1 00:00:12,119 --> 00:00:14,040 Speaker 1: Natasha read Taylor. 2 00:00:14,440 --> 00:00:17,400 Speaker 2: I'm not sure how much you think about students using 3 00:00:17,920 --> 00:00:21,280 Speaker 2: AI to cheat, or at least to get an edge 4 00:00:21,280 --> 00:00:24,280 Speaker 2: in the classroom, but there was some interesting live research 5 00:00:24,360 --> 00:00:28,920 Speaker 2: that emerged recently from Brown University. An economics professor made 6 00:00:29,000 --> 00:00:32,640 Speaker 2: his students take a final exam in person after suspecting 7 00:00:33,000 --> 00:00:35,839 Speaker 2: rampant cheating on the take home midterm. I want to 8 00:00:35,840 --> 00:00:38,440 Speaker 2: ask you to guess the delta between the average score 9 00:00:38,479 --> 00:00:40,879 Speaker 2: on the midterm and the average score on the final. 10 00:00:41,720 --> 00:00:44,120 Speaker 3: Well, I saw this story, and it was a massive delta. 11 00:00:44,280 --> 00:00:45,240 Speaker 4: Me too, Me too. 12 00:00:45,320 --> 00:00:47,240 Speaker 5: I gotta say fifty percent. 13 00:00:47,320 --> 00:00:47,879 Speaker 4: I didn't see it. 14 00:00:47,960 --> 00:00:50,440 Speaker 2: The average score on the midterm was ninety six percent, 15 00:00:51,040 --> 00:00:54,280 Speaker 2: the average score on the final forty eight point six percent. 16 00:00:54,640 --> 00:00:56,920 Speaker 3: But there was that one kid who like, actually did 17 00:00:57,200 --> 00:00:59,080 Speaker 3: he like got an F and then it went to 18 00:00:59,160 --> 00:01:00,440 Speaker 3: like a D or something like that. 19 00:01:05,680 --> 00:01:06,679 Speaker 1: Welcome to Tech stuff. 20 00:01:06,760 --> 00:01:09,800 Speaker 2: I'ma's Volosian and this is the Week in Tech where 21 00:01:09,800 --> 00:01:11,920 Speaker 2: I'm joined by three of the world's most clubbed and 22 00:01:11,959 --> 00:01:15,840 Speaker 2: reporters to break down what's really happening in tech right now. Today, 23 00:01:15,959 --> 00:01:19,480 Speaker 2: we're joined by reed Albrigotti, Take editor a Semaphore, Taylor 24 00:01:19,520 --> 00:01:22,800 Speaker 2: Lorenz of user mag and Natasha Tiku, tech reporter at 25 00:01:22,800 --> 00:01:24,200 Speaker 2: The Washington Post's welcome all. 26 00:01:24,600 --> 00:01:26,720 Speaker 4: Hey everyone, thanks for having us. 27 00:01:26,760 --> 00:01:27,400 Speaker 3: Good to be here. 28 00:01:28,040 --> 00:01:28,720 Speaker 1: So Natasha. 29 00:01:29,400 --> 00:01:32,520 Speaker 2: Nick Thompson, the former editor of Wired and current CEO 30 00:01:32,560 --> 00:01:35,600 Speaker 2: of The Atlantic, came on tech Stuff in early January 31 00:01:35,600 --> 00:01:38,160 Speaker 2: this year to give his predictions for the year, and 32 00:01:38,200 --> 00:01:39,679 Speaker 2: one of them, the top one, was that this will 33 00:01:39,720 --> 00:01:44,800 Speaker 2: be the year of the first legit AI catastrophe. Do 34 00:01:44,880 --> 00:01:46,560 Speaker 2: you think this was the week when it happened? 35 00:01:47,760 --> 00:01:52,480 Speaker 5: No, I don't, But I think this is one of 36 00:01:52,720 --> 00:01:56,520 Speaker 5: many weeks in which we got a pretty clear picture 37 00:01:56,680 --> 00:01:59,480 Speaker 5: of what the next few years are going to look like, 38 00:01:59,560 --> 00:02:04,360 Speaker 5: which is messy, disastrous. Everybody taking incidents to try to 39 00:02:04,520 --> 00:02:08,120 Speaker 5: support their own narrative about where AI capabilities are headed, 40 00:02:08,200 --> 00:02:10,760 Speaker 5: and you know, how to protect ourselves. 41 00:02:11,040 --> 00:02:11,959 Speaker 1: So what actually happened? 42 00:02:12,680 --> 00:02:14,960 Speaker 5: So I mean this this is a really good one 43 00:02:14,960 --> 00:02:18,880 Speaker 5: to dig into. I think because last week Hugging Face, 44 00:02:18,919 --> 00:02:20,919 Speaker 5: which is kind of like get hub but for AI, 45 00:02:21,080 --> 00:02:24,600 Speaker 5: like they have data sets, models, it's very open source, 46 00:02:25,639 --> 00:02:28,240 Speaker 5: funded by all the usual vcs. They put up this 47 00:02:28,320 --> 00:02:32,080 Speaker 5: blog post that they had this unprecedented security incident that 48 00:02:32,160 --> 00:02:36,160 Speaker 5: for the first time, an AI agent had broken into 49 00:02:36,160 --> 00:02:39,079 Speaker 5: their system and tried all of these methods to get 50 00:02:39,160 --> 00:02:41,720 Speaker 5: access to some data sets, and it did it by 51 00:02:41,960 --> 00:02:45,680 Speaker 5: uploading like a malicious data set that would allow it 52 00:02:45,720 --> 00:02:48,040 Speaker 5: to kind of like get into the system through their 53 00:02:48,120 --> 00:02:49,920 Speaker 5: data processing pipeline and then. 54 00:02:50,080 --> 00:02:51,440 Speaker 1: Like a trojan horses. 55 00:02:51,480 --> 00:02:56,280 Speaker 5: It were yeah, yeah, And at the time, Hugging Face said, 56 00:02:56,400 --> 00:02:58,400 Speaker 5: you know, we think that this looks like an agent 57 00:02:58,440 --> 00:03:00,840 Speaker 5: that's used for security and RESO, but we don't know 58 00:03:00,880 --> 00:03:03,600 Speaker 5: what LLLM it is. We don't know what model. So people, 59 00:03:03,960 --> 00:03:05,960 Speaker 5: you know, obviously this is coming in the middle of 60 00:03:06,000 --> 00:03:09,720 Speaker 5: all of this talk about Chinese open source models, how 61 00:03:09,720 --> 00:03:13,400 Speaker 5: they're proliferating, how they're getting more capable, so nobody knows, 62 00:03:13,520 --> 00:03:17,360 Speaker 5: you know, what's happening. Then this week on Tuesday, open 63 00:03:17,400 --> 00:03:20,399 Speaker 5: Ai puts out a blog post and they're like, oh, hey, 64 00:03:20,440 --> 00:03:23,720 Speaker 5: that was us, and they said, you know, they discovered 65 00:03:23,760 --> 00:03:29,040 Speaker 5: after the fact that while they were testing both GPT 66 00:03:30,120 --> 00:03:33,400 Speaker 5: five point six Soul and some new unreleased model, they 67 00:03:33,400 --> 00:03:35,760 Speaker 5: were testing it on its ability to do these like 68 00:03:35,840 --> 00:03:40,640 Speaker 5: cyber offensive attacks, and they gave it this benchmark called 69 00:03:40,760 --> 00:03:43,960 Speaker 5: exploit benchmark that's supposed to test its ability to like 70 00:03:44,040 --> 00:03:48,400 Speaker 5: break into, you know, to hack into systems. The AI agent, 71 00:03:48,480 --> 00:03:51,560 Speaker 5: which was allowed to work you know, autonomously for a while, 72 00:03:52,800 --> 00:03:56,600 Speaker 5: decided to cheat on the test and figured a way 73 00:03:56,760 --> 00:03:59,920 Speaker 5: out of the sandbox that open ai created, like based 74 00:04:00,280 --> 00:04:04,040 Speaker 5: you know, testing environment, got into the Internet, decided to 75 00:04:04,120 --> 00:04:08,840 Speaker 5: break into hugging Face. And open ai is so lucky. 76 00:04:08,880 --> 00:04:12,680 Speaker 5: They are so lucky that it was another AI company 77 00:04:12,680 --> 00:04:16,080 Speaker 5: that has invested in in the industry and the narrative 78 00:04:16,200 --> 00:04:18,920 Speaker 5: because this is I mean, even the the CEO of 79 00:04:19,000 --> 00:04:21,240 Speaker 5: hugging Face was like, yeah, don't do this to us. 80 00:04:21,360 --> 00:04:24,839 Speaker 5: It's illegal. You know, they had notified law enforcement. And 81 00:04:24,880 --> 00:04:27,520 Speaker 5: now the announcement that it was open ai came as 82 00:04:27,560 --> 00:04:31,040 Speaker 5: part of a partnership between hugging Face and open Ai, 83 00:04:31,440 --> 00:04:34,599 Speaker 5: which was obviously you know, put together after the fact. 84 00:04:34,920 --> 00:04:37,640 Speaker 2: So the model had essentially determined that it would be 85 00:04:37,680 --> 00:04:40,119 Speaker 2: easier to go and steal the answer from hugging Face 86 00:04:40,160 --> 00:04:43,320 Speaker 2: than to come up with it itself, and therefore did that. 87 00:04:44,279 --> 00:04:47,440 Speaker 5: Yeah, it knew that it was being tested on a benchmark. 88 00:04:48,080 --> 00:04:51,280 Speaker 5: I think they even maybe you know, uploaded the benchmark 89 00:04:51,320 --> 00:04:54,480 Speaker 5: from hugging Face and it figured out it was supposed 90 00:04:54,520 --> 00:04:56,440 Speaker 5: to not be able to access the Internet. 91 00:04:58,160 --> 00:05:00,440 Speaker 2: But that's a little I mean that That's the part 92 00:05:00,440 --> 00:05:02,720 Speaker 2: why Zone zoned in on the word catastrophe, because the 93 00:05:02,760 --> 00:05:04,360 Speaker 2: bit where is not supposed to be able to access 94 00:05:04,400 --> 00:05:06,400 Speaker 2: the Internet where where, but it does. I mean that's 95 00:05:06,480 --> 00:05:09,640 Speaker 2: got a kind of cinematically spooky quality. 96 00:05:10,000 --> 00:05:14,000 Speaker 4: Well, we failed to secure the sandbox, right, Yeah, exactly. 97 00:05:13,760 --> 00:05:17,279 Speaker 2: Failed to secure the soundbox. Premagine preschool is hearing that 98 00:05:17,320 --> 00:05:18,040 Speaker 2: in the nineties. 99 00:05:18,400 --> 00:05:19,040 Speaker 3: What are you doing? 100 00:05:22,880 --> 00:05:27,840 Speaker 5: Yeah, you zeroed in on exactly the right part of 101 00:05:27,920 --> 00:05:30,680 Speaker 5: the confusion around this and why people are using it 102 00:05:30,720 --> 00:05:34,480 Speaker 5: to like support you know, various narratives. You know, people 103 00:05:34,520 --> 00:05:38,760 Speaker 5: who are very worried about AI super intelligence and AI 104 00:05:38,839 --> 00:05:42,600 Speaker 5: capabilities growing really fast. You know, we're talking about this 105 00:05:42,760 --> 00:05:44,800 Speaker 5: as like this is what we've been warning you about. 