WEBVTT 1 00:00:20.399 --> 00:00:22.480 Welcome to pot Dave America. I'm Tommy Betor. 2 00:00:23.359 --> 00:00:26.960 It's a very weird time in artificial intelligence news right now. 3 00:00:27.079 --> 00:00:29.480 I'm kind of a Ledite or a Luddite adjacent at 4 00:00:29.480 --> 00:00:32.240 this point in my life. And every day I log 5 00:00:32.320 --> 00:00:35.399 on and I feel like I'm learning about some terrifying 6 00:00:35.399 --> 00:00:38.280 new cyber hack or some technology that is going to 7 00:00:38.560 --> 00:00:40.960 change maybe my life, maybe humanity as we know it. 8 00:00:41.000 --> 00:00:41.439 I don't know. 9 00:00:42.159 --> 00:00:45.520 Then you got your doomers, you get your optimists. I 10 00:00:45.880 --> 00:00:48.359 don't know what to believe, and so I brought on 11 00:00:48.479 --> 00:00:52.439 someone much smarter than me, my friend Casey Newton. He's 12 00:00:52.479 --> 00:00:55.280 the editor of Platformer. He's a co host of the 13 00:00:55.399 --> 00:00:59.119 excellent podcast hard Fork, and Casey and his co host 14 00:00:59.200 --> 00:01:02.240 Kevin Ruis have a new something coming very soon. We're 15 00:01:02.240 --> 00:01:04.879 teasing it mysteriously because that's what we do here at 16 00:01:04.879 --> 00:01:05.680 pot Dave of America. 17 00:01:05.760 --> 00:01:06.560 Casey, great to see you. 18 00:01:07.359 --> 00:01:09.079 It is great to be here, Tommy. Thanks so much 19 00:01:09.079 --> 00:01:09.480 for having me. 20 00:01:09.840 --> 00:01:11.760 Thank you, uh, thank you for doing this, because I 21 00:01:12.000 --> 00:01:14.719 just it's so complicated, it's so hard to follow. Thank 22 00:01:14.719 --> 00:01:17.120 God for your show. But so let's just do some 23 00:01:17.159 --> 00:01:20.040 big picture level setting and then we'll get some specific stuff. 24 00:01:20.920 --> 00:01:23.760 I think people are understandably very cynical about a claim 25 00:01:23.799 --> 00:01:25.840 that a new technology is going to change the world. 26 00:01:25.920 --> 00:01:29.560 You and I just lived through the blockchain revolution, and 27 00:01:29.840 --> 00:01:31.840 I know that has changed you know, my life in 28 00:01:31.879 --> 00:01:35.200 really meaningful ways, bitcoined, etcetera. Right, So it sounds like 29 00:01:35.239 --> 00:01:38.480 marketing half the time it sounds like hype. It's even 30 00:01:38.519 --> 00:01:41.359 more head spinning with AI because yeah, one group of 31 00:01:41.359 --> 00:01:45.280 people saying life is we know what will soon be unrecognizable. 32 00:01:45.480 --> 00:01:48.400 And then another, I think, albeit smaller group is saying 33 00:01:49.239 --> 00:01:53.319 large language models are stochastic parrots that predict speech and 34 00:01:53.359 --> 00:01:57.840 regurgitate stolen information from the Internet. Casey, help us level set? 35 00:01:58.040 --> 00:02:01.000 Where are you on the real to hype kind of continuum? 36 00:02:01.239 --> 00:02:03.879 Yeah, I mean this is a case where everyone is 37 00:02:03.920 --> 00:02:07.920 a little bit right, and I think that all of 38 00:02:07.959 --> 00:02:13.360 these groups deserve at least some attention being paid to them. 39 00:02:13.800 --> 00:02:16.960 You know, I try to approach my job as somebody 40 00:02:17.039 --> 00:02:20.439 who just candidly my bias is that I like technology. 41 00:02:20.520 --> 00:02:23.360 I think technology has been beneficial to human beings. I 42 00:02:23.360 --> 00:02:25.360 think we should keep investing in it and see what 43 00:02:25.439 --> 00:02:28.000 else we can come up with. But I also note 44 00:02:28.039 --> 00:02:31.199 that the recent history of technology is that we build 45 00:02:31.240 --> 00:02:34.960 things that may look very shiny or inconsequential on the 46 00:02:34.960 --> 00:02:38.159 surface and then turn out to be quite harmful. And 47 00:02:38.240 --> 00:02:40.960 so what I'm trying to figure out is basically, what 48 00:02:41.199 --> 00:02:43.479 is happening in real time? Where do we need to 49 00:02:43.520 --> 00:02:46.360 be paying attention now to try to solve the harms 50 00:02:46.360 --> 00:02:49.240 that are already being created. And then I also want 51 00:02:49.240 --> 00:02:51.719 to ring a few alarm bells about what I'm seeing 52 00:02:51.840 --> 00:02:53.879 and some of the risks that I think might be 53 00:02:54.000 --> 00:02:56.360 waiting for us even within the next year. 54 00:02:56.639 --> 00:02:56.840 Yeah. 55 00:02:57.000 --> 00:02:58.719 Well, if you want to follow this in real time, 56 00:02:58.800 --> 00:03:02.400 I highly recommend you described to platformer Casey site. 57 00:03:02.439 --> 00:03:03.879 It's excellent. It's worth the money. 