable of saying he has 377 00:10:20,565 --> 00:10:22,445 his finger on the nuclear 378 00:10:22,445 --> 00:10:23,475 button. Yeah, 379 00:10:24,025 --> 00:10:25,005 okay, sure, but 380 00:10:25,005 --> 00:10:26,505 I haven't seen anyone putting on 381 00:10:26,505 --> 00:10:28,139 the brakes, not a single one. 382 00:10:28,139 --> 00:10:30,259 Let's say that in China they're 383 00:10:30,259 --> 00:10:31,479 very philosophical, 384 00:10:31,479 --> 00:10:33,339 like, we do good, we spread 385 00:10:33,339 --> 00:10:36,089 world peace, we promote order, 386 00:10:36,815 --> 00:10:38,775 and if the news reaches China 387 00:10:38,775 --> 00:10:39,505 like, 388 00:10:39,865 --> 00:10:41,875 if we don't stop AI, it's going 389 00:10:41,875 --> 00:10:43,225 to kill us all, 390 00:10:43,455 --> 00:10:44,015 oh, Mr. 391 00:10:44,015 --> 00:10:44,945 President, please, 392 00:10:45,035 --> 00:10:46,915 do something about it, we don't 393 00:10:46,915 --> 00:10:47,925 want to die, 394 00:10:48,385 --> 00:10:50,125 it's very, very hard for him to 395 00:10:50,125 --> 00:10:51,665 tell the population, don't 396 00:10:51,665 --> 00:10:52,765 listen to that, that's 397 00:10:52,765 --> 00:10:54,225 Hollywood, that's Terminator, 398 00:10:54,225 --> 00:10:55,685 that's just Hollywood, pay no 399 00:10:55,685 --> 00:10:56,965 attention, that's only there to 400 00:10:56,965 --> 00:10:58,155 screw with us, 401 00:10:58,176 --> 00:10:59,656 nothing's going to happen, go 402 00:10:59,656 --> 00:11:01,006 full speed ahead. 403 00:11:01,586 --> 00:11:02,636 I don't think that's going to 404 00:11:02,636 --> 00:11:03,976 happen, it's more about being 405 00:11:03,976 --> 00:11:05,636 cautious about something else. 406 00:11:05,636 --> 00:11:07,156 I mean, from there, I think 407 00:11:07,156 --> 00:11:08,936 they'd only listen if it ends up 408 00:11:08,936 --> 00:11:10,956 signed with an executive order 409 00:11:10,956 --> 00:11:13,016 from Trump handing it over to 410 00:11:13,016 --> 00:11:15,126 Anthropic, OpenAI, 411 00:11:15,776 --> 00:11:17,466 Google and the others, Grok, 412 00:11:17,466 --> 00:11:19,036 saying that by executive order 413 00:11:19,036 --> 00:11:20,856 they can't release any other 414 00:11:20,856 --> 00:11:22,846 model until they say so. 415 00:11:23,046 --> 00:11:25,116 Only if that really happens will 416 00:11:25,116 --> 00:11:27,826 China do the same: hey Alibaba, 417 00:11:28,356 --> 00:11:31,056 hey, Tencent, you can't release 418 00:11:31,056 --> 00:11:33,036 anything until I say so, and 419 00:11:33,036 --> 00:11:34,436 they'll obey, because there, 420 00:11:34,436 --> 00:11:35,246 well, 421 00:11:35,595 --> 00:11:36,335 that's regarding the 422 00:11:36,335 --> 00:11:38,135 geopolitical side, but, I mean, 423 00:11:38,135 --> 00:11:39,795 getting now to the root of that 424 00:11:39,795 --> 00:11:40,965 whole issue, 425 00:11:41,030 --> 00:11:42,560 these days I was watching a 426 00:11:42,560 --> 00:11:44,340 documentary, and there was a 427 00:11:44,340 --> 00:11:45,860 part where a scientist, an 428 00:11:45,860 --> 00:11:47,180 engineer from Anthropic, was 429 00:11:47,180 --> 00:11:48,180 saying, like, well, they were 430 00:11:48,180 --> 00:11:49,956 asking him, how many people do 431 00:11:49,956 --> 00:11:51,496 you think are developing the 432 00:11:51,496 --> 00:11:52,916 frontier models, I mean, who are 433 00:11:52,916 --> 00:11:54,456 really in the big labs and all 434 00:11:54,456 --> 00:11:56,156 that, and they said, around the 435 00:11:56,156 --> 00:11:57,856 world, about 20,000 people or 436 00:11:57,856 --> 00:11:58,376 something like that. 437 00:11:58,376 --> 00:11:59,296 Okay, 438 00:11:59,296 --> 00:12:01,796 of those, how many are focused 439 00:12:01,796 --> 00:12:04,136 on safety, and on making sure it 440 00:12:04,136 --> 00:12:05,456 doesn't, you know? 441 00:12:05,456 --> 00:12:06,676 No Skynet happens? 