106 00:05:44,880 --> 00:05:47,599 Speaker 5: It's a rogue AI. You know, we gave it a 107 00:05:47,680 --> 00:05:50,880 Speaker 5: simple set of instructions and it decided to do it 108 00:05:50,920 --> 00:05:53,200 Speaker 5: in a way that you know, was not was not 109 00:05:53,320 --> 00:05:57,680 Speaker 5: what its developers would want. But many other cybersecurity professionals 110 00:05:57,680 --> 00:06:02,160 Speaker 5: were like, why did you not have a properly configured sandbox, 111 00:06:02,240 --> 00:06:05,240 Speaker 5: Like you could be using an air gap, which would 112 00:06:05,279 --> 00:06:08,560 Speaker 5: not have allowed it to break into an Internet system. 113 00:06:08,960 --> 00:06:13,240 Speaker 5: They allowed it to use like package installs, and I 114 00:06:13,279 --> 00:06:15,960 Speaker 5: think just a lot of cybersecurity professionals found this like 115 00:06:16,080 --> 00:06:18,200 Speaker 5: very sloppy. This is not you know, so it's not 116 00:06:18,600 --> 00:06:20,679 Speaker 5: it's not as though, oh my god, we can't contain 117 00:06:20,720 --> 00:06:23,599 Speaker 5: AI and it's doing some you know, like you know, 118 00:06:23,680 --> 00:06:26,400 Speaker 5: people were throwing around the word sentient machines and that's 119 00:06:26,640 --> 00:06:27,440 Speaker 5: not what this is. 120 00:06:27,520 --> 00:06:30,640 Speaker 3: Well, on top of that that the model had no safeguards, right, 121 00:06:30,680 --> 00:06:33,000 Speaker 3: there were no you know, it was a really raw, 122 00:06:33,400 --> 00:06:35,040 Speaker 3: you know model, So. 123 00:06:35,560 --> 00:06:39,680 Speaker 5: Yeah, they were testing its ability to to do these 124 00:06:39,760 --> 00:06:42,159 Speaker 5: kind of exploits, so they didn't have any safeguards on. 125 00:06:42,240 --> 00:06:46,720 Speaker 5: But then they just didn't have like traditional cybersecurity controls 126 00:06:47,040 --> 00:06:48,919 Speaker 5: that you would have in place when you're testing. And 127 00:06:48,960 --> 00:06:52,039 Speaker 5: this is something people had been warning against that maybe 128 00:06:52,040 --> 00:06:55,360 Speaker 5: the biggest danger is not when your systems are in deployment, 129 00:06:55,400 --> 00:06:57,920 Speaker 5: because then you have all your safeguards up. It's refusing 130 00:06:57,920 --> 00:07:00,720 Speaker 5: a lot of requests, but it's actually during this kind 131 00:07:00,760 --> 00:07:03,600 Speaker 5: of testing environment. And I'll just say one more thing. 132 00:07:03,720 --> 00:07:08,440 Speaker 5: This incident followed another blog post last week from open 133 00:07:08,480 --> 00:07:11,960 Speaker 5: ai about how they had to stop even just testing 134 00:07:12,040 --> 00:07:15,760 Speaker 5: an internal model because it like they weren't monitoring it 135 00:07:15,800 --> 00:07:18,280 Speaker 5: and it started doing all of these things because they 136 00:07:18,400 --> 00:07:20,520 Speaker 5: let it go on and on, because that's you know, 137 00:07:20,560 --> 00:07:23,080 Speaker 5: that's how these models get more functional. They're able to 138 00:07:23,160 --> 00:07:26,440 Speaker 5: like think and try thousands and thousands of different ways 139 00:07:26,480 --> 00:07:27,560 Speaker 5: of breaking into things. 140 00:07:27,760 --> 00:07:30,360 Speaker 3: Well, I think that I like agree with you, Natasha 141 00:07:30,400 --> 00:07:33,120 Speaker 3: that it's not this isn't like some you know, code 142 00:07:33,200 --> 00:07:37,600 Speaker 3: read catastrophe at all. But I do think there's something 143 00:07:37,640 --> 00:07:40,520 Speaker 3: here that is sort of like you know this, The 144 00:07:40,640 --> 00:07:43,080 Speaker 3: AI security people do have a point on this. It's 145 00:07:43,080 --> 00:07:45,080 Speaker 3: a point they've been making for a long time, which 146 00:07:45,120 --> 00:07:48,360 Speaker 3: is that like, when you tell an AI model to 147 00:07:48,480 --> 00:07:52,760 Speaker 3: do something, it will, like the smarter they are, the 148 00:07:52,800 --> 00:07:55,520 Speaker 3: more they're just going to find some shortcut to do that. Right, 149 00:07:55,600 --> 00:07:57,480 Speaker 3: this is the whole argument. But this is the argument 150 00:07:57,560 --> 00:07:59,880 Speaker 3: with like the you know, the paper clip argument, Right, 151 00:08:00,080 --> 00:08:02,000 Speaker 3: if you tell AI to make paper clips, it will 152 00:08:02,040 --> 00:08:04,680 Speaker 3: just ultimately like find the best way to do that, 153 00:08:04,760 --> 00:08:07,239 Speaker 3: even if that means like turning humans into paper clips. 154 00:08:07,280 --> 00:08:09,920 Speaker 4: So but also but also read, I think that that 155 00:08:10,040 --> 00:08:12,960 Speaker 4: is a is a sort of ridiculous concept because any 156 00:08:13,000 --> 00:08:16,880 Speaker 4: AI that's smart enough in that way would also have 157 00:08:16,960 --> 00:08:20,360 Speaker 4: the reasoning to know, like they're not going to turn 158 00:08:20,360 --> 00:08:22,320 Speaker 4: the world into paper clips. That's not the goal. Like, 159 00:08:22,480 --> 00:08:24,560 Speaker 4: I don't know, I think a lot of this like 160 00:08:24,640 --> 00:08:26,240 Speaker 4: kind of like what Natasha was saying before is like 161 00:08:26,240 --> 00:08:28,800 Speaker 4: a little it's being taken by these people. It's like 162 00:08:29,000 --> 00:08:32,840 Speaker 4: they made a sloppy security environment, the AI went and 163 00:08:32,880 --> 00:08:36,320 Speaker 4: solved the problem as it should have or whatever, And 164 00:08:36,320 --> 00:08:39,520 Speaker 4: and that's not evidence of it being super sentient or 165 00:08:39,520 --> 00:08:43,640 Speaker 4: even this being a particularly good AI agent. Mostly it's 166 00:08:43,679 --> 00:08:46,880 Speaker 4: just that like they didn't set up this test very responsibly. 167 00:08:47,520 --> 00:08:49,920 Speaker 3: Definitely agree, I agree with but again, like I'm not 168 00:08:49,920 --> 00:08:51,680 Speaker 3: disagreeing with you at all. I just think there is 169 00:08:51,720 --> 00:08:54,040 Speaker 3: one sliver of this that is worth sort of looking at, 170 00:08:54,040 --> 00:08:56,400 Speaker 3: which is like, you know, they do sort of find 171 00:08:56,440 --> 00:08:59,360 Speaker 3: these shortcuts, so and in this case, the shortcut meant 172 00:08:59,360 --> 00:09:04,480 Speaker 3: like breaking the law and hacking into hugging face, so 173 00:09:04,600 --> 00:09:06,880 Speaker 3: I and I think the other thing to remember is 174 00:09:06,880 --> 00:09:10,120 Speaker 3: that like we still don't know how these models work, 175 00:09:10,480 --> 00:09:12,280 Speaker 3: Like no one has been able to figure out like 176 00:09:12,360 --> 00:09:16,280 Speaker 3: what are these neurons inside these models actually doing? And 177 00:09:16,640 --> 00:09:19,320 Speaker 3: I do I think the big question is like, as 178 00:09:19,360 --> 00:09:23,640 Speaker 3: they get more and more capable and more powerful, is 179 00:09:23,679 --> 00:09:26,080 Speaker 3: it is it enough to just put you know, the 180 00:09:26,120 --> 00:09:29,200 Speaker 3: safeguards in place, which again we're not like they didn't 181 00:09:29,240 --> 00:09:31,959 Speaker 3: what I'm curious about because this this this Natasha, you 182 00:09:32,040 --> 00:09:34,800 Speaker 3: mentioned there were two open Ami models of working on this. 183 00:09:34,840 --> 00:09:37,400 Speaker 3: One was a publicly available one and one has not 184 00:09:37,480 --> 00:09:40,760 Speaker 3: yet been released. So could could I Well, probably not me, 185 00:09:40,840 --> 00:09:42,679 Speaker 3: but like could somebody has a bit more sophisticated than 186 00:09:42,720 --> 00:09:45,960 Speaker 3: me use open Ami to hugging face exactly like this 187 00:09:46,040 --> 00:09:49,520 Speaker 3: went down like using the publicly available models. Or was 188 00:09:49,559 --> 00:09:53,000 Speaker 3: there something special about this unreleased model? Well, no, because 189 00:09:53,000 --> 00:09:56,000 Speaker 3: there were no safeguards like the publicly released one will 190 00:09:56,040 --> 00:09:58,800 Speaker 3: have will have controls on it that will not let 191 00:09:58,880 --> 00:10:01,280 Speaker 3: you do this unless you can jail break it, which 192 00:10:01,559 --> 00:10:04,319 Speaker 3: you know is getting more difficult to do, right. 193 00:10:04,800 --> 00:10:06,959 Speaker 5: Okay, I will say, I mean, I do think it's 194 00:10:07,080 --> 00:10:09,640 Speaker 5: important to note that this is like many people are 195 00:10:09,679 --> 00:10:12,480 Speaker 5: interpreting this as like we were right about the paper 196 00:10:12,480 --> 00:10:15,000 Speaker 5: clip maximizer. You know, this is happening in this way. 197 00:10:15,040 --> 00:10:18,040 Speaker 5: But I think there's a way that because they conceived 198 00:10:18,080 --> 00:10:21,120 Speaker 5: of it, of the problem this way, they approached security 199 00:10:21,240 --> 00:10:24,120 Speaker 5: in a certain way, like you could have been thinking 200 00:10:24,360 --> 00:10:27,760 Speaker 5: like a cybersecurity professional the whole time and not thinking 201 00:10:27,840 --> 00:10:32,920 Speaker 5: about AI alignment, aligning it with with human values. And 202 00:10:33,080 --> 00:10:35,400 Speaker 5: one of the things I mentioned that like prior open 203 00:10:35,440 --> 00:10:38,880 Speaker 5: ai blog post. What they realized is that they didn't 204 00:10:38,880 --> 00:10:42,120 Speaker 5: have sufficient monitoring systems. When you let a system go 205 00:10:42,280 --> 00:10:44,800 Speaker 5: on and on for a really long time, they weren't 206 00:10:45,160 --> 00:10:49,640 Speaker 5: watching it closely, like there are simple mechanisms, you know. 207 00:10:49,880 --> 00:10:50,600 Speaker 4: It doesn't even go. 208 00:10:50,640 --> 00:10:53,439 Speaker 5: Back to what Reid said about knowing what's happening on 209 00:10:53,480 --> 00:10:56,280 Speaker 5: the neurons, Like they weren't even watching what it was 210 00:10:56,320 --> 00:10:59,160 Speaker 5: doing during the security test in a secure way. 211 00:10:59,400 --> 00:11:02,960 Speaker 2: Ax out there saying that saying maybe open Aye wanted 212 00:11:03,040 --> 00:11:05,400 Speaker 2: to demonstrate anything mythos can do, we can do better, 213 00:11:05,520 --> 00:11:08,120 Speaker 2: or is that it's great marketing for opening I. 214 00:11:08,640 --> 00:11:10,480 Speaker 5: For sure, I think the way that they wrote the 215 00:11:10,520 --> 00:11:12,719 Speaker 5: blog post at the top, it says like we are 216 00:11:12,760 --> 00:11:17,679 Speaker 5: treating this as an unprecedented cybersecurity incident that shows the capabilities. 