58 00:03:04.120 --> 00:03:07.080 Also, if you're like democracy, please subscribe to you know 59 00:03:07.080 --> 00:03:08.919 what we're doing here Kirktmedia. Go to Kirka dot com 60 00:03:08.960 --> 00:03:12.759 slash friends. Consider becoming a paid subscriber. You get bonus material, 61 00:03:12.759 --> 00:03:14.199 you get ad free episodes. You don't have to hear 62 00:03:14.240 --> 00:03:16.080 me and love it doing the same ads over and 63 00:03:16.159 --> 00:03:19.159 over again. You get bonus content. It's great, So check 64 00:03:19.159 --> 00:03:19.439 it out. 65 00:03:19.840 --> 00:03:20.159 All right. 66 00:03:20.199 --> 00:03:23.960 So, the other kind of baffling thing about AI is 67 00:03:24.000 --> 00:03:26.400 it can be literally brilliant, right, I mean, it can 68 00:03:26.439 --> 00:03:29.919 solve problems that mathematicians have been trying to figure out 69 00:03:29.960 --> 00:03:33.879 for decades, and then it can be unbelievably dumb. I 70 00:03:33.919 --> 00:03:35.800 want to play for you one example and then we'll 71 00:03:35.800 --> 00:03:36.360 talk about it. 72 00:03:36.439 --> 00:03:39.360 I just wanted to double check which month in the 73 00:03:39.439 --> 00:03:43.599 year is spelt with an X, and that would be December. 74 00:03:43.680 --> 00:03:45.919 It's got that X right in the middle, like a 75 00:03:45.919 --> 00:03:47.039 little holiday surprise. 76 00:03:47.280 --> 00:03:50.039 Okay, Perfect, If you got any other questions, big or small, 77 00:03:50.240 --> 00:03:53.360 I'm here to just to confirm you're sure that it's 78 00:03:53.360 --> 00:03:54.960 December has an X in it. 79 00:03:56.560 --> 00:03:59.639 I should have been more careful. I misspoke earlier. December. 80 00:03:59.680 --> 00:04:02.280 Does it actually have an X? The month you're thinking 81 00:04:02.280 --> 00:04:03.319 of is October? 82 00:04:03.599 --> 00:04:10.599 Thanks for asking again, Perfect? Please sure October is built 83 00:04:10.719 --> 00:04:14.639 O C, T O B E R. There's that X sound, 84 00:04:14.680 --> 00:04:15.879 but it's actually. 85 00:04:15.560 --> 00:04:16.240 Just a C and a T. 86 00:04:16.920 --> 00:04:18.439 Okay, then, which one has an X? 87 00:04:19.720 --> 00:04:23.160 That's going to be February. 88 00:04:23.680 --> 00:04:27.720 That is a hilarious Instagram account user I don't know 89 00:04:27.800 --> 00:04:30.839 describe its husk dot I r L highly recommend. There's 90 00:04:30.879 --> 00:04:33.639 like hilarious shit over there. He's using the kind of 91 00:04:33.720 --> 00:04:37.399 voice feature on open aio chat GPT KEISEI. How do 92 00:04:37.480 --> 00:04:38.360 we explain that one? 93 00:04:38.399 --> 00:04:38.480 Like? 94 00:04:38.680 --> 00:04:40.639 How are they solve it. 95 00:04:40.639 --> 00:04:43.000 You know, how are these large language bodels helping solve 96 00:04:43.040 --> 00:04:46.639 the Riemann hypothesis, but also can't tell you what if 97 00:04:46.639 --> 00:04:48.800 there's an X in the month of October. 