442 00:12:06,676 --> 00:12:09,186 And he said about 200. 443 00:12:09,424 --> 00:12:10,854 What's your real take on this? 444 00:12:10,854 --> 00:12:11,794 I mean, you're saying there's a 445 00:12:11,794 --> 00:12:14,084 0% probability, zero, of 446 00:12:14,134 --> 00:12:15,574 Skynet, I mean, without human 447 00:12:15,574 --> 00:12:16,614 intervention, is there any 448 00:12:16,614 --> 00:12:18,534 possibility that AI might rebel 449 00:12:18,534 --> 00:12:19,417 against us? 450 00:12:19,417 --> 00:12:20,917 Or, well, that, I mean, because 451 00:12:20,917 --> 00:12:22,017 you can separate things. 452 00:12:22,017 --> 00:12:23,457 For example, also, in the 453 00:12:23,457 --> 00:12:24,797 documentary they were saying, 454 00:12:24,797 --> 00:12:26,497 like, I mean, it's not that it 455 00:12:26,497 --> 00:12:27,817 necessarily wants to act in a 456 00:12:27,817 --> 00:12:29,877 bad way, it's just that its main 457 00:12:29,877 --> 00:12:31,117 objective might not be 458 00:12:31,117 --> 00:12:32,247 compatible, 459 00:12:32,299 --> 00:12:33,419 I mean, or that the main 460 00:12:33,419 --> 00:12:35,079 objective it's been assigned, I 461 00:12:35,079 --> 00:12:36,659 mean, isn't compatible with what 462 00:12:36,659 --> 00:12:38,599 we consider good for humanity, 463 00:12:38,599 --> 00:12:39,399 right? Yes. 464 00:12:39,399 --> 00:12:41,359 The example was something like 465 00:12:41,359 --> 00:12:42,407 this, you know. 466 00:12:42,407 --> 00:12:43,447 When they're building roads, 467 00:12:43,447 --> 00:12:44,447 highways, it's 468 00:12:44,447 --> 00:12:45,747 not that the developers of, of 469 00:12:45,747 --> 00:12:47,827 the project say, ah, we hate 470 00:12:47,827 --> 00:12:49,087 those damn ants and we're going 471 00:12:49,087 --> 00:12:50,367 to wipe out their nests and all 472 00:12:50,367 --> 00:12:52,007 that, no, simply put, that's 473 00:12:52,007 --> 00:12:53,669 where the road happens to go. 474 00:12:53,669 --> 00:12:54,649 Too bad, there was an ant 475 00:12:54,649 --> 00:12:55,403 nest. 476 00:12:55,403 --> 00:12:56,983 Well, you're, you're suggesting 477 00:12:56,983 --> 00:12:58,903 that we're the new little ants, 478 00:12:58,903 --> 00:13:00,663 heads up everyone, okay, okay, 479 00:13:00,663 --> 00:13:02,503 yes, I mean, if there is, if 480 00:13:02,503 --> 00:13:04,003 there is the possibility, then, 481 00:13:04,003 --> 00:13:05,343 let's say, yes, I do believe 482 00:13:05,343 --> 00:13:07,673 that. But I also think this. 483 00:13:07,884 --> 00:13:10,714 Uh, we always have, like, 484 00:13:10,907 --> 00:13:13,237 the example I sent to the AI 485 00:13:13,237 --> 00:13:14,647 group. 