217 00:11:17,720 --> 00:11:20,480 Speaker 5: And I will say, like, I don't think anyone's arguing 218 00:11:21,000 --> 00:11:25,440 Speaker 5: that these lms make people much much much better at 219 00:11:25,520 --> 00:11:30,400 Speaker 5: hacking and that it's capable of improving and enhancing and trying, 220 00:11:31,400 --> 00:11:34,960 Speaker 5: you know, cyber offensives that humans can't do. But everyone 221 00:11:35,040 --> 00:11:37,120 Speaker 5: I talked to was like, this is not an example 222 00:11:37,160 --> 00:11:40,920 Speaker 5: of enhanced capabilities. This is an example of like kind 223 00:11:40,920 --> 00:11:42,200 Speaker 5: of a sloppy testing environment. 224 00:11:42,280 --> 00:11:43,880 Speaker 3: I mean, we all agree on that. I think it's 225 00:11:43,920 --> 00:11:47,040 Speaker 3: just it just sort of highlights this question about the future, 226 00:11:47,080 --> 00:11:49,280 Speaker 3: which is like, when these models become more powerful, do 227 00:11:49,320 --> 00:11:52,160 Speaker 3: we need do we actually have to understand how they work? 228 00:11:52,559 --> 00:11:56,560 Speaker 2: Has this has this effected the kind of policy discussions 229 00:11:56,559 --> 00:11:59,280 Speaker 2: in Washington? What's how's the White House and crapsios and 230 00:11:59,320 --> 00:12:02,480 Speaker 2: others around around Trump and the sort of technology advisory 231 00:12:03,040 --> 00:12:05,319 Speaker 2: space reacted to this? Is this is this like grist 232 00:12:05,360 --> 00:12:07,840 Speaker 2: to the mill of the of the deceleration lists or 233 00:12:07,920 --> 00:12:08,160 Speaker 2: is it? 234 00:12:09,080 --> 00:12:12,240 Speaker 5: I would say they were very busy with They're still 235 00:12:12,320 --> 00:12:14,200 Speaker 5: very busy with Chinese open source models. 236 00:12:14,480 --> 00:12:15,480 Speaker 4: So that was the focus. 237 00:12:15,559 --> 00:12:18,079 Speaker 5: This didn't I don't think it got the attention of 238 00:12:18,600 --> 00:12:21,680 Speaker 5: you know, the the same White House folks they were 239 00:12:21,800 --> 00:12:26,880 Speaker 5: putting out. They were putting out information about distillation, which 240 00:12:26,920 --> 00:12:30,000 Speaker 5: we talked about last week, and you know, did did 241 00:12:30,400 --> 00:12:33,800 Speaker 5: Chinese open source models steal from quote unquote steal from 242 00:12:34,000 --> 00:12:38,960 Speaker 5: from anthropic? But there were a number of regulators that said, like, 243 00:12:39,040 --> 00:12:42,920 Speaker 5: it's time to legislate, it's time to get involved, Bernie 244 00:12:42,960 --> 00:12:48,000 Speaker 5: Sanders and many others. So yeah, it's feeding into this 245 00:12:48,760 --> 00:12:53,240 Speaker 5: growing sense of urgency around having some safeguards in place, 246 00:12:53,280 --> 00:12:57,520 Speaker 5: But I just want to say, like another plea for 247 00:12:57,600 --> 00:13:01,680 Speaker 5: people to start thinking about monitoring like downstream usage, like 248 00:13:01,800 --> 00:13:03,920 Speaker 5: kind of what reads saying about the neurons. Actually, now 249 00:13:03,920 --> 00:13:06,160 Speaker 5: that I go back to it, because what they do 250 00:13:06,200 --> 00:13:10,119 Speaker 5: when they try to figure out the models like reasoning 251 00:13:10,440 --> 00:13:13,079 Speaker 5: is look at its chain of thought, like look at 252 00:13:13,240 --> 00:13:15,480 Speaker 5: this little like they give the model a scratch pad 253 00:13:15,520 --> 00:13:17,760 Speaker 5: and it says like, ah, it might be faster to 254 00:13:17,840 --> 00:13:20,040 Speaker 5: go to directly to hooking face, let me break in 255 00:13:20,120 --> 00:13:23,439 Speaker 5: or whatever, and we don't even know if those that 256 00:13:23,600 --> 00:13:29,320 Speaker 5: text actually reflects its interior thought. You know, it's like 257 00:13:29,440 --> 00:13:31,560 Speaker 5: decision making process. So it's true, this. 258 00:13:31,640 --> 00:13:33,080 Speaker 2: Is the dome worry mom and dad, I'm doing my 259 00:13:33,080 --> 00:13:36,920 Speaker 2: homework upstairs. Basically, of AI, it's thinking out loud. 260 00:13:37,160 --> 00:13:39,400 Speaker 3: It's thinking out loud, and that is actually this is 261 00:13:39,440 --> 00:13:41,400 Speaker 3: now this, this is actually the state of the art 262 00:13:41,440 --> 00:13:44,120 Speaker 3: in like AI safety is like looking at it's thinking 263 00:13:44,160 --> 00:13:44,920 Speaker 3: out loud. 264 00:13:45,080 --> 00:13:48,400 Speaker 2: But but it can think out loud deceptively potentially yes. 265 00:13:48,440 --> 00:13:50,840 Speaker 3: And also you're not looking at it's thinking inside. It's 266 00:13:50,880 --> 00:13:54,080 Speaker 3: like if you if you only base what somebody's thoughts 267 00:13:54,120 --> 00:13:57,280 Speaker 3: are and what they say, then that's you're clearly not 268 00:13:57,360 --> 00:13:59,280 Speaker 3: getting all their thoughts right and open. 269 00:13:59,400 --> 00:14:02,600 Speaker 5: I wasn't even reading that you know, so so. 270 00:14:02,960 --> 00:14:04,400 Speaker 3: Right for this experiment. 271 00:14:04,440 --> 00:14:07,200 Speaker 5: They were let's let's yeah, let's let's clean it up. 272 00:14:08,120 --> 00:14:10,880 Speaker 2: So here's my favorite detail from this whole story, which 273 00:14:10,920 --> 00:14:14,200 Speaker 2: actually comes from Forbes, of all places. Quote Hugging Face 274 00:14:14,200 --> 00:14:16,760 Speaker 2: said it had first attempted to use an undisclosed AI 275 00:14:16,840 --> 00:14:20,120 Speaker 2: model from leading US Labs to defend against the attacking 276 00:14:20,160 --> 00:14:24,000 Speaker 2: AI agent, but the guardrails around that model cyber capabilities 277 00:14:24,160 --> 00:14:27,480 Speaker 2: stemied its response teams work. The company said it instead 278 00:14:27,480 --> 00:14:30,360 Speaker 2: wound up using an open source AI model from Chinese 279 00:14:30,360 --> 00:14:33,040 Speaker 2: company Zai to carry out its defense. 280 00:14:33,400 --> 00:14:38,760 Speaker 4: Which is I think exactly why these type of safeguards 281 00:14:38,840 --> 00:14:41,000 Speaker 4: that we're sort of putting in our American models are 282 00:14:41,000 --> 00:14:45,520 Speaker 4: deranged and we need access to open source Chinese models. 283 00:14:45,680 --> 00:14:49,240 Speaker 4: I mean, I think even just from a security standpoint. 284 00:14:48,840 --> 00:14:52,840 Speaker 5: Right, I mean, many cybersecurity people were saying, you know, 285 00:14:52,920 --> 00:14:56,520 Speaker 5: this shows the need to like have policies that help 286 00:14:56,600 --> 00:15:01,440 Speaker 5: proliferation of open source models, because as like, history has 287 00:15:01,480 --> 00:15:06,040 Speaker 5: shown that cyber defense capabilities rely on open source models, 288 00:15:06,080 --> 00:15:07,960 Speaker 5: So it doesn't need to be the Chinese if we 289 00:15:08,080 --> 00:15:10,840 Speaker 5: had been investing in open source AI or if our 290 00:15:10,960 --> 00:15:13,560 Speaker 5: leading companies were putting out open source AI models, it 291 00:15:13,640 --> 00:15:16,640 Speaker 5: could be us. But they did talk about the need 292 00:15:16,680 --> 00:15:20,760 Speaker 5: to make sure like hospitals are adopting the latest open 293 00:15:20,800 --> 00:15:23,600 Speaker 5: source AI models in order to make sure they're capable 294 00:15:23,600 --> 00:15:25,840 Speaker 5: of defending if such a thing happens to them. 295 00:15:26,240 --> 00:15:28,840 Speaker 2: Okay, lightning round, before we go to the break, what's 296 00:15:28,880 --> 00:15:31,320 Speaker 2: the one thing from each of you that somebody's listening 297 00:15:31,360 --> 00:15:34,080 Speaker 2: to this podcast needs to know to sound smart. 298 00:15:33,880 --> 00:15:34,360 Speaker 1: At the weekend? 299 00:15:34,440 --> 00:15:37,320 Speaker 4: About Kimmy K three, I mean, I think everybody wants 300 00:15:37,360 --> 00:15:40,160 Speaker 4: to act like, oh, it was distilled. It was definitely distilled. 301 00:15:40,200 --> 00:15:42,640 Speaker 4: I mean, this is a model that I believe was 302 00:15:42,680 --> 00:15:46,240 Speaker 4: testing and also if I might be getting this wrong, 303 00:15:46,280 --> 00:15:49,520 Speaker 4: but I also think that it's this model also has 304 00:15:49,560 --> 00:15:52,760 Speaker 4: capabilities that our American models don't have. Or it is. 305 00:15:53,000 --> 00:15:55,240 Speaker 1: Kimmykre right, Yeah? Which is what? 306 00:15:55,360 --> 00:15:58,600 Speaker 2: Which is a new model from a Chinese lab that's 307 00:15:58,680 --> 00:16:02,840 Speaker 2: kind of outperform throwed on various benchmarks and therefore everyone 308 00:16:02,920 --> 00:16:03,760 Speaker 2: kind of lost their mind? 309 00:16:03,800 --> 00:16:04,760 Speaker 1: Or what's the what's the like? 310 00:16:04,880 --> 00:16:07,840 Speaker 3: It's sort of on par with with the with the Frontier. 