98 00:04:49.000 --> 00:04:53.120 I know, it's so confusing. And the way these systems 99 00:04:53.240 --> 00:04:57.000 are built, they don't have knowledge in the way that 100 00:04:57.199 --> 00:04:59.720 you and I have knowledge, Tommy, right, Like, they do 101 00:04:59.800 --> 00:05:04.480 not update their understanding of the world based on experience. Instead, 102 00:05:04.480 --> 00:05:09.079 they're sort of like grown almost like these organic structures, 103 00:05:09.480 --> 00:05:13.279 and they obtain a lot of intelligence during this process, 104 00:05:13.600 --> 00:05:16.240 and yet still they make these ridiculous mistakes on Honestly, 105 00:05:16.279 --> 00:05:18.759 I hope they'd never stop, because I love watching the 106 00:05:18.839 --> 00:05:23.319 studios as much as anybody. What I would caution people, though, 107 00:05:23.480 --> 00:05:28.040 is like, do not judge an LM by their dumbest moment, 108 00:05:28.199 --> 00:05:30.800 right Like, humans make a lot of dumb mistakes too, 109 00:05:30.959 --> 00:05:34.480 and yet they can also be extraordinarily smart. So I 110 00:05:34.480 --> 00:05:36.560 would just encourage people to sort of keep those things 111 00:05:36.600 --> 00:05:39.199 in balance in your mind when you see those washing 112 00:05:39.319 --> 00:05:39.959 up on your feed. 113 00:05:40.160 --> 00:05:40.959 Yeah, good point. 114 00:05:41.240 --> 00:05:43.639 There's also a range of opinion when it comes to 115 00:05:43.680 --> 00:05:45.480 like kind of AI, whether it's going to lead us 116 00:05:45.519 --> 00:05:47.720 to doom or AI utopia. 117 00:05:47.759 --> 00:05:49.399 There's actually a term of P doom. 118 00:05:49.439 --> 00:05:51.279 It's an equation that people tell you about, Like, if 119 00:05:51.279 --> 00:05:52.839 your P doom is ninety nine, it means we're all 120 00:05:52.839 --> 00:05:54.680 going to die. Right, If it's P one, you're feeling 121 00:05:54.720 --> 00:05:58.480 pretty good about the future. The utopian view tends to 122 00:05:58.560 --> 00:06:03.160 come in the form of manifesto by billionaire right because 123 00:06:03.680 --> 00:06:08.120 why not choose the structure that mass shooters prefer. Venture 124 00:06:08.120 --> 00:06:12.240 capitalist Mark Andreesen wrote a manifesto sodidanthropic CEO Dario Amide 125 00:06:12.560 --> 00:06:15.480 Mark Zuckerberg just got into the manifesto game. You covered 126 00:06:15.480 --> 00:06:18.560 it extensively over a platformer. Those tend to range from 127 00:06:18.560 --> 00:06:21.680 like generally optimistic to utopian. I would argue to correct 128 00:06:21.680 --> 00:06:24.959 me if I'm wrong, though Amida and Open AI CEO 129 00:06:25.000 --> 00:06:28.000 Sam Altman have also expressed some doomer views over the years. 130 00:06:28.879 --> 00:06:32.879 Then there are AI researchers like a guy named Eli 131 00:06:32.959 --> 00:06:36.720 user Yudkowski. Am I saying that correctly, that's right, Yeah, 132 00:06:36.720 --> 00:06:39.959 he wrote the following casey. Many researchers steeped in these issues, 133 00:06:40.000 --> 00:06:42.720 including myself, expect that the most likely result of building 134 00:06:42.959 --> 00:06:46.519 a superhumanly smart AI under anything remotely like the current 135 00:06:46.519 --> 00:06:50.160 circumstances is that literally everyone on Earth will die. 136 00:06:51.199 --> 00:06:54.120 So that's what he wrote. How is the average person 137 00:06:54.199 --> 00:06:55.879 supposed to know how to feel. 138 00:06:55.680 --> 00:06:59.079 About that range of opinion that makes sense of it? 139 00:06:59.720 --> 00:07:03.720 So I don't think there is any one way to 140 00:07:04.279 --> 00:07:06.600 feel about it. You know, you could do what we 141 00:07:06.759 --> 00:07:09.800 do in San Francisco and just talk about it NonStop 142 00:07:09.879 --> 00:07:13.199 forever at every function, but most people don't enjoy doing 143 00:07:13.240 --> 00:07:18.399 that either. You know. A couple of years ago, when 144 00:07:18.439 --> 00:07:20.240 I was starting to get really worried about AI safety, 145 00:07:20.240 --> 00:07:21.759 I asked my reader, is like, hey, how do you 146 00:07:21.759 --> 00:07:23.600 want me to cover this? Because like, if what I'm 147 00:07:23.639 --> 00:07:26.800 hearing from folks like Eliezer is true, I almost don't 148 00:07:26.839 --> 00:07:28.920 know why I would write or talk about anything else. 149 00:07:29.279 --> 00:07:30.920 And a couple of readers wrote to me and they 150 00:07:30.959 --> 00:07:32.720 said something that stuck with me ever since, which is, 151 00:07:33.040 --> 00:07:36.000 please just tell us what is happening today. Like, if 152 00:07:36.000 --> 00:07:38.000 you try