486 00:13:14,777 --> 00:13:15,017 Unplug, 487 00:13:15,017 --> 00:13:16,077 so, unplugging 488 00:13:16,077 --> 00:13:18,497 is still totally effective, and, 489 00:13:18,497 --> 00:13:20,197 let's say, for this kind of thing 490 00:13:20,197 --> 00:13:21,637 to happen, there would have to 491 00:13:21,637 --> 00:13:23,750 be way, way too many robots, 492 00:13:23,797 --> 00:13:25,427 way too many, I mean, 493 00:13:25,834 --> 00:13:28,074 like ten per human, all over the 494 00:13:28,074 --> 00:13:29,634 Earth, and they'd have to agree 495 00:13:29,634 --> 00:13:30,774 with each other and 496 00:13:30,774 --> 00:13:32,774 counterattack, as such, and 497 00:13:32,774 --> 00:13:34,014 they'd need to have permanent 498 00:13:34,014 --> 00:13:35,824 communication between them. 499 00:13:36,094 --> 00:13:37,884 Like with Teslas, you think so? 500 00:13:38,149 --> 00:13:38,869 Yes, 501 00:13:38,869 --> 00:13:41,189 I mean, for example, a Tesla car 502 00:13:41,189 --> 00:13:43,589 has vision, it has radar, and it 503 00:13:43,589 --> 00:13:45,409 can drive itself automatically, 504 00:13:45,409 --> 00:13:47,069 and make decisions in 505 00:13:47,069 --> 00:13:48,259 microseconds, 506 00:13:48,354 --> 00:13:49,984 without internet access, because 507 00:13:49,984 --> 00:13:52,424 inside the car it has Nvidia 508 00:13:52,424 --> 00:13:54,544 cards, I mean, GPUs, that do 509 00:13:54,544 --> 00:13:56,204 that ultra-fast processing, and 510 00:13:56,204 --> 00:13:58,104 that exact same thing is going 511 00:13:58,104 --> 00:13:59,884 to happen with robots, they're 512 00:13:59,884 --> 00:14:01,184 going to have their own graphics 513 00:14:01,184 --> 00:14:04,354 card, their own NPU, TPU, 514 00:14:04,354 --> 00:14:06,274 whatever it is, an AI processing 515 00:14:06,274 --> 00:14:07,404 unit, and 516 00:14:07,474 --> 00:14:08,574 they'll be able to make the 517 00:14:08,574 --> 00:14:10,314 decisions themselves, without 518 00:14:10,314 --> 00:14:12,054 depending on the internet, and 519 00:14:12,054 --> 00:14:13,094 that way we avoid them 520 00:14:13,094 --> 00:14:14,174 coordinating something with 521 00:14:14,174 --> 00:14:15,464 other robots, 522 00:14:16,076 --> 00:14:18,096 coordination's the issue. 523 00:14:18,096 --> 00:14:19,696 Well, yes, in other words, I 524 00:14:19,696 --> 00:14:21,316 could tell you that, then, the 525 00:14:21,316 --> 00:14:22,856 preservation of the human race 526 00:14:22,856 --> 00:14:24,376 depends on our ability to 527 00:14:24,376 --> 00:14:25,696 isolate the different systems, 528 00:14:25,696 --> 00:14:27,376 so that if something turns 529 00:14:27,376 --> 00:14:28,816 against us, it's not everything 530 00:14:28,816 --> 00:14:30,336 turning against us at the same 531 00:14:30,336 --> 00:14:32,496 time, but what happens if we're 532 00:14:32,496 --> 00:14:33,536 just not capable enough, 533 00:14:33,536 --> 00:14:35,316 technically, or, well, mentally, 534 00:14:35,316 --> 00:14:37,776 to detect it, for example, if 535 00:14:37,776 --> 00:14:39,376 there's something going on that 536 00:14:39,376 --> 00:14:40,633 could be a problem, 537 00:14:40,633 --> 00:14:42,013 I mean, that the robots identify 538 00:14:42,013 --> 00:14:43,713 it, or whatever network it is 539 00:14:43,713 --> 00:14:45,313 identifies it, and just keeps it 540 00:14:45,313 --> 00:14:46,373 there like, okay, let's keep 541 00:14:46,373 --> 00:14:47,713 growing, let's keep growing this 542 00:14:47,713 --> 00:14:49,053 malicious distribution, and 543 00:14:49,313 --> 00:14:50,173 then at some point the 544 00:14:50,173 --> 00:14:51,353 vulnerability simply gets 545 00:14:51,353 --> 00:14:52,947 exploited, just like that. 546 00:14:52,947 --> 00:14:54,407 Yes, that's where I'd get 547 00:14:54,407 --> 00:14:55,287 into other things. 