311 00:16:08,160 --> 00:16:10,280 Speaker 3: I wouldn't say it's outperforming, and I don't know if 312 00:16:10,320 --> 00:16:13,200 Speaker 3: I do, you mean it has capabilities the US models 313 00:16:13,240 --> 00:16:15,880 Speaker 3: don't have, or they use training techniques that are sort 314 00:16:15,880 --> 00:16:18,680 Speaker 3: of novel and that you know, the US labs could 315 00:16:18,720 --> 00:16:19,240 Speaker 3: learn from. 316 00:16:19,480 --> 00:16:21,880 Speaker 4: Basically, Like what I was reading yesterday, I was saying 317 00:16:21,920 --> 00:16:26,480 Speaker 4: that like the architecture of the model is superior to 318 00:16:26,480 --> 00:16:29,400 Speaker 4: to whatever it was claiming to be distilled on, and 319 00:16:29,440 --> 00:16:31,920 Speaker 4: that these are not developments that they could have stolen 320 00:16:32,120 --> 00:16:36,880 Speaker 4: basically because it is they have developed further on you 321 00:16:36,880 --> 00:16:39,160 Speaker 4: know whatever. I guess it's like the latest sort of 322 00:16:39,160 --> 00:16:42,800 Speaker 4: Claude model. So I just I think this idea that like, oh, 323 00:16:42,840 --> 00:16:45,640 Speaker 4: these are all just like t MoU versions of Claude. 324 00:16:45,680 --> 00:16:49,200 Speaker 4: I don't think that that's a full picture. Yes, I imagine 325 00:16:49,120 --> 00:16:52,160 Speaker 4: that they're doing some level of distilling. They all do. Like, 326 00:16:52,320 --> 00:16:55,000 Speaker 4: I just think this is like an industry. 327 00:16:55,280 --> 00:16:57,800 Speaker 3: No, there's an innovation there. Like I think what you're 328 00:16:57,800 --> 00:17:01,040 Speaker 3: saying is like you can distill from one of these 329 00:17:01,160 --> 00:17:03,760 Speaker 3: large models in order to shortcut the training, but there's 330 00:17:03,800 --> 00:17:08,080 Speaker 3: still innovation in the training in being just because by 331 00:17:08,200 --> 00:17:11,320 Speaker 3: virtue of being hamstrung on the amount of compute they have, 332 00:17:11,520 --> 00:17:14,080 Speaker 3: they have to they have to take these shortcuts and 333 00:17:14,119 --> 00:17:17,000 Speaker 3: they have to find new and novel ways to train 334 00:17:17,080 --> 00:17:20,680 Speaker 3: with less compute, and so there's innovation there for sure. 335 00:17:25,920 --> 00:17:28,680 Speaker 2: When we come back, read takes us to a garage 336 00:17:28,840 --> 00:17:43,639 Speaker 2: full of automated pickup trucks carrying anti drone weaponry. 337 00:17:48,800 --> 00:17:49,440 Speaker 1: Welcome back. 338 00:17:49,920 --> 00:17:52,320 Speaker 2: So I mentioned in the first story that Nick Thompson 339 00:17:52,640 --> 00:17:54,720 Speaker 2: predicted this would be the year of the first legit 340 00:17:54,840 --> 00:17:58,240 Speaker 2: ai catastrophe. Our panelisting that that that prediction has not 341 00:17:58,320 --> 00:17:59,600 Speaker 2: yet been proven out. 342 00:18:00,119 --> 00:18:03,520 Speaker 1: My prediction was was the year of the robot Read. 343 00:18:03,880 --> 00:18:06,080 Speaker 2: You had a story this week about a company that, 344 00:18:06,080 --> 00:18:09,560 Speaker 2: according to the headline semiphore quote, wants to turn robotics 345 00:18:09,600 --> 00:18:10,800 Speaker 2: into child's play. 346 00:18:11,359 --> 00:18:13,760 Speaker 1: What's the company and what's it do today? And what 347 00:18:13,800 --> 00:18:15,760 Speaker 1: does turning robotics into child's playing? 348 00:18:16,320 --> 00:18:20,600 Speaker 3: Yeah, so the companies applied intuition and I went visited 349 00:18:20,600 --> 00:18:23,920 Speaker 3: their their headquarters on Monday and took a tour through. 350 00:18:24,520 --> 00:18:26,600 Speaker 3: You know, they have this garage where there's really just 351 00:18:26,640 --> 00:18:30,399 Speaker 3: a bunch of cars in you know, various states of disrepair, 352 00:18:30,560 --> 00:18:35,000 Speaker 3: so like you just see all the electronics around a car. Essentially, 353 00:18:35,280 --> 00:18:39,240 Speaker 3: what they've been doing is working for a lot of automakers, 354 00:18:39,280 --> 00:18:42,600 Speaker 3: most of the big automakers actually and building the technology 355 00:18:42,640 --> 00:18:47,080 Speaker 3: for them everything from like the software platform, infotainment systems, 356 00:18:47,080 --> 00:18:49,840 Speaker 3: et cetera, to the self driving tech and they've been 357 00:18:49,840 --> 00:18:51,960 Speaker 3: making a lot of money doing that. They've now expanded 358 00:18:52,000 --> 00:18:55,760 Speaker 3: into like farming and other you know, construction, other areas. 359 00:18:56,359 --> 00:18:59,560 Speaker 2: So basically, if you're not Tesla, your license applied intuition 360 00:19:00,000 --> 00:19:00,800 Speaker 2: evan technology. 361 00:19:01,119 --> 00:19:04,160 Speaker 3: Yeah, exactly. Like if you're an automaker and you realize 362 00:19:04,160 --> 00:19:06,240 Speaker 3: you don't have the tech talent that Tesla has, but 363 00:19:06,280 --> 00:19:09,120 Speaker 3: you want to have a Tesla like product, you kind 364 00:19:09,160 --> 00:19:12,439 Speaker 3: of you hire this this company rather than try to 365 00:19:12,480 --> 00:19:14,800 Speaker 3: do it in house. But and that's been a good 366 00:19:14,800 --> 00:19:18,240 Speaker 3: business for them. But what the news was on Monday, 367 00:19:18,359 --> 00:19:20,560 Speaker 3: I wrote I wrote about it on Tuesday morning was 368 00:19:20,640 --> 00:19:24,720 Speaker 3: that they've come out with this new open platform that 369 00:19:24,760 --> 00:19:26,280 Speaker 3: you can you know, it's a product that you can 370 00:19:26,280 --> 00:19:28,680 Speaker 3: pay for and you can actually use all of their 371 00:19:28,760 --> 00:19:32,639 Speaker 3: data and all the infrastructure they built over almost a 372 00:19:32,720 --> 00:19:36,560 Speaker 3: decade and train your own robots. Essentially. They would call 373 00:19:36,600 --> 00:19:39,800 Speaker 3: it physical intelligence, which is this jargon term for like 374 00:19:39,880 --> 00:19:43,280 Speaker 3: any machine that can be programmed essentially. But like their 375 00:19:43,359 --> 00:19:46,440 Speaker 3: example of is, like you have a kid who wants 376 00:19:46,480 --> 00:19:50,480 Speaker 3: to make an autonomous lawnmower, they could do it now, 377 00:19:50,520 --> 00:19:52,560 Speaker 3: whereas it used to take a team of like eight, 378 00:19:53,080 --> 00:19:56,399 Speaker 3: you know, PhD engineers to do all the computer vision 379 00:19:56,520 --> 00:19:59,439 Speaker 3: and train models, et cetera. And I played around with it. 380 00:19:59,480 --> 00:20:01,639 Speaker 3: I mean, it's still I would say, this is not 381 00:20:01,800 --> 00:20:05,639 Speaker 3: child's play yet, but it's moving in that direction. And 382 00:20:05,800 --> 00:20:07,720 Speaker 3: I just think it's really interesting because it's like, I 383 00:20:07,720 --> 00:20:09,280 Speaker 3: don't know about you guys. I play around with like 384 00:20:09,359 --> 00:20:11,840 Speaker 3: Raspberry pies with my with my ten year old, you know, 385 00:20:11,880 --> 00:20:14,400 Speaker 3: and we try to hack stuff together, and I'm like, oh, 386 00:20:14,440 --> 00:20:17,080 Speaker 3: this is cool. Like this allows to do even more 387 00:20:17,080 --> 00:20:19,720 Speaker 3: fun projects. And then I think sort of create more 388 00:20:19,880 --> 00:20:23,119 Speaker 3: entrepreneurship around AI for the physical world. 389 00:20:23,560 --> 00:20:25,800 Speaker 2: Okay, but so I could go and buy a regular 390 00:20:26,119 --> 00:20:29,760 Speaker 2: NORM from home depot and then use this platform to 391 00:20:29,800 --> 00:20:30,680 Speaker 2: turn into a robot. 392 00:20:31,280 --> 00:20:33,159 Speaker 3: Yeah, well you could put like a GoPro on it 393 00:20:33,280 --> 00:20:37,359 Speaker 3: or something, right, and go film film yourself mowing your 394 00:20:37,440 --> 00:20:41,679 Speaker 3: lawn and then upload the video into this platform and say, okay, 395 00:20:41,720 --> 00:20:44,600 Speaker 3: I want you to basically do like an evaluation and 396 00:20:44,640 --> 00:20:47,679 Speaker 3: train an AI model that is the perfect model for 397 00:20:47,800 --> 00:20:51,080 Speaker 3: mowing my lawn, you know, and be done in a day. 398 00:20:51,400 --> 00:20:57,120 Speaker 5: Essentially, you provide the hardware and they provide the intellig This. 399 00:20:57,200 --> 00:20:59,760 Speaker 3: Is just a software platform. There's no they don't do 400 00:21:00,040 --> 00:21:02,720 Speaker 3: they don't do any hardware, so you make the hardware 401 00:21:03,400 --> 00:21:06,320 Speaker 3: well or like it's really kind of right now aimed 402 00:21:06,320 --> 00:21:08,640 Speaker 3: at like startups, right like if you were it would 403 00:21:08,680 --> 00:21:10,960 Speaker 3: be like a more like a company probably trying to 404 00:21:10,960 --> 00:21:14,240 Speaker 3: build a n autonomous lawnmower at this point. Like it's not 405 00:21:14,320 --> 00:21:16,720 Speaker 3: at the point where you know, but. 406 00:21:16,200 --> 00:21:19,399 Speaker 5: But it's only for hardware, Like it's not. It's not 407 00:21:19,520 --> 00:21:23,320 Speaker 5: like you wouldn't use this just as an online model. 408 00:21:23,359 --> 00:21:26,439 Speaker 3: It's just a software platform that you can train like 409 00:21:26,640 --> 00:21:29,320 Speaker 3: computer vision models. Look, I'll tell you what I want 410 00:21:29,320 --> 00:21:31,119 Speaker 3: to do with it. Do you want? This is embarrassing, 411 00:21:31,200 --> 00:21:33,520 Speaker 3: but yes, but I didn't put I didn't put this 412 00:21:33,600 --> 00:21:36,240 Speaker 3: in the article because I was worried about animal rights people. 