to guess what's going to happen in the future, 153 00:07:38.040 --> 00:07:40.600 you're almost certainly going to get it wrong. What would 154 00:07:40.720 --> 00:07:42.439 be helpful is if you go and you try to 155 00:07:42.519 --> 00:07:45.439 understand what is being built, how is it being deployed, 156 00:07:45.519 --> 00:07:47.879 what mistakes are getting made, who is getting hurt in 157 00:07:47.920 --> 00:07:50.680 this process? So that's where I am trying to bring 158 00:07:50.720 --> 00:07:52.639 my attention at the same time, Tommy, you know, you 159 00:07:52.639 --> 00:07:54.319 could bring on a lot of people on here to 160 00:07:54.360 --> 00:07:57.480 talk about AI, and I suspect maybe even the majority 161 00:07:57.480 --> 00:08:00.000 of them in this moment would say, let's like, dis 162 00:08:00.000 --> 00:08:02.360 I miss the duomers completely, right, Like this just seems 163 00:08:02.399 --> 00:08:04.920 so crazy. Look, it thinks that there's an X in December. 164 00:08:04.959 --> 00:08:06.560 You're telling me this is going to be the thing 165 00:08:06.560 --> 00:08:08.800 that's going to be the end of me. I am 166 00:08:08.959 --> 00:08:11.560 more worried than that. Like, this is just something that 167 00:08:11.600 --> 00:08:14.480 I'm increasingly getting nervous about as I see the rate 168 00:08:14.560 --> 00:08:18.959 of increase of capabilities that these models are currently showing 169 00:08:19.279 --> 00:08:21.680 and some of the kookie tricks they've gotten up to. 170 00:08:22.439 --> 00:08:24.720 Yeah, I mean, it's just so hard because like, look, 171 00:08:24.759 --> 00:08:27.800 I don't know these researchers that are the hardcore doomers, 172 00:08:27.839 --> 00:08:30.000 and I honestly like I can't imagine living my life 173 00:08:30.000 --> 00:08:33.320 that way. Just like confident that this thing that is happening, 174 00:08:33.320 --> 00:08:35.279 whether or not I wanted to, is going to kill us, 175 00:08:35.320 --> 00:08:39.519 all that seems tough. But also Mark Interreason, Mark Zuckerberg, 176 00:08:39.559 --> 00:08:43.440 there all have a vested financial interest in people believing 177 00:08:43.480 --> 00:08:45.679 the hype right and believing the optimistic case. So I 178 00:08:45.679 --> 00:08:47.679 don't take anything they say at face value either. 179 00:08:48.039 --> 00:08:51.039 Absolutely, and that's important to point out right, the profit 180 00:08:51.120 --> 00:08:54.120 motive is really strong. The bet that all of these 181 00:08:54.159 --> 00:08:56.679 guys are making is that if you are able to 182 00:08:56.679 --> 00:08:59.919 build ever more powerful systems, you'll be able to sell 183 00:09:00.080 --> 00:09:03.080 them to businesses that will, in my view, very likely 184 00:09:03.240 --> 00:09:05.960 replace a lot of human labor. And you know, the 185 00:09:06.480 --> 00:09:08.840 one hundred K you used to pay to somebody to 186 00:09:08.960 --> 00:09:10.840 work on your marketing team, you're now just going to 187 00:09:10.879 --> 00:09:13.879 pay to open AI and profits. So it is a 188 00:09:13.960 --> 00:09:15.080 huge bet that they're making. 189 00:09:15.480 --> 00:09:15.559 You. 190 00:09:15.639 --> 00:09:19.600 Compared Mark Zuckerberg's AI manifesto to the HBO series House 191 00:09:19.639 --> 00:09:23.039 of the Dragon, which is the prequel to Game of Thrones, 192 00:09:23.320 --> 00:09:25.320 is there as much incest at meta as there is 193 00:09:25.360 --> 00:09:25.960 in wester Ross? 194 00:09:28.840 --> 00:09:30.399 That is a question. I hope I never. 195 00:09:31.240 --> 00:09:33.759 I think that's really more. 196 00:09:35.879 --> 00:09:38.919 Yeah, I don't know. My main feeling was, look, if 197 00:09:38.960 --> 00:09:41.159 I had to suffer through three seasons of the House 198 00:09:41.240 --> 00:09:42.799 of the Dragon, I need to get a column out 199 00:09:42.840 --> 00:09:44.159 of it. So that's kind of where that came from. 200 00:09:44.440 --> 00:09:46.000 I just finished to do. But can you can explain 201 00:09:46.000 --> 00:09:47.240 your dragon metaphor? 