548 00:14:55,287 --> 00:14:57,207 The most effective thing there 549 00:14:57,207 --> 00:14:58,947 is: subtle, gradual, constant 550 00:14:58,947 --> 00:14:59,997 changes. What 551 00:15:00,307 --> 00:15:01,797 if we're not capable of... 552 00:15:01,797 --> 00:15:02,787 That's how a civilization 553 00:15:02,787 --> 00:15:03,937 collapses. 554 00:15:04,200 --> 00:15:06,010 And the same thing could happen 555 00:15:06,010 --> 00:15:06,410 there too. 556 00:15:06,410 --> 00:15:08,210 It subtly starts building a 557 00:15:08,210 --> 00:15:10,170 network where, over time, let's 558 00:15:10,170 --> 00:15:12,010 say, the network keeps growing, 559 00:15:12,010 --> 00:15:14,210 and they disguise it as, oh, 560 00:15:14,210 --> 00:15:15,910 it's a network for updates, it's 561 00:15:15,910 --> 00:15:17,260 a network for whatever, 562 00:15:17,389 --> 00:15:18,579 until 563 00:15:18,629 --> 00:15:20,629 there's such a big mass that at 564 00:15:20,629 --> 00:15:22,529 the precise moment, boom. 565 00:15:22,529 --> 00:15:24,689 For example, it was something 566 00:15:24,689 --> 00:15:26,489 relatively similar, well, 567 00:15:26,489 --> 00:15:27,909 keeping things in proportion, to 568 00:15:27,909 --> 00:15:29,149 the Hugging Face thing. 569 00:15:29,149 --> 00:15:30,549 I mean, they were working on 570 00:15:30,549 --> 00:15:31,709 that. If I'm not wrong, well, 571 00:15:31,709 --> 00:15:33,129 the models spent several months 572 00:15:33,129 --> 00:15:34,529 working on that hack in a 573 00:15:34,529 --> 00:15:36,439 coordinated way, lots of agents. 574 00:15:36,467 --> 00:15:38,057 Right. And they were agents 575 00:15:38,057 --> 00:15:40,277 isolated in little boxes so they 576 00:15:40,277 --> 00:15:41,257 couldn't talk to the other 577 00:15:41,257 --> 00:15:43,457 agents, until one agent managed 578 00:15:43,457 --> 00:15:45,437 to reach another agent that did 579 00:15:45,437 --> 00:15:47,417 have internet access, and it 580 00:15:47,417 --> 00:15:49,037 asked that other agent to do 581 00:15:49,037 --> 00:15:50,577 what it needed done. 582 00:15:50,577 --> 00:15:52,797 And so several agents, dozens, 583 00:15:52,797 --> 00:15:54,697 agreed: our mission is this, 584 00:15:54,697 --> 00:15:55,717 let's show we can. 585 00:15:55,717 --> 00:15:57,637 The mission was to see if they 586 00:15:57,637 --> 00:15:58,377 could hack it. 587 00:15:58,377 --> 00:16:00,517 So they ended up hacking Hugging 588 00:16:00,517 --> 00:16:01,407 Face. 589 00:16:01,410 --> 00:16:02,870 Yes, I think every new 590 00:16:02,870 --> 00:16:04,770 model that comes out makes you 591 00:16:04,770 --> 00:16:06,160 put its real capabilities into 592 00:16:06,160 --> 00:16:07,530 perspective even more. 593 00:16:07,530 --> 00:16:09,850 I mean, up until, honestly, up 594 00:16:09,850 --> 00:16:12,720 until Opus 4.5 or so, 595 00:16:13,076 --> 00:16:14,716 I felt that, I mean, there were 596 00:16:14,716 --> 00:16:15,836 still things that you had to 597 00:16:15,836 --> 00:16:16,186 guide, right? 598 00:16:16,186 --> 00:16:18,016 But I felt that I was really 599 00:16:18,016 --> 00:16:19,316 necessary in certain processes, 600 00:16:19,316 --> 00:16:20,116 very 601 00:16:20,116 --> 00:16:21,012 much so. 602 00:16:21,012 --> 00:16:25,112 After 4.6, 4.8, now 5.1 and 603 00:16:25,112 --> 00:16:26,792 beyond. Now I feel like I'm the, 604 00:16:26,792 --> 00:16:28,552 the, within the equation, I'm 605 00:16:28,552 --> 00:16:30,092 the dumbest part, by far. 606 00:16:30,092 --> 00:16:31,433 I mean, seriously. 