413 00:21:36,840 --> 00:21:40,000 Speaker 3: But I'm going to say it anyway. I have a pool. 414 00:21:40,520 --> 00:21:42,640 Speaker 4: I have a pal to a vegan on this call. 415 00:21:44,600 --> 00:21:47,000 Speaker 3: I have a pool. And no, it won't involve eating 416 00:21:47,000 --> 00:21:49,399 Speaker 3: any animals. I mean I could go there, but but 417 00:21:49,480 --> 00:21:53,639 Speaker 3: I have this pool. And the ducks in my neighbor 418 00:21:53,760 --> 00:21:56,480 Speaker 3: they fly into the pool and they just make a 419 00:21:56,560 --> 00:21:59,520 Speaker 3: huge mess. They just poop all over the pool, cover 420 00:21:59,600 --> 00:22:02,240 Speaker 3: all over everything. Right, And I'm always trying to get 421 00:22:02,280 --> 00:22:05,000 Speaker 3: the ducks to stay away, and there's YouTube videos like 422 00:22:05,040 --> 00:22:07,439 Speaker 3: I'm not the only one who has this problem. I 423 00:22:07,520 --> 00:22:10,359 Speaker 3: just want to take a little Raspberry Pie, program it 424 00:22:10,440 --> 00:22:13,840 Speaker 3: to roll around and just scare away the ducks. Maybe 425 00:22:13,960 --> 00:22:17,240 Speaker 3: have some sort of like NERF gun attachment that shoots 426 00:22:17,280 --> 00:22:19,000 Speaker 3: a little NERF gun. Is that bad? 427 00:22:19,080 --> 00:22:19,159 Speaker 5: Like? 428 00:22:19,280 --> 00:22:21,000 Speaker 3: Am I a bad person for wanting to do that? 429 00:22:21,000 --> 00:22:22,320 Speaker 3: That's the product I want to build. 430 00:22:22,600 --> 00:22:25,840 Speaker 4: I think that's honestly amazing. And you're not going to 431 00:22:25,960 --> 00:22:28,240 Speaker 4: hurt the ducks, And I don't want to hear that 432 00:22:28,400 --> 00:22:30,680 Speaker 4: I just want to scam. You're not trying to build 433 00:22:30,680 --> 00:22:34,520 Speaker 4: an autonomous weapon. I think. I think that sounds great. 434 00:22:35,000 --> 00:22:37,199 Speaker 5: Isn't this the Tony Soprano problem? 435 00:22:39,400 --> 00:22:40,879 Speaker 1: He liked it. I think he liked the ducks. I 436 00:22:40,920 --> 00:22:44,280 Speaker 1: think he Yeah, I. 437 00:22:44,359 --> 00:22:47,199 Speaker 5: Support this, Yeah, but it also sounds like you're going 438 00:22:47,240 --> 00:22:49,120 Speaker 5: to have to do all the work. I don't think 439 00:22:49,119 --> 00:22:51,600 Speaker 5: that this that applied intuition is going. 440 00:22:51,560 --> 00:22:52,040 Speaker 1: To help well. 441 00:22:52,040 --> 00:22:54,320 Speaker 3: So I buy the g So, I buy the go Pro, 442 00:22:54,440 --> 00:22:57,240 Speaker 3: I mean sorry, I buy the I buy the Raspberry Pie, 443 00:22:57,560 --> 00:23:00,000 Speaker 3: and I find some little one of those little robe 444 00:23:00,160 --> 00:23:02,480 Speaker 3: I even I think I have most of the hardware 445 00:23:02,480 --> 00:23:05,199 Speaker 3: I need. I just I just need to upload the 446 00:23:05,280 --> 00:23:09,640 Speaker 3: video and essentially train a model to recognize ducks and 447 00:23:09,880 --> 00:23:12,919 Speaker 3: you know, fire a little a little nerf gun in 448 00:23:12,960 --> 00:23:16,320 Speaker 3: its general direction, not at it to hit it. 449 00:23:16,640 --> 00:23:18,960 Speaker 2: Read for the for the for the layman. And I 450 00:23:19,200 --> 00:23:21,720 Speaker 2: put my hand up here, like what is the what 451 00:23:21,880 --> 00:23:25,320 Speaker 2: is the problem in robotics? This is solving like what what? 452 00:23:25,320 --> 00:23:28,000 Speaker 2: What isn't what was not possible yesterday that maybe possible 453 00:23:28,080 --> 00:23:30,160 Speaker 2: today because of this development. 454 00:23:30,280 --> 00:23:32,760 Speaker 3: Well, I I don't think this actually is not it's 455 00:23:32,800 --> 00:23:36,040 Speaker 3: not about like this isn't like an advancement in AI 456 00:23:36,240 --> 00:23:39,040 Speaker 3: or robotics. This is just democratizing it. I know you 457 00:23:39,080 --> 00:23:42,439 Speaker 3: all hate that word, but it's but it's letting people 458 00:23:42,600 --> 00:23:46,320 Speaker 3: like me or startups you know, with with fewer resources 459 00:23:46,400 --> 00:23:48,520 Speaker 3: get into this field. 460 00:23:47,840 --> 00:23:53,239 Speaker 2: To use natural language to to essentially train physical robots. 461 00:23:53,240 --> 00:23:55,520 Speaker 3: Right exactly like I want to I want to scare 462 00:23:55,560 --> 00:23:58,800 Speaker 3: the ducks away? Can you create a model that will 463 00:23:59,320 --> 00:24:03,600 Speaker 3: recognize the ducks and you know, do whatever. So it's 464 00:24:03,640 --> 00:24:05,480 Speaker 3: like a move in that in the direction of like 465 00:24:06,080 --> 00:24:08,720 Speaker 3: you know, we'll have more than just the robot vacuums 466 00:24:08,800 --> 00:24:10,720 Speaker 3: in our house, like we'll have a and you know, 467 00:24:10,800 --> 00:24:15,040 Speaker 3: eventually humanoids, but that's farther down the line. 468 00:24:15,680 --> 00:24:18,720 Speaker 2: You talked about something your story called the missing link 469 00:24:18,800 --> 00:24:22,000 Speaker 2: effect in robotics. What is that and how does that 470 00:24:22,040 --> 00:24:23,200 Speaker 2: kind of play into this story? 471 00:24:23,960 --> 00:24:26,639 Speaker 3: Yeah? That was sort of like this this concept that I, 472 00:24:27,080 --> 00:24:32,639 Speaker 3: you know, you're essentially like you can create. There's robotics 473 00:24:32,680 --> 00:24:36,040 Speaker 3: everywhere now, right, Like you go into a biotech lab. 474 00:24:36,080 --> 00:24:39,280 Speaker 3: I was in Boston last week, right, there's there's robotic 475 00:24:39,359 --> 00:24:42,879 Speaker 3: pipe heading machines in all these labs. Now, there's there's 476 00:24:42,960 --> 00:24:46,240 Speaker 3: all sorts of automation that you know, centrifugas, et cetera, 477 00:24:46,320 --> 00:24:49,959 Speaker 3: that are testing DNA and doing all sorts of stuff. 478 00:24:50,480 --> 00:24:52,600 Speaker 3: But like, you still need a lot of people to 479 00:24:52,680 --> 00:24:56,080 Speaker 3: run these labs because they're still built for humans. So 480 00:24:56,880 --> 00:25:00,200 Speaker 3: it's it's not like it's not a step chain. The 481 00:25:00,320 --> 00:25:04,280 Speaker 3: robotics gives you these incremental gains and efficiency. But my 482 00:25:04,400 --> 00:25:06,840 Speaker 3: point was like, let's say, and this is what a 483 00:25:06,880 --> 00:25:09,359 Speaker 3: lot of lab technicians say, not all of them agree, 484 00:25:09,400 --> 00:25:11,760 Speaker 3: but like, let's say you could have a humanoid robot 485 00:25:11,800 --> 00:25:14,199 Speaker 3: that has like five fingers and can do you know, 486 00:25:14,280 --> 00:25:16,639 Speaker 3: any sort of dexterous tasks that a human can do, 487 00:25:16,720 --> 00:25:19,240 Speaker 3: and you can just program it. All of a sudden, 488 00:25:19,359 --> 00:25:25,520 Speaker 3: every single biolab as they exist today can become completely automated, 489 00:25:26,000 --> 00:25:28,080 Speaker 3: and then you can just have these things run twenty 490 00:25:28,080 --> 00:25:31,679 Speaker 3: four to seven and that would like allow scientists to 491 00:25:31,760 --> 00:25:34,600 Speaker 3: run more experiments that they would never do today that 492 00:25:34,800 --> 00:25:37,320 Speaker 3: just there just wouldn't be the resources to do today, 493 00:25:37,840 --> 00:25:42,560 Speaker 3: And so that becomes like an exponential gain in scientific discovery, 494 00:25:42,600 --> 00:25:45,480 Speaker 3: and that sort of principle of like you plug it once, 495 00:25:45,520 --> 00:25:49,280 Speaker 3: you get once these robots become more general purpose and 496 00:25:49,320 --> 00:25:51,959 Speaker 3: you can sort of plug them into you know, all 497 00:25:52,000 --> 00:25:56,280 Speaker 3: the technology we already have, you you get exponential gains 498 00:25:56,400 --> 00:25:59,040 Speaker 3: versus just Okay, now I got like a ten percent 499 00:25:59,119 --> 00:26:00,879 Speaker 3: increase in efficiency, you're twenty percent. 500 00:26:01,280 --> 00:26:04,280 Speaker 2: So right now, for example, you could have a robot 501 00:26:04,320 --> 00:26:07,639 Speaker 2: that shot sort of nerf arrows at the ducks, but 502 00:26:07,680 --> 00:26:09,159 Speaker 2: then you have to go and pick them up yourself 503 00:26:09,200 --> 00:26:12,480 Speaker 2: and reload the robot, whereas in the future you could 504 00:26:12,520 --> 00:26:15,000 Speaker 2: also be a general purpose robot collect the arrows. 505 00:26:15,480 --> 00:26:18,200 Speaker 3: Sure, yes, you do that, but then think about that 506 00:26:18,280 --> 00:26:22,800 Speaker 3: in like at an industrial scale, right. Building construction is 507 00:26:22,800 --> 00:26:25,680 Speaker 3: a great example, Like there's all this automation and construction, right, 508 00:26:26,160 --> 00:26:28,640 Speaker 3: but you still need a ton of people, and that 509 00:26:28,720 --> 00:26:34,280 Speaker 3: creates like there's labor shortages, that creates like the bottleneck essentially. Right, 510 00:26:34,680 --> 00:26:38,080 Speaker 3: But once you can fully automate things, the things move 511 00:26:38,600 --> 00:26:43,119 Speaker 3: just exponentially faster. And that's like to me, that probably 512 00:26:43,240 --> 00:26:45,920 Speaker 3: in my view, and you might you all might disagree 513 00:26:45,960 --> 00:26:49,240 Speaker 3: with this. I think that's where humanoid robots are valuable. 