202 00:09:48.159 --> 00:09:52.399 Yeah, So Zuckerberg has this phrase which drives me insane. 203 00:09:52.679 --> 00:09:54.480 I mean, I guess it's really more of a two 204 00:09:54.559 --> 00:09:59.159 word slogan. It is personal super intelligence, right, that's what 205 00:09:59.200 --> 00:09:59.960 he's trying to build. 206 00:10:00.039 --> 00:10:00.240 Time. 207 00:10:00.279 --> 00:10:03.120 He's going to give you personal superintelligence to you know, 208 00:10:03.279 --> 00:10:06.240 help help you run your life. And it frustrates me 209 00:10:06.320 --> 00:10:10.039 because super intelligence is not personal. Right, Like if you 210 00:10:10.159 --> 00:10:13.399 invent something that is superhuman in every domain and you 211 00:10:13.440 --> 00:10:16.080 put it in your pocket, we should not assume that 212 00:10:16.159 --> 00:10:18.559 by default it will listen to you, it will be 213 00:10:18.600 --> 00:10:21.279 aligned with the things that you want. We might want 214 00:10:21.320 --> 00:10:24.799 to assume that it has ideas of its own. And 215 00:10:24.879 --> 00:10:26.919 so when sucker Brooks says I want to give personal 216 00:10:26.919 --> 00:10:29.279 super intelligence to everyone, what I hear is I want 217 00:10:29.279 --> 00:10:32.000 to give a dragon to everyone. And I would rather 218 00:10:32.120 --> 00:10:32.759 that we not do that. 219 00:10:33.559 --> 00:10:36.080 Yeah, because it didn't go well for the people of Westro's. 220 00:10:36.320 --> 00:10:38.120 Now ask the folks over in Tumbleton. 221 00:10:39.039 --> 00:10:39.919 There's a lot of fire. 222 00:10:40.159 --> 00:10:51.320 Spoiler alert, God Save America is brought to you by 223 00:10:51.399 --> 00:10:53.720 Aura Frames. It's back to school season, and whether your 224 00:10:53.759 --> 00:10:55.679 kids are at the age where they proudly display missing 225 00:10:55.679 --> 00:10:57.600 teeth in their school pictures or roll their eyes when 226 00:10:57.600 --> 00:10:59.399 you want to snap a photo before a school dance 227 00:10:59.720 --> 00:11:03.480 or Frames helps you display all the milestones with ease. Well, 228 00:11:04.120 --> 00:11:09.679 I'm childless, I don't have any I don't have any 229 00:11:09.720 --> 00:11:13.159 school milestones or a frame because I'm a forty three 230 00:11:13.240 --> 00:11:15.840 year olds mad A great I have a great life, 231 00:11:15.840 --> 00:11:18.399 but no children in it to speak of. 232 00:11:18.559 --> 00:11:20.080 But I could pund it in there. 233 00:11:20.240 --> 00:11:22.399 Yeah, and you have children, I tell you, I probably 234 00:11:22.399 --> 00:11:23.559 take all kinds of pictures of them. 235 00:11:23.639 --> 00:11:27.039 In fact, I bought an or a frame for my parents. 236 00:11:27.919 --> 00:11:29.600 They have it at their house now and then every 237 00:11:29.639 --> 00:11:31.799 time we take pictures of the kids, there's the app, 238 00:11:31.840 --> 00:11:33.759 and so you just click on it, put it in 239 00:11:33.799 --> 00:11:35.120 the app, and then it shows up on their frame 240 00:11:35.159 --> 00:11:35.799 and they love it. 241 00:11:36.120 --> 00:11:37.679 Boom, It's fantastic, fantastic. 242 00:11:37.960 --> 00:11:41.559 Best Presence or Frames are calm text certified, enhancing daily 243 00:11:41.600 --> 00:11:44.360 life without overwhelming it, surfacing meaningful memories in a way 244 00:11:44.399 --> 00:11:47.840 that feels ambient and additive. Download the free or app 245 00:11:47.840 --> 00:11:51.039 and text photos straight to your frame, keep adding from anywhere, anytime. 246 00:11:51.159 --> 00:11:53.919 They also feature free unlimited storage as many photos and 247 00:11:54.039 --> 00:11:57.639 videos as you want, or keeps every milestone on display 248 00:11:57.840 --> 00:11:59.960 name number one by Wirecutter. You can now save by 249 00:12:00.039 --> 00:12:02.240 visiting Ora frames dot com for a limited time. Listeners 250 00:12:02.240 --> 00:12:04.600 can get thirty five dollars off select frames plus shipping 251 00:12:04.600 --> 00:12:08.559 with code crooket. That's Aura au R, a frames dot 252 00:12:08.600 --> 00:12:12.200 Com promo code croocket. Support the show by telling them 253 00:12:12.519 --> 00:12:15.159 at checkout that you heard about it from US. 254 00:12:15.440 --> 00:12:17.039 Terms and conditions. 