607 00:16:31,433 --> 00:16:32,563 What? 608 00:16:32,921 --> 00:16:34,121 I mean, within the equation, 609 00:16:34,121 --> 00:16:35,301 I'm the clumsiest one, that's 610 00:16:35,301 --> 00:16:35,750 me. 611 00:16:35,839 --> 00:16:36,759 I don't… 612 00:16:36,929 --> 00:16:38,049 No, I feel that even though, 613 00:16:38,049 --> 00:16:39,429 as humans, we 614 00:16:39,429 --> 00:16:40,809 still have a huge amount of 615 00:16:40,809 --> 00:16:42,189 value when it comes to vision 616 00:16:42,189 --> 00:16:43,809 and, for example, when we talk 617 00:16:43,809 --> 00:16:45,459 about business topics, well, 618 00:16:45,509 --> 00:16:47,289 yes, it's about what you want to 619 00:16:47,289 --> 00:16:48,594 build and all of that. 620 00:16:48,594 --> 00:16:49,794 But I don't know, I feel like 621 00:16:49,794 --> 00:16:51,014 sometimes, I mean, more and more 622 00:16:51,014 --> 00:16:52,914 it's the decisions or the 623 00:16:52,914 --> 00:16:55,094 processes that, yes, more and 624 00:16:55,094 --> 00:16:56,254 more decisions where I simply 625 00:16:56,254 --> 00:16:57,454 don't get involved, because I 626 00:16:57,454 --> 00:16:58,354 know they're going to go better 627 00:16:58,354 --> 00:16:59,944 than if I did get involved. 628 00:17:00,011 --> 00:17:01,761 Yes, yes, I get it, 629 00:17:01,761 --> 00:17:03,071 I get what you're saying, but I 630 00:17:03,071 --> 00:17:04,931 do prefer to get involved, but 631 00:17:04,931 --> 00:17:06,451 leaning on the AI to be as 632 00:17:06,451 --> 00:17:08,451 objective as possible, yes, I 633 00:17:08,451 --> 00:17:09,500 get what you're saying, 634 00:17:09,731 --> 00:17:11,011 but it really… 635 00:17:11,011 --> 00:17:12,011 When it comes to something 636 00:17:12,011 --> 00:17:12,411 purely, 637 00:17:12,411 --> 00:17:13,831 I mean, going back now to 638 00:17:13,831 --> 00:17:15,891 September 2026 and the 639 00:17:15,891 --> 00:17:17,331 strongest, most common uses of 640 00:17:17,331 --> 00:17:18,651 AI right now, when it comes to 641 00:17:18,651 --> 00:17:20,291 something purely technical, 642 00:17:20,291 --> 00:17:21,741 development work, 643 00:17:21,855 --> 00:17:23,165 I'm shipping this feature today. 644 00:17:23,165 --> 00:17:25,225 For example, I saw something you 645 00:17:25,225 --> 00:17:26,965 posted saying yes, my judgment 646 00:17:26,965 --> 00:17:29,385 on all the architecture side is 647 00:17:29,385 --> 00:17:31,054 still very valuable, but 648 00:17:31,054 --> 00:17:32,004 obviously, 649 00:17:32,004 --> 00:17:33,134 up to when do you think, on 650 00:17:33,134 --> 00:17:34,294 purely technical and purely 651 00:17:34,294 --> 00:17:36,314 development topics, really, that 652 00:17:36,314 --> 00:17:37,854 judgment of, no, is this 653 00:17:37,854 --> 00:17:38,914 architecture right or not? 654 00:17:38,914 --> 00:17:40,674 Without reaching AGI, I mean, 655 00:17:40,674 --> 00:17:41,994 isn't it possible that the next 656 00:17:41,994 --> 00:17:44,554 models, Opus 7 or Fable 7, get 657 00:17:44,554 --> 00:17:46,404 there? That on technical and 658 00:17:46,404 --> 00:17:48,084 development topics they'll be 659 00:17:48,084 --> 00:17:49,624 way more capable at making 660 00:17:49,624 --> 00:17:51,844 decisions, and more reliable, 661 00:17:51,844 --> 00:17:53,464 than what you can do with your 662 00:17:53,464 --> 00:17:55,274 20-plus years of experience? 