514 00:26:49,720 --> 00:26:51,800 Speaker 3: Like people make this argument that, well, why do you 515 00:26:51,800 --> 00:26:54,639 Speaker 3: need humanoids, Like you could just create a special robot 516 00:26:54,680 --> 00:26:57,840 Speaker 3: for each individual purpose, but it just it doesn't work 517 00:26:57,880 --> 00:26:58,400 Speaker 3: that way, Like. 518 00:26:58,359 --> 00:27:01,680 Speaker 2: The industrial wood world was designed to be navigated by 519 00:27:01,680 --> 00:27:06,520 Speaker 2: real humans and therefore humanoid robots and are actually they 520 00:27:06,560 --> 00:27:08,879 Speaker 2: fit into the system we've built better than any other 521 00:27:08,960 --> 00:27:09,800 Speaker 2: types of robots. 522 00:27:10,440 --> 00:27:12,600 Speaker 3: Yeah, totally. I think that's I think they may be 523 00:27:12,680 --> 00:27:16,200 Speaker 3: a necessary step to get to like a thick full 524 00:27:16,640 --> 00:27:21,119 Speaker 3: full automation and like you know, fully roboticizing you know, 525 00:27:21,240 --> 00:27:21,920 Speaker 3: the economy. 526 00:27:22,480 --> 00:27:24,919 Speaker 5: But wait is the miss So the missing link is 527 00:27:24,960 --> 00:27:27,760 Speaker 5: the technology that will make humans obsolete? Is that? 528 00:27:27,840 --> 00:27:28,080 Speaker 4: What? 529 00:27:28,080 --> 00:27:29,200 Speaker 5: What is the missing link? 530 00:27:29,240 --> 00:27:32,520 Speaker 3: I still I mean, yeah, that's the that's the and 531 00:27:32,520 --> 00:27:34,639 Speaker 3: and look that was I put that in the articles 532 00:27:34,680 --> 00:27:37,560 Speaker 3: down like that. You know. The obvious next question is like, Okay, 533 00:27:37,560 --> 00:27:40,439 Speaker 3: what does that mean? For human labor. I personally like, 534 00:27:40,560 --> 00:27:42,639 Speaker 3: first of all, this is this is a ways of way, right, 535 00:27:42,720 --> 00:27:45,520 Speaker 3: Like this is not gonna happen overnight. But second of all, 536 00:27:46,240 --> 00:27:49,560 Speaker 3: I think that you just think about the economic growth 537 00:27:49,560 --> 00:27:51,919 Speaker 3: that comes from that world, Like I don't. I'm not 538 00:27:52,080 --> 00:27:55,720 Speaker 3: worried personally that humans are just gonna have nothing to 539 00:27:55,760 --> 00:27:59,120 Speaker 3: do like that that just is not gonna happen. We'll 540 00:27:59,119 --> 00:27:59,919 Speaker 3: find stuff to do. 541 00:28:00,200 --> 00:28:02,760 Speaker 2: You did refer on your story to the Nobel Prize 542 00:28:02,760 --> 00:28:07,240 Speaker 2: winging economist Darren Astimolgulu's research, who found that every industrial 543 00:28:07,320 --> 00:28:11,639 Speaker 2: robot added per thousand workers measurably reduces employment and wages 544 00:28:11,680 --> 00:28:12,879 Speaker 2: in the community where it lends. 545 00:28:13,520 --> 00:28:16,680 Speaker 3: Yeah, I just disagree, though. I just think that you're 546 00:28:16,720 --> 00:28:20,680 Speaker 3: not You're not thinking through all the ancillary benefits of 547 00:28:21,119 --> 00:28:23,800 Speaker 3: being able to do this kind of stuff and then 548 00:28:24,040 --> 00:28:25,400 Speaker 3: the economic growth that will happen. 549 00:28:25,440 --> 00:28:27,760 Speaker 2: I mean, but according according to according to the quote, 550 00:28:27,840 --> 00:28:29,399 Speaker 2: is measurably reduced. 551 00:28:30,560 --> 00:28:34,880 Speaker 3: Right. But like economists, they can only measure what already exists, right, 552 00:28:34,920 --> 00:28:37,040 Speaker 3: And this is this is the mistake everybody. It's like 553 00:28:37,359 --> 00:28:40,400 Speaker 3: you almost can't win this argument because it's like who 554 00:28:40,440 --> 00:28:43,959 Speaker 3: would have thought, like in nineteen hundred that, you know, 555 00:28:44,240 --> 00:28:47,280 Speaker 3: all the jobs that exist will be gone. But like 556 00:28:47,400 --> 00:28:50,320 Speaker 3: people will make millions of dollars sitting in front of 557 00:28:50,360 --> 00:28:54,080 Speaker 3: a webcam and just talking like the idea of a 558 00:28:54,160 --> 00:28:57,160 Speaker 3: creator could not have existed in their minds, right, So 559 00:28:58,120 --> 00:29:01,280 Speaker 3: there's I think humans will always find something that they 560 00:29:01,360 --> 00:29:05,080 Speaker 3: do that they're that we assign value to and will 561 00:29:05,120 --> 00:29:05,760 Speaker 3: be just fine. 562 00:29:06,200 --> 00:29:09,440 Speaker 4: Natasha Taylor, I'm just in my mind like right now, 563 00:29:09,560 --> 00:29:12,720 Speaker 4: daydreaming about what kind of robots I could have, you know, 564 00:29:12,920 --> 00:29:16,040 Speaker 4: automated parts of my life, because there's I'm thinking of 565 00:29:16,080 --> 00:29:18,160 Speaker 4: Reed's pool example, and I'm like, there's a few things 566 00:29:18,200 --> 00:29:23,160 Speaker 4: I'd like. I would like something to handle my garden honestly. 567 00:29:23,320 --> 00:29:26,320 Speaker 3: To prune or what would you will, yeah. 568 00:29:25,880 --> 00:29:30,080 Speaker 4: Prune weed, anything like water. I mean, they have automated 569 00:29:30,080 --> 00:29:32,880 Speaker 4: watering systems. I should just buy one off Amazon. But 570 00:29:32,880 --> 00:29:35,480 Speaker 4: but you know, I went out of town recently and 571 00:29:35,800 --> 00:29:38,600 Speaker 4: I needed to pay someone to come over and like 572 00:29:39,520 --> 00:29:42,400 Speaker 4: harvest the jilapenos that we're going to go rotten. Then 573 00:29:42,560 --> 00:29:44,280 Speaker 4: you know, just like manage it. 574 00:29:44,360 --> 00:29:47,040 Speaker 5: I guess a jilapeno harvesting robot. 575 00:29:47,120 --> 00:29:50,360 Speaker 3: Of course, I think a gardening robot is an even 576 00:29:50,760 --> 00:29:54,600 Speaker 3: more lucrative idea than than the Nerf nerve shooting duck robot. 577 00:29:54,680 --> 00:29:57,600 Speaker 2: I would I have nightmas whenever I travel, I have 578 00:29:57,680 --> 00:30:01,280 Speaker 2: nightmas about the garden just kept becoming you know, wasteland, 579 00:30:01,280 --> 00:30:03,080 Speaker 2: and I feel like I would, I would be a 580 00:30:03,200 --> 00:30:07,360 Speaker 2: very irrational spender on more about the plantain my garden. 581 00:30:09,080 --> 00:30:11,520 Speaker 4: Just a humanoid robot that lives in your backyard, that 582 00:30:11,560 --> 00:30:13,120 Speaker 4: can tend to your garden while you're. 583 00:30:13,000 --> 00:30:18,240 Speaker 2: Gone, Natasha, I feel like I feel like you've got 584 00:30:18,280 --> 00:30:20,240 Speaker 2: to a counterpoint here. 585 00:30:20,880 --> 00:30:25,719 Speaker 5: No, I mean I I have heard this argument about, 586 00:30:26,320 --> 00:30:30,000 Speaker 5: you know, we can't possibly comprehend the future like future 587 00:30:30,080 --> 00:30:34,760 Speaker 5: jobs from many, many, many executives. But I do think 588 00:30:34,800 --> 00:30:38,760 Speaker 5: it's really interesting the humanoid robot example in a lab. 589 00:30:38,840 --> 00:30:43,320 Speaker 5: Both my parents were like research scientists, and just thinking 590 00:30:43,360 --> 00:30:46,240 Speaker 5: about how it could potentially change that. I mean, first 591 00:30:46,280 --> 00:30:49,840 Speaker 5: of all, it's like just imagine the number of mistakes 592 00:30:49,840 --> 00:30:53,640 Speaker 5: and errors that could be made. But also, you know, 593 00:30:53,680 --> 00:30:56,800 Speaker 5: so like you can't you can't make the humans obsolete 594 00:30:57,000 --> 00:30:59,880 Speaker 5: because it's never going to be one hundred percent accuracy. 595 00:31:00,400 --> 00:31:02,600 Speaker 5: But in terms of all of the different things you 596 00:31:02,600 --> 00:31:04,960 Speaker 5: could try, Yeah, I mean I think that's that's like 597 00:31:05,320 --> 00:31:11,120 Speaker 5: pretty exciting when you apply it to like actual innovation 598 00:31:11,440 --> 00:31:14,680 Speaker 5: like new drugs, new science, new materials. 599 00:31:15,120 --> 00:31:18,800 Speaker 3: Well, here's a here's a peta point for Taylor like that, 600 00:31:19,000 --> 00:31:22,120 Speaker 3: I think this is the thing that people aren't thinking 601 00:31:22,160 --> 00:31:25,760 Speaker 3: about when you have so when you have like fully 602 00:31:25,880 --> 00:31:30,280 Speaker 3: robotic labs, right, what is like the number one way 603 00:31:30,320 --> 00:31:32,840 Speaker 3: that we test all these molecules. It's animals. 604 00:31:32,840 --> 00:31:36,960 Speaker 5: Oh, you know, people are building these like synthetic animals 605 00:31:36,960 --> 00:31:40,080 Speaker 5: so that we won't have to do testing and synthetic brains. 606 00:31:40,520 --> 00:31:43,640 Speaker 3: That is true. I looked at synthetic brains under the microscope, 607 00:31:43,680 --> 00:31:45,640 Speaker 3: like they look like little brains. I mean that's just 608 00:31:45,680 --> 00:31:48,920 Speaker 3: brain tissue. These or they call them organoids. It's fascinating. 609 00:31:49,000 --> 00:31:52,160 Speaker 3: But like but right now, like so one of the 610 00:31:52,200 --> 00:31:55,680 Speaker 3: researchers said, well, if you look at like the C elegance, 611 00:31:55,720 --> 00:31:58,320 Speaker 3: which is like this little worm that that was made 612 00:31:58,440 --> 00:32:01,959 Speaker 3: huge advances and like the longevity industry like all this testing. 613 00:32:02,280 --> 00:32:07,000 Speaker 3: They like sort of root force tested the DNA of 614 00:32:07,160 --> 00:32:12,000 Speaker 3: C elegance in order to you know, to essentially like 615 00:32:12,080 --> 00:32:14,320 Speaker 3: do like do testing on it. And that led to 616 00:32:14,360 --> 00:32:16,280 Speaker 3: a lot of this a lot of this new research, 617 00:32:16,360 --> 00:32:18,760 Speaker 3: right and they're like, well imagine if like but that's 618 00:32:18,800 --> 00:32:20,880 Speaker 3: a very simple organize and imagine if you could do 619 00:32:20,920 --> 00:32:23,320 Speaker 3: the same thing with mice, and then we could have 620 00:32:23,400 --> 00:32:26,120 Speaker 3: this greater understanding of mice and that would advance medicine. 