1 00:00:01,560 --> 00:00:03,720 Welcome to the Cerebral Valley Podcast. 2 00:00:03,760 --> 00:00:06,600 I'm Eric Newcomer. We're calling this fear 3 00:00:06,600 --> 00:00:09,760 mongering and forecasting. We're looking to the future. 4 00:00:09,760 --> 00:00:12,800 We're making predictions, or really I think we're going to be 5 00:00:12,800 --> 00:00:17,360 relying on AI and chat bots to make our predictions and then we 6 00:00:17,360 --> 00:00:20,000 will shit all over them. I'm here with Max Child. 7 00:00:20,240 --> 00:00:22,040 Hey, Eric. And James Wilserman. 8 00:00:22,560 --> 00:00:26,200 Hey, hey guys. Ready to make some hot take 9 00:00:26,200 --> 00:00:29,600 predictions here? We're gonna use both Open AI, 10 00:00:29,640 --> 00:00:33,840 Chaji, BT and Anthropic, so we'll also get sort of the meta 11 00:00:34,240 --> 00:00:37,800 sense of who's which bot is making better predictions. 12 00:00:38,080 --> 00:00:42,280 But yeah, excited what? What is AI if not constantly 13 00:00:42,280 --> 00:00:46,160 looking to the future and charting out what could be given 14 00:00:46,160 --> 00:00:48,520 the rate of growth we've seen so far? 15 00:00:48,520 --> 00:00:52,440 So it's very fitting that that in our last episode before this 16 00:00:52,440 --> 00:00:56,360 Ruebel Valley AI Summit on November 20th in San Francisco, 17 00:00:56,760 --> 00:01:02,080 we are going to fiendishly fantasize about what could be to 18 00:01:02,080 --> 00:01:05,120 come. This episode is presented by 19 00:01:05,120 --> 00:01:09,080 Breck's The Financial Stack that founders and VCs can truly bank 20 00:01:09,080 --> 00:01:11,280 on. Imagine what your founders could 21 00:01:11,280 --> 00:01:14,560 do with their runway if they had a banking solution that had no 22 00:01:14,560 --> 00:01:19,040 minimums, no transaction fees, and 20 times the standard FDIC 23 00:01:19,040 --> 00:01:21,880 protection. Plus, they could earn an 24 00:01:21,920 --> 00:01:25,120 industry leading yield while maintaining access to funds 25 00:01:25,120 --> 00:01:27,880 whenever needed. Brex simplifies financial 26 00:01:27,880 --> 00:01:30,960 services for startups so they can focus on building. 27 00:01:31,120 --> 00:01:33,880 Connect your portfolio to the financial stack that one in 28 00:01:33,880 --> 00:01:36,640 three US venture backed startups already use. 29 00:01:37,000 --> 00:01:42,200 Check out brex.com/banking Dash solutions. 30 00:01:43,080 --> 00:01:47,920 Without further ado, James. So for this episode I asked both 31 00:01:47,920 --> 00:01:53,080 ChatGPT and Claude the same question to come up with a list 32 00:01:53,080 --> 00:01:55,960 of predictions about AI for the year 2025. 33 00:01:56,360 --> 00:02:00,080 I also asked the predictions to be hyper specific and 34 00:02:00,080 --> 00:02:04,560 falsifiable so that by the end of 2025 we could easily 35 00:02:04,560 --> 00:02:06,600 determine whether the predictions were correct or 36 00:02:06,600 --> 00:02:08,520 incorrect. The predictions I said could be 37 00:02:08,520 --> 00:02:13,600 spicy, medium, mild. However, however hot Thai spicy. 38 00:02:14,400 --> 00:02:17,840 I didn't get any pepper emojis out on the on the output. 39 00:02:17,880 --> 00:02:22,680 Maybe I should ask for that. But the, the, the predictions I 40 00:02:22,680 --> 00:02:26,840 was, I was pretty happy with how they came out and I asked, you 41 00:02:26,840 --> 00:02:30,160 know, relevant to the AI industry and then I also asked 42 00:02:30,160 --> 00:02:32,520 to get a probability estimate for each prediction. 43 00:02:32,520 --> 00:02:36,440 So then you guys are going to be able to provide me kind of the 44 00:02:36,440 --> 00:02:38,600 over undertake on each prediction. 45 00:02:38,960 --> 00:02:41,400 So we have our own prediction market basically. 