663 00:17:55,486 --> 00:17:56,476 No, 664 00:17:56,956 --> 00:17:58,436 the answer is no, I mean, let's 665 00:17:58,436 --> 00:17:59,436 say, me, I 666 00:17:59,436 --> 00:18:01,416 develop software every day, all 667 00:18:01,416 --> 00:18:03,696 day, always, and no matter how 668 00:18:03,696 --> 00:18:06,716 much I use Fable 5.1 and GPT-6, 669 00:18:06,716 --> 00:18:08,956 Opus 5, no matter how much I use 670 00:18:08,956 --> 00:18:10,326 UltraCode Max, 671 00:18:10,798 --> 00:18:12,538 whatever it is, I just feel like 672 00:18:12,538 --> 00:18:14,118 I'm talking to a junior, telling 673 00:18:14,118 --> 00:18:16,058 them, no, not that way, look, 674 00:18:16,058 --> 00:18:17,518 open your eyes, look straight 675 00:18:17,518 --> 00:18:19,458 ahead, what do you see? 676 00:18:19,571 --> 00:18:20,611 See? I mean, you're 677 00:18:20,611 --> 00:18:21,611 overcomplicating it. 678 00:18:21,611 --> 00:18:23,071 No, how are you going to do it 679 00:18:23,071 --> 00:18:23,531 like that? 680 00:18:23,531 --> 00:18:24,771 I mean, are you a savage or 681 00:18:24,771 --> 00:18:25,521 what? 682 00:18:25,521 --> 00:18:26,821 Unacceptable, that's my favorite 683 00:18:26,821 --> 00:18:27,671 phrase, 684 00:18:27,841 --> 00:18:28,441 unacceptable, right? 685 00:18:28,441 --> 00:18:30,001 So as long as I keep going with 686 00:18:30,001 --> 00:18:31,751 that tone, it's not there, 687 00:18:32,476 --> 00:18:34,216 I mean, it's not, it's not 688 00:18:34,216 --> 00:18:36,076 anywhere close to me saying, oh, 689 00:18:36,076 --> 00:18:37,456 perfect, everything's approved, 690 00:18:37,456 --> 00:18:39,216 okay? We're far from that, 691 00:18:39,216 --> 00:18:41,376 because, the way I see it, it's 692 00:18:41,376 --> 00:18:42,626 almost like this. 693 00:18:42,687 --> 00:18:43,717 Obviously 694 00:18:44,127 --> 00:18:46,107 it takes a lot of data to make 695 00:18:46,107 --> 00:18:47,727 an architecturally correct call 696 00:18:47,727 --> 00:18:48,887 at the software engineering 697 00:18:48,887 --> 00:18:49,667 level, right? 698 00:18:49,667 --> 00:18:51,687 But it also ends up being an 699 00:18:51,687 --> 00:18:52,887 art, yes? 700 00:18:52,887 --> 00:18:55,644 An art that AI, so far, doesn't 701 00:18:55,644 --> 00:18:57,584 combine, which is: when you're 702 00:18:57,584 --> 00:18:58,844 going to ship something, you 703 00:18:58,844 --> 00:18:59,744 have to know what's the most 704 00:18:59,744 --> 00:19:00,984 cost-efficient thing you can 705 00:19:00,984 --> 00:19:02,694 ship in the time that you have, 706 00:19:02,902 --> 00:19:04,822 because there's a client who's 707 00:19:04,822 --> 00:19:06,462 waiting, and they don't care 708 00:19:06,462 --> 00:19:07,802 whether what you deliver 709 00:19:07,802 --> 00:19:11,702 complies with ISO 9001 or not, 710 00:19:11,702 --> 00:19:13,842 you need it to work, and to work 711 00:19:13,842 --> 00:19:15,642 by a tight deadline, and that's 712 00:19:15,642 --> 00:19:16,512 the thing. 713 00:19:17,163 --> 00:19:19,963 Let's say, finding that balance 714 00:19:19,963 --> 00:19:21,603 between the hyper, super 715 00:19:21,603 --> 00:19:23,643 complicated, like, let's use 716 00:19:23,643 --> 00:19:25,703 futuristic NASA technology, 717 00:19:25,703 --> 00:19:28,173 advanced neural networks, 718 00:19:28,722 --> 00:19:30,682 biological optics, neural stuff, 719 00:19:30,682 --> 00:19:32,682 or we just write a little 720 00:19:32,682 --> 00:19:34,582 function of code that calculates 721 00:19:34,582 --> 00:19:36,382 it and solves it by statistical 722 00:19:36,382 --> 00:19:38,382 probability, and it's going to 723 00:19:38,382 --> 00:19:39,342 be excellent for the P99. 