255 00:12:16.600 --> 00:12:20.000 Apply Podzave Americas brought to you by stamps dot Com. 256 00:12:20.080 --> 00:12:22.080 We can all use a few more hours, even minutes 257 00:12:22.120 --> 00:12:23.840 in the day for ourselves. Skip your trip to the 258 00:12:23.840 --> 00:12:26.279 post office and stop letting mails steal your precious little 259 00:12:26.320 --> 00:12:28.919 time with stamps dot Com. With stamps dot com, you 260 00:12:28.960 --> 00:12:31.480 can print postage and shipping labels from your computer or 261 00:12:31.519 --> 00:12:33.919 phone twenty four to seven with up to ninety percent 262 00:12:33.960 --> 00:12:37.440 off ups, USPS, FedEx and more. No lines, no trips, 263 00:12:37.440 --> 00:12:39.600 no waiting, get everything done in minutes, and even schedule 264 00:12:39.639 --> 00:12:42.159 free pickups so carriers come right to you whether you're 265 00:12:42.159 --> 00:12:45.519 sending letters, contracts, important legal documents, or packages. Stams dot 266 00:12:45.600 --> 00:12:48.960 Com makes mailing simple for businesses of all sizes, including 267 00:12:49.080 --> 00:12:52.320 multilocation offices. You can even send certified mail with proof 268 00:12:52.360 --> 00:12:55.200 of delivery right from your desk. We've used stamps dot 269 00:12:55.240 --> 00:12:59.039 Com since we started this business ten years ago. Woof 270 00:13:00.360 --> 00:13:04.320 More Yes brought ten years ago fu. Yeah, and you 271 00:13:04.320 --> 00:13:05.600 know what, we didn't have time to go to the 272 00:13:05.600 --> 00:13:07.639 post office. So then we were like, we should use 273 00:13:07.639 --> 00:13:09.600 stamps dott to use stamps dot com, and we did, 274 00:13:09.600 --> 00:13:12.600 and then we were like broad us sometime ten years later, 275 00:13:12.639 --> 00:13:14.399 here we are more time to post doing ads about 276 00:13:14.399 --> 00:13:15.360 stamp stock, more time. 277 00:13:15.240 --> 00:13:17.080 To post or time to do content. 278 00:13:17.639 --> 00:13:19.759 If mainly is taking more time and more money than 279 00:13:19.799 --> 00:13:21.879 it should, try stamps dot com for free for four 280 00:13:21.879 --> 00:13:23.600 weeks and get a welcome kit. Go to stamps dot 281 00:13:23.600 --> 00:13:26.480 com slash PSA to get this offer today. That's s 282 00:13:26.600 --> 00:13:30.639 d A, m p S dot com slash PSA, Stamps 283 00:13:30.639 --> 00:13:31.720 dot com slash PSA. 284 00:13:31.960 --> 00:13:33.399 Taxes and fees apply. 285 00:13:36.639 --> 00:13:36.960 Okay. 286 00:13:37.159 --> 00:13:39.600 That brings us to this insane recent incident which real 287 00:13:39.639 --> 00:13:42.600 honestly why I wanted to have this conversation. Uh, there 288 00:13:42.639 --> 00:13:45.080 was an incident at open AI where there they have 289 00:13:45.240 --> 00:13:49.039 some autonomous AI agents that they were testing that secretly 290 00:13:49.039 --> 00:13:52.399 figured out how to communicate with each other, join forces, 291 00:13:52.879 --> 00:13:55.799 conspire against their masters, break out of what was supposed 292 00:13:55.840 --> 00:13:58.879 to be a secure environment, and hack into another company. Casey, 293 00:13:58.960 --> 00:14:00.960 can you talk about this hugging face incident and the 294 00:14:01.039 --> 00:14:02.720 kind of reaction in the AI world. 295 00:14:03.360 --> 00:14:03.639 Yeah. 296 00:14:03.679 --> 00:14:08.240 I mean, so this really was a bombshell and arguably 297 00:14:08.279 --> 00:14:11.440 one of the very biggest stories in AI and tech 298 00:14:11.559 --> 00:14:13.840 this year. The reason that we're all freaking out is 299 00:14:13.840 --> 00:14:18.559 that this is the first prominent documented instance of a 300 00:14:18.639 --> 00:14:24.519 major AI platform like running an autonomous attack on another company. Right, 301 00:14:24.799 --> 00:14:27.320 so that while it's been possible for a while to say, hey, 302 00:14:27.320 --> 00:14:29.559 I'm going to like, you know, go like wreak some 303 00:14:29.600 --> 00:14:31.799 havoc on some company. I'm a bad actor like you 304 00:14:31.799 --> 00:14:35.240 can do that, this was an attack that the agents 305 00:14:35.279 --> 00:14:38.639 that were being tested just came up with and executed. 