46 00:02:41,400 --> 00:02:43,400 Exactly. We basically created a bunch of 47 00:02:43,760 --> 00:02:48,760 possible prediction markets around AI and you guys have to 48 00:02:48,760 --> 00:02:51,560 decide whether you're short or long each prediction. 49 00:02:52,040 --> 00:02:52,960 Nice, how's that? I love it. 50 00:02:53,200 --> 00:02:56,840 Is this is this GPT 01 or is this? 51 00:02:56,880 --> 00:03:04,040 Yeah, this is the ChatGPT 01 preview and Claude Sonnet 3.5. 52 00:03:04,400 --> 00:03:07,880 OK, so the the most thinking oriented AI models. 53 00:03:07,920 --> 00:03:11,600 We're paying the big bucks here. We're premium subscribers for. 54 00:03:12,640 --> 00:03:19,880 We're we're bankrupt now they're running, but no, these these are 55 00:03:19,880 --> 00:03:22,760 the best AI can do in a consumer consumer product. 56 00:03:22,760 --> 00:03:25,320 It it feels if I was pretty impressed, like I was kind of 57 00:03:25,320 --> 00:03:27,440 skeptical actually, that this would work that well. 58 00:03:27,440 --> 00:03:32,600 You know, I think AI has a tendency to provide sort of 59 00:03:32,600 --> 00:03:37,520 generic takes in a lot of ways, but if you prompt it correctly, 60 00:03:37,520 --> 00:03:40,920 I was able to get these really specific, really specific 61 00:03:40,920 --> 00:03:43,800 predictions that. They could keep this is for 12 62 00:03:43,800 --> 00:03:45,480 months from now. Like what's the date? 63 00:03:45,600 --> 00:03:48,680 You have the year 25. Yeah, by the end of 2025, 64 00:03:48,680 --> 00:03:51,120 December 31st. OK, cool. 65 00:03:51,120 --> 00:03:55,800 All right, let's do it. All right, we'll start with O1 66 00:03:55,800 --> 00:03:58,400 preview from Open AI in Chacha BT. 67 00:03:58,400 --> 00:04:04,800 So here's the first prediction So. 68 00:04:05,960 --> 00:04:08,280 We have not heard these before. To be clear, Max and I are 69 00:04:08,280 --> 00:04:09,960 coming in blind. Unless. 70 00:04:10,000 --> 00:04:11,840 You guys, you guys are Co founders so maybe you're in 71 00:04:12,000 --> 00:04:14,240 cahoots. Nope, I got nothing running 72 00:04:14,280 --> 00:04:17,440 blind. All right, so first prediction, 73 00:04:17,959 --> 00:04:24,760 Open AI will release ChatGPT or GBT 5 exceeding 10 trillion 74 00:04:24,760 --> 00:04:28,240 parameters. What do you guys think? 75 00:04:28,440 --> 00:04:31,040 How many parameters is 4? Again, one trillion. 76 00:04:31,160 --> 00:04:32,720 The bots are already smarter than me. 77 00:04:32,720 --> 00:04:34,600 What is that? You're like what's a parameter 78 00:04:34,600 --> 00:04:36,880 isn't. Isn't it like 400 billion? 79 00:04:36,880 --> 00:04:40,240 Can you can we get the number of the current it's like 4 or 500 80 00:04:40,240 --> 00:04:42,200 billion, right? 1.5 trillion. 81 00:04:42,200 --> 00:04:45,400 OK, let's add the. Elevator .76 trillion Is the 82 00:04:45,400 --> 00:04:46,600 rumours OK? 1. 83 00:04:47,280 --> 00:04:49,800 One to two, OK. So this would be, you know. 84 00:04:50,200 --> 00:04:51,920 What's it predicting to rate how many? 85 00:04:51,960 --> 00:04:55,240 A 10X like like A7X basically or whatever. 86 00:04:55,240 --> 00:04:57,640 And we think the current is 1.7 something. 87 00:04:57,880 --> 00:05:00,760 Yeah. Also the the probability for 88 00:05:00,760 --> 00:05:05,560 this prediction is 50% so. Oh, over smash the over on that. 89 00:05:05,560 --> 00:05:08,560 Wait, wait, wait, wait. Sorry, it's is it making a claim 90 00:05:08,560 --> 00:05:11,520 about the name of it? Because I feel like I, I think I 91 00:05:11,520 --> 00:05:15,560 was just listening to Dario on Lex Friedman and like, I feel 92 00:05:15,560 --> 00:05:18,640 like half of his, like Hemming and hawing about what they're 93 00:05:18,640 --> 00:05:22,120 gonna do is like not committing to the name of various models 94 00:05:22,120 --> 00:05:24,520 because you know, they're doing all these like half updates and 95 00:05:24,520 --> 00:05:26,080 stuff. So how much is this a prediction 96 00:05:26,080 --> 00:05:28,240 about what the model is going to be called? 