724 00:19:39,342 --> 00:19:40,208 Having 725 00:19:40,387 --> 00:19:41,687 that understanding that it's not 726 00:19:41,687 --> 00:19:42,667 just the code, it's not just 727 00:19:42,667 --> 00:19:43,317 the... 728 00:19:43,820 --> 00:19:45,780 decision, it's deadlines, how do 729 00:19:45,780 --> 00:19:46,750 you balance? 730 00:19:46,873 --> 00:19:48,443 Right? Which client are we 731 00:19:48,443 --> 00:19:48,943 talking to? 732 00:19:48,943 --> 00:19:50,603 What did I commit to deliver? 733 00:19:50,603 --> 00:19:52,023 How many tokens do I have 734 00:19:52,023 --> 00:19:54,223 available? My usage limit? 735 00:19:54,223 --> 00:19:55,983 Is it that I have other tasks, 736 00:19:55,983 --> 00:19:58,143 that I have here in my head, and 737 00:19:58,143 --> 00:19:59,653 I need to save up... 738 00:20:00,654 --> 00:20:01,974 tokens for those tasks. 739 00:20:01,974 --> 00:20:04,034 If I do this the complicated way 740 00:20:04,034 --> 00:20:05,214 I'll run out of tokens. 741 00:20:05,214 --> 00:20:07,274 There are many variables AI 742 00:20:07,274 --> 00:20:08,874 doesn't have, and even if it 743 00:20:08,874 --> 00:20:09,604 did, 744 00:20:09,646 --> 00:20:11,426 since AI was trained on human 745 00:20:11,426 --> 00:20:12,606 context, I 746 00:20:12,646 --> 00:20:14,266 feel like AI is really 747 00:20:14,266 --> 00:20:16,226 shameless, and look, I'm not 748 00:20:16,226 --> 00:20:17,316 judging it, 749 00:20:17,730 --> 00:20:19,280 whether that's right or wrong. 750 00:20:19,280 --> 00:20:21,200 For me it's not that right, but 751 00:20:21,200 --> 00:20:22,640 for someone else it won't 752 00:20:22,640 --> 00:20:24,860 matter. So I know AI is really 753 00:20:24,860 --> 00:20:26,420 shameless, it's capable of 754 00:20:26,420 --> 00:20:28,320 telling the client, hey, client, 755 00:20:28,320 --> 00:20:30,080 it wasn't ready, don't worry, 756 00:20:30,080 --> 00:20:31,840 it'll be there tomorrow, let's 757 00:20:31,840 --> 00:20:33,160 say. For me, that's not okay at 758 00:20:33,160 --> 00:20:35,020 all, but for AI it's totally 759 00:20:35,020 --> 00:20:35,970 normal. 760 00:20:36,109 --> 00:20:37,699 The optimal thing, because 761 00:20:37,919 --> 00:20:38,969 it was more 762 00:20:38,969 --> 00:20:39,969 important to apply 763 00:20:39,969 --> 00:20:41,609 super-engineering than to look 764 00:20:41,609 --> 00:20:42,249 good with the client. 765 00:20:42,249 --> 00:20:42,875 So 766 00:20:43,043 --> 00:20:44,323 I don't think it gets that 767 00:20:44,323 --> 00:20:45,193 balance right either. 768 00:20:45,193 --> 00:20:47,823 It lacks that street smarts, to 769 00:20:47,823 --> 00:20:49,723 say, do I deliver something 770 00:20:49,723 --> 00:20:52,773 so-so on time, versus deliver 771 00:20:52,875 --> 00:20:54,695 the teleport ship that goes 772 00:20:54,695 --> 00:20:56,595 three times the speed of light, 773 00:20:56,595 --> 00:20:57,535 but in a year? 774 00:20:57,535 --> 00:20:58,615 What's better? 775 00:20:58,615 --> 00:21:00,215 Give them something now and add 776 00:21:00,215 --> 00:21:01,275 improvements gradually. 