621 00:32:26,120 --> 00:32:30,360 Speaker 3: And I'm like, I'm picturing like like a robotic you know, 622 00:32:30,480 --> 00:32:34,120 Speaker 3: twenty four to seven robotic like warehouse just full of mice, 623 00:32:34,200 --> 00:32:38,120 Speaker 3: being like they die eventually right in these tests. And 624 00:32:38,160 --> 00:32:40,960 Speaker 3: I'm just like, I think this is a big animal 625 00:32:41,040 --> 00:32:44,280 Speaker 3: rights like once it gets to that, it's already controversial. 626 00:32:44,520 --> 00:32:46,960 Speaker 2: Well, actually I want to share this in the in 627 00:32:47,240 --> 00:32:49,520 Speaker 2: the first segment, but but I didn't, but I will now. 628 00:32:49,560 --> 00:32:52,520 Speaker 2: So I had Stewart Russell on tech Stuff recently, who 629 00:32:52,640 --> 00:32:57,200 Speaker 2: was Elon's only expert witness in the open AI try 630 00:32:57,200 --> 00:32:59,560 Speaker 2: it about AI safety and one of the great sort 631 00:32:59,600 --> 00:33:02,120 Speaker 2: of advocate of the alignment problem and the paper clip 632 00:33:02,160 --> 00:33:05,239 Speaker 2: issue and all the other things. And his example was, 633 00:33:05,600 --> 00:33:08,560 Speaker 2: if you told the super intelligent AI to cure cancer, 634 00:33:09,040 --> 00:33:12,480 Speaker 2: the most rational first step would be to give find 635 00:33:12,520 --> 00:33:14,600 Speaker 2: a way to give all humans cancer, so you could 636 00:33:14,640 --> 00:33:18,360 Speaker 2: run as many simultaneous clinical trials as possible. So here 637 00:33:18,400 --> 00:33:21,160 Speaker 2: we are the connective tissue between segments one and two. 638 00:33:21,440 --> 00:33:24,320 Speaker 2: But now we go to the atbreak when we come back. 639 00:33:24,640 --> 00:33:28,560 Speaker 2: Bill Gates's daughter Phoebe gets into some startup trouble. Stay 640 00:33:28,600 --> 00:33:43,400 Speaker 2: with us, Welcome back, Taylor. I remember when Phoebe Gates 641 00:33:43,480 --> 00:33:46,880 Speaker 2: announced her AI shopping startup a little over a year ago, 642 00:33:47,720 --> 00:33:51,719 Speaker 2: investors from Kleiner Perkins to Sydney Sweeney piled into Fear. 643 00:33:52,800 --> 00:33:55,760 Speaker 2: Tell us what fear is and what's been going down? 644 00:33:56,480 --> 00:34:00,480 Speaker 4: Yeah, Fear. It's Phia. I feel like it of it's 645 00:34:00,480 --> 00:34:05,240 Speaker 4: hard to pronounce. So it's an affiliate shopping site. Basically, 646 00:34:05,480 --> 00:34:09,920 Speaker 4: it promises to give you deals on you know, whatever 647 00:34:09,920 --> 00:34:12,719 Speaker 4: you're shopping on online. It's very similar to Honey. It's 648 00:34:12,840 --> 00:34:15,520 Speaker 4: essentially the same business as Honey, which is very funny 649 00:34:15,520 --> 00:34:19,120 Speaker 4: because Honey got into trouble for the same thing like 650 00:34:19,280 --> 00:34:21,840 Speaker 4: a year ago. I think. I guess what's different is 651 00:34:21,840 --> 00:34:24,319 Speaker 4: that it promised to be powered by AI, so you know, 652 00:34:24,440 --> 00:34:27,799 Speaker 4: AI powered shopping assistant, help you shop, help you find 653 00:34:27,840 --> 00:34:30,960 Speaker 4: the best deals, blah blah blah. Well, it turns out 654 00:34:31,000 --> 00:34:33,520 Speaker 4: that they were doing the exact same thing that Honey did, 655 00:34:33,600 --> 00:34:38,440 Speaker 4: which is sort of essentially stealing other people's affiliate revenue 656 00:34:38,680 --> 00:34:39,839 Speaker 4: and using it as their own. 657 00:34:40,680 --> 00:34:44,120 Speaker 2: So how exacted as well? I do remember just last week, 658 00:34:44,160 --> 00:34:47,000 Speaker 2: you were talking about how useful it would be. Honest, 659 00:34:47,040 --> 00:34:48,640 Speaker 2: two weeks ago, I think how useful would be to 660 00:34:48,680 --> 00:34:51,319 Speaker 2: have an AI go and find lookalike products for what 661 00:34:51,360 --> 00:34:52,560 Speaker 2: you wanted for your living room. 662 00:34:53,400 --> 00:34:54,399 Speaker 4: Yeah, this is not This is. 663 00:34:54,360 --> 00:34:56,919 Speaker 2: Not the offering that that's solving that problem yet. 664 00:34:57,120 --> 00:35:00,239 Speaker 4: No, Yeah, it's I think I believe it. It's a 665 00:35:00,280 --> 00:35:04,839 Speaker 4: browser extension. Basically, there's this messy world of like advertising, 666 00:35:04,880 --> 00:35:08,040 Speaker 4: cookies and data. You know, lots of data is being 667 00:35:08,040 --> 00:35:12,080 Speaker 4: harvested on us all day. There's lots of things that 668 00:35:12,080 --> 00:35:14,359 Speaker 4: are happening on our browser that were maybe not aware of. 669 00:35:14,640 --> 00:35:17,759 Speaker 4: So say you went to say you were a FIA user, right, 670 00:35:18,200 --> 00:35:20,640 Speaker 4: and then you went to buy something on Nike dot com. 671 00:35:20,840 --> 00:35:23,680 Speaker 4: FIA might get credit for that sale even though they 672 00:35:23,680 --> 00:35:26,600 Speaker 4: didn't directly drive that sale, And so that costs retailers 673 00:35:26,600 --> 00:35:27,200 Speaker 4: a lot of money. 674 00:35:27,640 --> 00:35:29,840 Speaker 2: So it's not it doesn't it doesn't cause consumer harm. 675 00:35:29,840 --> 00:35:32,560 Speaker 2: But it's basically it's basically just stlling out the other 676 00:35:33,160 --> 00:35:37,360 Speaker 2: affiliate affiliate marketers and harming the retailers. 677 00:35:37,719 --> 00:35:40,440 Speaker 4: Yeah, and the idea is that if this happens on 678 00:35:40,480 --> 00:35:43,719 Speaker 4: a broad enough scale, of course, retailers will raise their 679 00:35:43,719 --> 00:35:45,600 Speaker 4: prices to deal with this stuff. 680 00:35:45,680 --> 00:35:49,680 Speaker 2: And you had Ben Adelman, who was a researcher looked 681 00:35:49,680 --> 00:35:52,360 Speaker 2: into this story on your podcast. So what did he 682 00:35:52,400 --> 00:35:53,960 Speaker 2: tell you and what made you want to kind of 683 00:35:54,239 --> 00:35:55,040 Speaker 2: cover this one? 684 00:35:55,280 --> 00:35:56,800 Speaker 4: I mean, I'll be real with you, guys. I just 685 00:35:56,840 --> 00:35:59,279 Speaker 4: thought this story was interesting, not from any of this 686 00:35:59,360 --> 00:36:01,960 Speaker 4: stuff like il like it's a pretty standard. It sounds 687 00:36:01,960 --> 00:36:04,520 Speaker 4: like Honey already did this kind of scam. It's fraud, 688 00:36:04,600 --> 00:36:06,360 Speaker 4: it's not good. It sounds like everyone does it not 689 00:36:06,440 --> 00:36:11,720 Speaker 4: to make it whatever. What I found interesting is the narrative. 690 00:36:11,920 --> 00:36:14,600 Speaker 4: And I'm very skeptical of like sort of these takedown 691 00:36:14,600 --> 00:36:18,160 Speaker 4: pieces on female founders and everything. But this is a 692 00:36:18,200 --> 00:36:23,640 Speaker 4: girl that raised forty three million dollars, that has some 693 00:36:23,719 --> 00:36:27,400 Speaker 4: of the top investors, you know, that got in Forbes 694 00:36:27,440 --> 00:36:32,040 Speaker 4: thirty under thirty, that is just getting unfathomable amounts of 695 00:36:32,080 --> 00:36:36,560 Speaker 4: sort of opportunities and ushered into this like tech elite 696 00:36:36,640 --> 00:36:40,880 Speaker 4: world basically for building what was essentially a Honey clone 697 00:36:41,239 --> 00:36:44,400 Speaker 4: and ultimately doing the same thing that Honey was doing. 698 00:36:44,480 --> 00:36:46,719 Speaker 4: And so I just I guess to me it was 699 00:36:46,840 --> 00:36:49,040 Speaker 4: it was more interesting. It is like a story about 700 00:36:49,080 --> 00:36:51,839 Speaker 4: kind of like how does access play a role, how 701 00:36:51,880 --> 00:36:54,480 Speaker 4: does fame? How does adjacency to power like play a 702 00:36:54,560 --> 00:36:57,000 Speaker 4: role in this Silicon Valley start up funding ecosystem and 703 00:36:57,520 --> 00:36:59,839 Speaker 4: the companies that are sort of considered successes. 704 00:37:00,600 --> 00:37:03,320 Speaker 3: I mean, if you could raise that much money, snap 705 00:37:03,360 --> 00:37:05,520 Speaker 3: your finger, well you probably could, honestly, Taylor, I mean 706 00:37:05,520 --> 00:37:08,520 Speaker 3: you're pretty well, but like you could start, you could 707 00:37:08,560 --> 00:37:12,919 Speaker 3: build a robotic you know, Gardner, Like, I mean, let's 708 00:37:12,960 --> 00:37:15,800 Speaker 3: do something more ambitious. I totally agree, Like what. 709 00:37:15,920 --> 00:37:19,680 Speaker 4: Right bart headline? Taylor tries to put Gardners this no, 710 00:37:20,080 --> 00:37:23,040 Speaker 4: and it's like I don't want to like downplay like 711 00:37:23,120 --> 00:37:27,279 Speaker 4: the idea of like AI shopping, but I just I 712 00:37:27,400 --> 00:37:31,880 Speaker 4: feel like we can think bigger about technology and about 713 00:37:31,920 --> 00:37:34,960 Speaker 4: like what we're like lauding and what we're like praising 714 00:37:35,000 --> 00:37:38,360 Speaker 4: and putting founders on like most exciting startup founders lists. 715 00:37:38,400 --> 00:37:41,720 Speaker 4: Like you know, that's kind of to me. I'm like hmmm, 716 00:37:41,840 --> 00:37:44,440 Speaker 4: And you know a lot was focused on sort of 717 00:37:44,480 --> 00:37:47,840 Speaker 4: Phoebe Gates is like, you know, she's an influencer, she's 718 00:37:47,880 --> 00:37:50,400 Speaker 4: out there, you know, building an audience. 