97 00:05:28,880 --> 00:05:33,440 I think we can decide there, but I would say both have to come 98 00:05:33,440 --> 00:05:36,040 true. The GPT 5 naming and the 10 99 00:05:36,040 --> 00:05:38,040 trillion. I think, you know, if I'm 100 00:05:38,240 --> 00:05:41,680 reading the tea leaves of what open what 01 was thinking and 101 00:05:41,680 --> 00:05:44,640 reasoning about here, maybe they're saying, hey, 10 trillion 102 00:05:44,640 --> 00:05:47,880 is a big step up to the point where it would have to almost be 103 00:05:47,880 --> 00:05:50,960 called, you know, GBT. Five, I mean, I don't know, this 104 00:05:50,960 --> 00:05:54,240 is kind of pedantic, but I would smash the over on a model that 105 00:05:54,240 --> 00:05:56,720 it has 10 trillion parameters, but I don't know if they're 106 00:05:56,720 --> 00:06:00,360 gonna call GBT 5 exactly. I feel like Altman has been you. 107 00:06:00,360 --> 00:06:02,200 Committed to over already, I think the. 108 00:06:02,400 --> 00:06:03,280 Name. OK, fine, I'll commit. 109 00:06:03,280 --> 00:06:06,640 Why are you guys so skeptical that it's gonna be called GBT 5? 110 00:06:06,920 --> 00:06:09,760 I'm just chaotic on the names. The names are bad at the moment. 111 00:06:10,040 --> 00:06:15,320 Like a circle that we get you introduced as like open AI chat 112 00:06:15,320 --> 00:06:19,640 GBTO one like they they need to like I I do I mean anthropic has 113 00:06:19,640 --> 00:06:23,640 more colorful names, but they're getting chaotic sure, but so I'm 114 00:06:23,840 --> 00:06:26,560 just take the under on the basis of the name in terms of the 115 00:06:26,560 --> 00:06:32,640 parameters yeah, I'm probably bullish. 116 00:06:32,640 --> 00:06:35,880 I mean, thankfully it's a compound prediction, but yeah, 117 00:06:36,000 --> 00:06:37,880 the under. OK, All right, Well, I will 118 00:06:37,880 --> 00:06:40,080 stick with my over, but yeah, I mean, I think they're basically 119 00:06:40,080 --> 00:06:45,160 running into like the iPhone 4S5 S 6S problem where Apple used to 120 00:06:45,160 --> 00:06:48,400 give the like in between phones, like, you know, an S instead of 121 00:06:48,400 --> 00:06:50,280 a new number. And then I think the marketing 122 00:06:50,280 --> 00:06:53,000 team realized like people want to see the number go up. 123 00:06:53,000 --> 00:06:56,200 Like we don't see, we don't we don't buy new iPhones unless the 124 00:06:56,200 --> 00:07:00,000 number goes from 14 to 15. So I think, I think This is why 125 00:07:00,000 --> 00:07:02,840 GBT is trying to move off these numbers because I think they're 126 00:07:02,840 --> 00:07:05,400 like, nobody wants the new model unless we give it a higher 127 00:07:05,400 --> 00:07:06,600 number, right? And they're like, they don't 128 00:07:06,600 --> 00:07:10,000 even care if we introduce like 4 O and voice mode and you know, 129 00:07:10,000 --> 00:07:12,840 O1 and all this stuff. So I just, I feel like Altman is 130 00:07:12,840 --> 00:07:15,520 severely foreshadowing that they're gonna drop these 131 00:07:15,520 --> 00:07:17,920 numbers. I do think you're predicting. 132 00:07:17,920 --> 00:07:20,960 Yourself. No, I'm taking the over. 50% is 133 00:07:20,960 --> 00:07:23,520 a good price. I just think I understand where 134 00:07:23,520 --> 00:07:24,520 they're coming from with the names. 135 00:07:24,520 --> 00:07:26,000 James, are you making predictions? 136 00:07:26,440 --> 00:07:28,080 You'll give us a quick your quick take. 137 00:07:29,120 --> 00:07:30,720 Yeah, I'll just react to your guys takes. 