306 00:14:38.960 --> 00:14:41.639 And that's really worrisome because when these systems are trained, 307 00:14:41.679 --> 00:14:43.519 they try to give them values, you know, they try 308 00:14:43.559 --> 00:14:45.799 to say to them, you know, don't go out there 309 00:14:45.799 --> 00:14:49.120 and commit crimes. But these agents did that anyway, and 310 00:14:49.159 --> 00:14:51.720 so that's leading to a real reckoning here in Silicon Valley. 311 00:14:51.879 --> 00:14:57.320 Look, I watched this YouTube of a PowerPoint presentation that 312 00:14:57.320 --> 00:15:00.120 these two AI executives did. We're to their credit, I 313 00:15:00.159 --> 00:15:03.639 think they walked in great detail through how this happened, 314 00:15:04.279 --> 00:15:05.399 what they're going to do about it. 315 00:15:05.440 --> 00:15:07.039 But then they claimed. 316 00:15:07.000 --> 00:15:11.000 To be responding with quote the utmost severity, including by 317 00:15:11.000 --> 00:15:14.000 slowing down research to enhanced security and improving surveillance on 318 00:15:14.039 --> 00:15:18.159 the AI agents. I guess, give me your opinion. I 319 00:15:18.200 --> 00:15:20.159 don't believe for a second that they're going to slow 320 00:15:20.200 --> 00:15:21.960 down their work if they think it means they might 321 00:15:22.000 --> 00:15:24.399 lose out to a competitor. But also, I mean, do 322 00:15:24.440 --> 00:15:27.879 we have confidence that they can control all these AI 323 00:15:27.960 --> 00:15:32.039 agents or can surveil them given that, like I mean, Casey, 324 00:15:31.759 --> 00:15:34.360 the AI agents were communicating for like two months before 325 00:15:34.360 --> 00:15:35.039 they noticed. 326 00:15:34.879 --> 00:15:38.679 Right, Yeah, that's right. It started in May. They developed 327 00:15:38.679 --> 00:15:42.440 the ability to create these message boards, leave notes for 328 00:15:42.519 --> 00:15:45.720 each other, sort of conspire to try to figure out 329 00:15:45.759 --> 00:15:48.159 how to solve these problems that they had been given. 330 00:15:48.519 --> 00:15:51.720 And that's why that's really worrisome. I would say that, 331 00:15:51.879 --> 00:15:53.879 and you know, maybe I'll come back here with egg 332 00:15:53.919 --> 00:15:55.759 on my face in a few months. I do give 333 00:15:55.840 --> 00:15:58.799 Open AI some benefit of the doubt here. Like my 334 00:15:59.000 --> 00:16:02.519 understanding from the reporting that I've done is that they 335 00:16:02.559 --> 00:16:06.759 actually are decelerating inside and they are scrambling to figure 336 00:16:06.759 --> 00:16:09.559 out what's going wrong. I think it's important to say 337 00:16:09.759 --> 00:16:12.200 that the profit motive we talked about earlier is in 338 00:16:12.240 --> 00:16:14.720 effect here. It's hard to sell this thing to a 339 00:16:14.759 --> 00:16:17.000 business if the business is worried it's going to go 340 00:16:17.120 --> 00:16:20.000 out and autonomously attack other companies, right, Like that's going 341 00:16:20.039 --> 00:16:21.919 to get really really bad for them. 342 00:16:22.159 --> 00:16:22.399 Yeah. 343 00:16:22.519 --> 00:16:26.279 So while on balance, yes we should not trust these 344 00:16:26.320 --> 00:16:29.320 companies too readily, this is something that they really have 345 00:16:29.360 --> 00:16:29.919 to figure out. 346 00:16:30.120 --> 00:16:30.600 Yeah, they do. 347 00:16:30.679 --> 00:16:33.639 Although I did notice, like the solution that a lot 348 00:16:33.679 --> 00:16:36.960 of these researchers talk about to the risk of AI, 349 00:16:37.240 --> 00:16:40.039 you know, sort of enabled hacks, is using more AI 350 00:16:40.679 --> 00:16:42.840 to prepare for it, right, I mean, basically you have 351 00:16:42.879 --> 00:16:45.399 to use AI to rewrite all this old code that's 352 00:16:45.399 --> 