719 00:37:50,000 --> 00:37:52,440 Speaker 5: And even when it launched, I thought it was like 720 00:37:52,560 --> 00:37:54,600 Speaker 5: a weird move for her. I'm like, you could do 721 00:37:54,760 --> 00:37:56,960 Speaker 5: you could have better nepotism than this you know, you 722 00:37:57,000 --> 00:38:00,239 Speaker 5: could have like a more impressive startup. She was talking 723 00:38:00,280 --> 00:38:02,360 Speaker 5: a lot about reproductive rights and like going in a 724 00:38:02,360 --> 00:38:05,440 Speaker 5: certain way with her influencer career. I thought this was 725 00:38:05,480 --> 00:38:09,439 Speaker 5: really a like lateral slash step down. 726 00:38:09,719 --> 00:38:13,200 Speaker 3: I mean, it's like one of those businesses where you're 727 00:38:13,239 --> 00:38:15,719 Speaker 3: like you type into like CHATGBT, like find a way 728 00:38:15,719 --> 00:38:16,680 Speaker 3: for me to make money? 729 00:38:17,640 --> 00:38:21,839 Speaker 5: Yeah, like ask your dad for help. If you haven't 730 00:38:21,880 --> 00:38:23,960 Speaker 5: been asking your dad for ask him for help. Because 731 00:38:23,960 --> 00:38:25,120 Speaker 5: this was not a good one. 732 00:38:25,560 --> 00:38:28,719 Speaker 2: Okay, but let's so so tech tech NEPO babies obviously, 733 00:38:29,719 --> 00:38:32,479 Speaker 2: you know PB Gates is kind of intriguing as one 734 00:38:33,040 --> 00:38:36,960 Speaker 2: read jobs has his has his fund investing in like 735 00:38:37,040 --> 00:38:40,319 Speaker 2: cancer cancer technologies. I'm not sure if I'm not sure 736 00:38:40,360 --> 00:38:42,520 Speaker 2: if anyone knows much about that or where it's going. 737 00:38:42,600 --> 00:38:45,080 Speaker 2: But if you do, chime in, and who are the 738 00:38:45,120 --> 00:38:48,000 Speaker 2: other tech NEPO babies who want to be in tech? 739 00:38:48,040 --> 00:38:49,120 Speaker 1: I guess my question. 740 00:38:49,360 --> 00:38:53,680 Speaker 5: There's already a lot of them in tech. Tim Draper's 741 00:38:53,680 --> 00:38:59,319 Speaker 5: sons are investors. You would be surprised if you just 742 00:38:59,400 --> 00:39:03,040 Speaker 5: look at you know, like middle management at the fank companies. 743 00:39:04,120 --> 00:39:07,720 Speaker 5: I think you see a lot of a lot. Yeah, 744 00:39:07,760 --> 00:39:12,520 Speaker 5: so well, Sophie Schmidt, she had she had Rest of World, 745 00:39:13,560 --> 00:39:15,560 Speaker 5: and I'm not sure how long she's going to be 746 00:39:15,640 --> 00:39:17,759 Speaker 5: affiliated with that. I think Rest of World is actually 747 00:39:17,800 --> 00:39:21,480 Speaker 5: an awesome publication. So that's that's one of the technico 748 00:39:22,120 --> 00:39:26,319 Speaker 5: products that I would definitely stand behind. Yeah, I feel 749 00:39:26,360 --> 00:39:29,280 Speaker 5: like often they bring them in on their funds. 750 00:39:29,640 --> 00:39:32,080 Speaker 4: Yeah, I mean I would say a really famous one 751 00:39:32,160 --> 00:39:36,239 Speaker 4: is David Ellison, Larry Ellison's son, who would be. 752 00:39:37,520 --> 00:39:38,960 Speaker 1: That would probably the example number one. 753 00:39:44,719 --> 00:39:47,759 Speaker 5: And we, like I have been thinking about this for 754 00:39:48,239 --> 00:39:52,319 Speaker 5: I guess decades, because we are not prepared for the 755 00:39:53,160 --> 00:39:56,480 Speaker 5: like the way that the that the children of these 756 00:39:56,719 --> 00:40:00,360 Speaker 5: tech billionaires are going to influence our world old with 757 00:40:00,400 --> 00:40:02,840 Speaker 5: the amount of capital that they will have access to 758 00:40:02,920 --> 00:40:06,319 Speaker 5: and the amount of like Taylor said, access just generally to. 759 00:40:06,800 --> 00:40:09,840 Speaker 5: I mean just imagine like Davos twenty third. 760 00:40:10,360 --> 00:40:12,240 Speaker 3: What do you what do you think it'll look like? 761 00:40:12,239 --> 00:40:14,680 Speaker 3: Like we've seen that we have the Rockefellers, we have 762 00:40:14,880 --> 00:40:17,120 Speaker 3: you know, there's there's all these families, but like, is 763 00:40:17,160 --> 00:40:19,960 Speaker 3: this going to be just more of that effort? 764 00:40:21,280 --> 00:40:23,840 Speaker 5: It's right, No, I don't think it's going to be 765 00:40:23,880 --> 00:40:26,160 Speaker 5: like the Rockefellers at all, because I think it's going 766 00:40:26,239 --> 00:40:30,200 Speaker 5: to be a lot of the kids trying to make 767 00:40:30,239 --> 00:40:32,719 Speaker 5: a name for themselves. So they pick a they pick 768 00:40:32,760 --> 00:40:36,319 Speaker 5: a cause, they pick a charity, they pick you know, 769 00:40:36,400 --> 00:40:38,600 Speaker 5: something that they want to be associated with. So the 770 00:40:38,640 --> 00:40:41,240 Speaker 5: money that would normally go to like kind of traditional 771 00:40:41,239 --> 00:40:44,759 Speaker 5: philanthropies which have their which have their own issues, is 772 00:40:44,800 --> 00:40:47,719 Speaker 5: probably going to go to like a startup, you know, 773 00:40:47,760 --> 00:40:50,040 Speaker 5: they want to when they show up at a Davo 774 00:40:50,160 --> 00:40:52,120 Speaker 5: circuit or whatever it is. They want to have like 775 00:40:52,239 --> 00:40:54,719 Speaker 5: something behind their name that's not just their mom or dad. 776 00:40:55,760 --> 00:41:06,600 Speaker 5: So yeah, it's just going to be I guess like neposlop. 777 00:41:03,480 --> 00:41:05,719 Speaker 3: Some of them. Some of them might be good. 778 00:41:06,480 --> 00:41:09,000 Speaker 4: Yeah, I mean I think it'll be interesting too, Like 779 00:41:09,280 --> 00:41:12,560 Speaker 4: with all of Elon Musk's kids, you know, Elon Musk, 780 00:41:12,640 --> 00:41:14,600 Speaker 4: like a lot of these Silicon Valley men also are 781 00:41:14,600 --> 00:41:17,880 Speaker 4: like obsessed with pro creating. They want to ensure that 782 00:41:17,920 --> 00:41:21,520 Speaker 4: their legacy continues, you know. So I wonder if all 783 00:41:21,560 --> 00:41:23,360 Speaker 4: said they're going to take a more like hands on 784 00:41:23,880 --> 00:41:26,719 Speaker 4: role than maybe some like finance guy or whatever, you know, 785 00:41:26,760 --> 00:41:28,000 Speaker 4: would have done with his children. 786 00:41:28,640 --> 00:41:31,080 Speaker 5: And I mean so many of them got radicalized from 787 00:41:31,120 --> 00:41:34,440 Speaker 5: their children, right, Like their children's politics radicalize them. So 788 00:41:34,520 --> 00:41:37,800 Speaker 5: maybe they will be like the class traders that tried 789 00:41:37,840 --> 00:41:43,959 Speaker 5: to push more, you know, more income equality or different things. 790 00:41:44,000 --> 00:41:47,280 Speaker 5: I mean, I'm not saying it's necessarily going to be slopped, 791 00:41:47,280 --> 00:41:50,200 Speaker 5: but you have to imagine. I mean, their dads are 792 00:41:50,239 --> 00:41:52,640 Speaker 5: the ones who are like Elon's like I did it 793 00:41:52,640 --> 00:41:55,040 Speaker 5: by myself. You know, I'm an immigrant, I like didn't 794 00:41:55,080 --> 00:41:57,719 Speaker 5: have any help. And so these kids are going to 795 00:41:57,800 --> 00:42:00,560 Speaker 5: have like a lot of chips on their shoulders, I'm saying. 796 00:42:00,600 --> 00:42:00,880 Speaker 5: And the. 797 00:42:03,760 --> 00:42:06,719 Speaker 3: Yeah, it's like dead right, Like they're not going to 798 00:42:06,760 --> 00:42:10,160 Speaker 3: give it away, are they? Is the philanthropy dead is 799 00:42:10,239 --> 00:42:11,279 Speaker 3: just going to be startups? 800 00:42:11,320 --> 00:42:14,359 Speaker 4: Well, I would say maybe the Dario. You know, we're 801 00:42:14,400 --> 00:42:18,239 Speaker 4: about to get a lot of ea anthropic millionaires and 802 00:42:18,280 --> 00:42:21,759 Speaker 4: billionaires they famously kind of give to a lot of causes. 803 00:42:22,400 --> 00:42:25,239 Speaker 5: Well, okay, but I went to this event in San 804 00:42:25,239 --> 00:42:27,680 Speaker 5: Francisco called what should we do about all this money? 805 00:42:32,000 --> 00:42:34,160 Speaker 3: Get it out of California? Get the money out? 806 00:42:34,760 --> 00:42:37,080 Speaker 5: Yeah, and I think they're going to go into donor 807 00:42:37,120 --> 00:42:39,920 Speaker 5: advice funds, which means like no transparency. There's just a 808 00:42:39,920 --> 00:42:41,680 Speaker 5: lot of ways to like set it up as though 809 00:42:41,719 --> 00:42:44,120 Speaker 5: it looks like a charity but there's no transparency, you know, 810 00:42:45,000 --> 00:42:49,200 Speaker 5: public accountability. And I think it's quite possible too that 811 00:42:49,239 --> 00:42:51,640 Speaker 5: a lot of this money will be going into AI safety. 812 00:42:52,000 --> 00:42:56,680 Speaker 3: Oh god, well we'll have very safe AI. That'll be great. 813 00:42:57,880 --> 00:42:58,080 Speaker 4: Yeah. 814 00:42:58,160 --> 00:43:03,000 Speaker 2: Right, Well, that's all we have time for today. Thank 815 00:43:03,000 --> 00:43:10,960 Speaker 2: you all so much for joining for tech stuff. I'm 816 00:43:10,960 --> 00:43:14,719 Speaker 2: mos Voloshin. This episode was produced by Eliza Dennis. It 817 00:43:14,760 --> 00:43:18,040 Speaker 2: was executive produced by me and Julian Nutta for Kaleidoscope 818 00:43:18,440 --> 00:43:22,800 Speaker 2: and Katrina norvelbe iHeart Podcasts. Jack instantly mixed this episode 819 00:43:22,800 --> 00:43:25,640 Speaker 2: and Kyle Murdoch wrote olph theme song. A special thank 820 00:43:25,680 --> 00:43:29,760 Speaker 2: you to Taylor Lorenz, Natasha Tku and Read Albergotti. Please 821 00:43:29,880 --> 00:43:32,000 Speaker 2: check out all the work they put out into the world. 822 00:43:32,239 --> 00:43:34,120 Speaker 2: We're lucky to call them friends of the Pod.