138 00:07:31,800 --> 00:07:34,960 I think that yeah, you guys are way too focused on the naming 139 00:07:34,960 --> 00:07:38,720 conventions. I think the reason they are 140 00:07:38,720 --> 00:07:43,080 calling things like, you know, 40 and 01 preview is to try to 141 00:07:43,080 --> 00:07:47,600 like, you know, downplay the model a little bit before the 142 00:07:47,600 --> 00:07:50,120 next evolution of the model. Like I think they will stick 143 00:07:50,120 --> 00:07:53,800 with the GPT 456 naming convention, but they will need 144 00:07:53,800 --> 00:07:58,280 it to be kind of step change, you know, improvement on the 145 00:07:58,280 --> 00:08:00,120 model. So I do think A10 trillion 146 00:08:00,120 --> 00:08:01,480 parameter model will get that name. 147 00:08:01,640 --> 00:08:03,960 But there's an open question. I mean, this is gonna be 148 00:08:03,960 --> 00:08:07,760 probably coming up throughout, just like how monumental is A10 149 00:08:07,760 --> 00:08:10,240 trillion parameter model change, right? 150 00:08:10,240 --> 00:08:14,320 Like if it's true if they have consumed basically all the data 151 00:08:14,320 --> 00:08:18,400 there is to consume and set that data has its limits. 152 00:08:19,000 --> 00:08:20,720 That's fair. Then if there's not a step 153 00:08:20,720 --> 00:08:23,000 change in the model capabilities, maybe they don't 154 00:08:23,000 --> 00:08:25,280 give the new name. OK, all right, let's move on. 155 00:08:25,800 --> 00:08:27,280 Good, good first discussion there. 156 00:08:27,600 --> 00:08:29,680 Are you over or under though? I never got that part. 157 00:08:30,720 --> 00:08:33,080 I'm over, OK? I mean if you want me to make 158 00:08:33,080 --> 00:08:34,760 make my own predictions. You've got to come in. 159 00:08:34,760 --> 00:08:35,640 I just. Say I just put. 160 00:08:35,640 --> 00:08:37,200 I'll say where you are. I'll take the over. 161 00:08:37,679 --> 00:08:45,800 All right, so next prediction from Chachi BT we have by the 162 00:08:45,840 --> 00:08:50,360 end of 2025, at least three countries will have implemented 163 00:08:50,760 --> 00:08:55,200 regulations at the national level specifically governing the 164 00:08:55,200 --> 00:08:59,400 use of AI to diagnose healthcare disease. 165 00:09:00,600 --> 00:09:04,200 And what's the price on that? 70% probability. 166 00:09:05,160 --> 00:09:07,160 Short. Way short. 167 00:09:09,280 --> 00:09:11,840 I just think it's really, it takes a long time to write 168 00:09:11,840 --> 00:09:13,600 legislation. I mean, you're saying like 169 00:09:13,960 --> 00:09:16,280 national level entities are going to write complex 170 00:09:16,280 --> 00:09:20,000 legislation about healthcare and AI in the next 14 months. 171 00:09:20,520 --> 00:09:22,680 Yeah, yeah, it's. Interesting, though I'll grab. 172 00:09:22,720 --> 00:09:24,720 Go ahead. I'll let Eric answer, but you're 173 00:09:24,720 --> 00:09:27,800 just I'm. Taking the over, I mean, I don't 174 00:09:27,800 --> 00:09:31,400 want to just do the opposite of Max every time here, but I do 175 00:09:31,520 --> 00:09:32,720 That's good. There are a lot of countries. 176 00:09:32,760 --> 00:09:36,600 There are a lot of countries we've actually seen aggressive 177 00:09:36,720 --> 00:09:39,080 AI regulation already. There's clearly a lot of 178 00:09:39,080 --> 00:09:40,920 appetite. And I don't think this 179 00:09:40,920 --> 00:09:43,720 prediction requires some sort of comprehensive, they just need 180 00:09:43,960 --> 00:09:48,640 some regulation touching sort of health, health diagnostics, 181 00:09:48,640 --> 00:09:50,160 right? Yeah, I, I guess. 182 00:09:50,320 --> 00:09:54,440 I guess the question is like, is some random EU regulator issuing 183 00:09:54,440 --> 00:09:56,680 a decree count as 35? Countries. 