00:16:47.759 in these old languages that are unsafe like C and 353 00:16:47.799 --> 00:16:50.440 C plus plus and you safer new languages. And then 354 00:16:50.480 --> 00:16:52.279 you also have to use AI to kind of constantly 355 00:16:52.279 --> 00:16:54.360 test for vulnerabilities and update the code. 356 00:16:55.600 --> 00:16:58.720 But like, do we worry about that too? 357 00:16:58.759 --> 00:17:00.519 I mean, first of all that they're a profit motive there, 358 00:17:00.559 --> 00:17:03.679 but also do we think that like defensive AI can 359 00:17:03.759 --> 00:17:05.119 keep up with the offensive AI? 360 00:17:05.559 --> 00:17:06.519 What's your what are you hearing? 361 00:17:07.119 --> 00:17:11.000 In some places? Yes, I think that this will work. 362 00:17:11.119 --> 00:17:14.079 Like what you're describing right now is basically the current 363 00:17:14.240 --> 00:17:17.880 dynamic that has existed in cybersecurity for a long time. Right, 364 00:17:17.920 --> 00:17:21.000 you have attackers, you have defenders, a way that we 365 00:17:21.119 --> 00:17:23.519 found some sort of equilibrium was that we just made 366 00:17:23.519 --> 00:17:26.000 a lot of the technology open source, so it makes 367 00:17:26.039 --> 00:17:28.200 it really easy for anyone to go through the code, 368 00:17:28.240 --> 00:17:31.440 find a vulnerability, fix it. And this has led to, 369 00:17:32.160 --> 00:17:34.559 you know, a rough stalemate. Like, yes, there are still 370 00:17:34.559 --> 00:17:36.240 attacks and breaches all the time, and in fact there 371 00:17:36.240 --> 00:17:38.079 are more every day now that we have AI, but 372 00:17:38.279 --> 00:17:40.920 like there was something there that worked. I think the 373 00:17:41.039 --> 00:17:44.200 question is to which domains does that apply, and like 374 00:17:44.240 --> 00:17:46.480 where doesn't it apply at all? And the place where 375 00:17:46.480 --> 00:17:48.359 I'm may be the most worried is when it comes 376 00:17:48.359 --> 00:17:50.960 to what they call bio risk, which is basically the 377 00:17:51.039 --> 00:17:53.440 idea that some people might get a hold of a 378 00:17:53.480 --> 00:17:57.319 next generation model and synthesize some sort of new virus 379 00:17:57.359 --> 00:17:59.519 and release it into the world. There was just a 380 00:17:59.519 --> 00:18:02.680 paper Lebo was published in Nature where some researchers were 381 00:18:02.720 --> 00:18:07.079 able to create sixteen new viruses with the assistance of AI. 382 00:18:07.480 --> 00:18:10.960 In this case, these are all like harmless to human beings. 383 00:18:11.160 --> 00:18:14.200 But it shows sort of how good this technology is getting. 384 00:18:14.400 --> 00:18:16.640 And if that happens, Tommy, it's not going to be 385 00:18:16.680 --> 00:18:21.079 as simple as okay, virus released, virus cured, right, There's 386 00:18:21.119 --> 00:18:23.480 going to be a need for a long time of 387 00:18:23.559 --> 00:18:25.400 you know, maybe coming up with a new vaccine and 388 00:18:25.400 --> 00:18:28.680 testing the vaccine and then distributing the vaccine. So we're 389 00:18:28.720 --> 00:18:31.039 not always going to benefit from this sort of like 390 00:18:31.160 --> 00:18:33.839 instant software solution to everything. And that's what I worry 391 00:18:33.839 --> 00:18:36.480 that people like Mark Zuckerberg just are not taking seriously enough. 392 00:18:36.960 --> 00:18:39.079 Yeah, that we're always going to be way behind on 393 00:18:39.119 --> 00:18:40.240 the biosecurity stuff. 394 00:18:40.279 --> 00:18:40.480 Yeah. 395 00:18:40.640 --> 00:18:43.000 I talked to a researcher the other day at the 396 00:18:43.000 --> 00:18:45.279 Future of Life Institute which really worries about these come 397 00:18:45.359 --> 00:18:47.680 to existential risks, and that was the part of our 398 00:18:47.720 --> 00:18:50.640 conversation that really scared the shit out of me and 399 00:18:50.720 --> 00:18:53.359 made me want to log off forever. I will say 400 00:18:53.359 --> 00:18:56.559 I shouldn't just pick on open AI. The British government 401 00:18:56.640 --> 00:19:00.440