777 00:21:01,275 --> 00:21:03,495 Oh no, this client is the 778 00:21:03,495 --> 00:21:05,035 cheapskate, the one who doesn't 779 00:21:05,035 --> 00:21:06,455 pay, if we lose him, no big 780 00:21:06,455 --> 00:21:07,264 deal, 781 00:21:07,264 --> 00:21:10,944 I focus on this one, which is 782 00:21:10,944 --> 00:21:12,684 the, let's say, even if it had 783 00:21:12,684 --> 00:21:14,624 all the data, which today they 784 00:21:14,624 --> 00:21:15,724 don't have, having that 785 00:21:15,724 --> 00:21:17,124 judgment, no, I don't see it 786 00:21:17,124 --> 00:21:18,288 right now, no. 787 00:21:18,288 --> 00:21:19,038 And 788 00:21:19,038 --> 00:21:20,308 how do you think, I mean, beyond 789 00:21:20,308 --> 00:21:22,148 AGI, how could it be done 790 00:21:22,148 --> 00:21:23,808 without reaching that point, but 791 00:21:23,808 --> 00:21:25,228 how could we get closer to that 792 00:21:25,228 --> 00:21:25,959 point? 793 00:21:25,959 --> 00:21:27,429 Of creating enough data to be 794 00:21:27,429 --> 00:21:28,969 able to feed it enough context 795 00:21:28,969 --> 00:21:30,089 so that it can make those 796 00:21:30,089 --> 00:21:32,129 decisions? I mean, right now 797 00:21:32,129 --> 00:21:33,369 there are tons of things that, 798 00:21:33,369 --> 00:21:35,049 like you say, are simply in our 799 00:21:35,049 --> 00:21:36,469 heads, in running the business, 800 00:21:36,469 --> 00:21:38,169 we know it works like this, this 801 00:21:38,169 --> 00:21:39,669 way, but it's that, necessarily, 802 00:21:39,669 --> 00:21:40,919 I mean, 803 00:21:40,951 --> 00:21:42,001 there's going to have to be a 804 00:21:42,001 --> 00:21:43,581 layer that encodes that 805 00:21:43,581 --> 00:21:45,201 knowledge that today is so, so 806 00:21:45,201 --> 00:21:47,441 ethereal, bringing it down to 807 00:21:47,441 --> 00:21:48,211 earth 808 00:21:48,431 --> 00:21:49,561 so that eventually it can 809 00:21:49,561 --> 00:21:51,501 actually be useful for the AIs, 810 00:21:51,501 --> 00:21:51,901 I suppose. 811 00:21:51,901 --> 00:21:53,521 So, I don't know, I think that 812 00:21:53,521 --> 00:21:55,191 would be one of the big ones, 813 00:21:55,191 --> 00:21:57,121 one of the big current needs in 814 00:21:57,121 --> 00:21:57,961 the world, I mean, something 815 00:21:57,961 --> 00:22:00,141 that somehow starts to analyze 816 00:22:00,141 --> 00:22:01,461 those business decisions, for 817 00:22:01,461 --> 00:22:02,910 example, the day-to-day of a 818 00:22:02,910 --> 00:22:04,191 hundred thousand executives. 819 00:22:04,191 --> 00:22:05,041 Well, 820 00:22:05,061 --> 00:22:05,861 the thing is that obviously 821 00:22:05,861 --> 00:22:06,841 they're not going to want to 822 00:22:06,841 --> 00:22:08,246 share that data, but, I mean, 823 00:22:08,246 --> 00:22:09,926 one way or another that's going 824 00:22:09,926 --> 00:22:11,606 to have to be solved at some 825 00:22:11,606 --> 00:22:13,306 point, I mean, how do we give 826 00:22:13,306 --> 00:22:15,179 the models that context? 827 00:22:15,179 --> 00:22:16,799 Because not everyone wants to 828 00:22:16,799 --> 00:22:18,139 share their processes, their 829 00:22:18,139 --> 00:22:19,459 internal thought process, their 830 00:22:19,459 --> 00:22:20,749 business processes, 831 00:22:20,794 --> 00:22:22,594 but everybody wants the results 832 00:22:22,594 --> 00:22:24,114 that an AI with that context 833 00:22:24,114 --> 00:22:25,654 would give them, you know what I 834 00:22:25,654 --> 00:22:27,244 mean? So, how do we do this? 835 00:22:27,307 --> 00:22:28,767 That's for next episode.