184 00:09:56,680 --> 00:09:58,320 Yeah, exactly. I think it should like I think 185 00:09:58,320 --> 00:09:59,000 it. Should. 186 00:09:59,000 --> 00:10:00,360 Well, yeah, of course you think it should. 187 00:10:01,440 --> 00:10:04,280 Like, yeah, but I'm just saying like, you know, I'm this. 188 00:10:04,320 --> 00:10:07,640 Episode isn't gonna be about the nature of predictions more than 189 00:10:07,640 --> 00:10:10,760 this underlying. Game. 190 00:10:11,720 --> 00:10:15,120 You know, for every good game, we've realized the rules are 191 00:10:15,120 --> 00:10:19,840 really what's most important to discuss, not the actual OK the 192 00:10:19,880 --> 00:10:21,880 gameplay. All right, well, I'm still I'm 193 00:10:21,880 --> 00:10:25,000 still grabbing the under here. But yeah, that just seems like a 194 00:10:25,000 --> 00:10:27,840 lot of work for three different nation states to execute and 195 00:10:27,840 --> 00:10:29,600 then in the 12 month time frame would be mine. 196 00:10:29,840 --> 00:10:33,080 Are we getting more point like or so like does he get more 197 00:10:33,080 --> 00:10:35,680 return on his investment for predicting something that was at 198 00:10:35,680 --> 00:10:37,200 70? Yeah, that's a good idea. 199 00:10:37,200 --> 00:10:39,400 I think that we could use something like that. 200 00:10:39,880 --> 00:10:43,760 If you take the over on a low probability prediction and then 201 00:10:43,760 --> 00:10:46,520 at the end of next year you're proven correct, I think we'll 202 00:10:46,520 --> 00:10:47,840 give you give you some more points. 203 00:10:48,080 --> 00:10:51,240 I think we need. ROI like let's let's let's save 204 00:10:51,240 --> 00:10:53,360 the exact one. Right, exactly. 205 00:10:54,600 --> 00:10:55,920 Could we get into another rule here? 206 00:10:55,920 --> 00:11:00,960 Yeah. James, what are you picking? 207 00:11:01,920 --> 00:11:05,040 OK, I am taking the over there and I think, you know, we're 208 00:11:05,040 --> 00:11:08,200 just going to see this come up a lot in the next year is like 209 00:11:08,200 --> 00:11:11,920 people using AI for their jobs, maybe in areas where they're not 210 00:11:11,920 --> 00:11:13,880 supposed to. And I just think that'll be a 211 00:11:13,880 --> 00:11:15,920 key kind of political discussion. 212 00:11:15,920 --> 00:11:20,120 Point is, hey, why is why is my doctor, you know, using ChatGPT, 213 00:11:20,560 --> 00:11:22,240 right? And is that a should that be 214 00:11:22,240 --> 00:11:23,880 legal? So I wouldn't be surprised by 215 00:11:23,880 --> 00:11:25,960 three countries. Taking I want them to be to be 216 00:11:25,960 --> 00:11:28,240 clear or like as an assistant. I think it would be good. 217 00:11:28,480 --> 00:11:30,880 I'm not saying that's a bad thing, but maybe it should be a 218 00:11:30,880 --> 00:11:32,800 little bit regulated, right? There should be something. 219 00:11:33,160 --> 00:11:35,600 I might just, I might just be biased based on living in the 220 00:11:35,600 --> 00:11:38,440 United States where we have just epic legislative gridlock. 221 00:11:38,440 --> 00:11:42,200 But yeah, I mean they. Passed AI regulations in 222 00:11:42,200 --> 00:11:44,600 California. Well, that's true. 223 00:11:45,560 --> 00:11:46,680 I mean, you're. Not the big one. 224 00:11:46,880 --> 00:11:48,960 But you're saying the state, the state government, Yeah. 225 00:11:48,960 --> 00:11:50,120 Yeah, which is like one of. The. 226 00:11:50,120 --> 00:11:51,520 Well, they didn't pass it because the. 227 00:11:51,880 --> 00:11:54,120 They passed other ones, they passed it all there other. 228 00:11:54,640 --> 00:11:57,440 Things OK, this prediction specifically says national, 229 00:11:57,440 --> 00:11:59,160 national. Yes, wouldn't count. 230 00:11:59,400 --> 00:12:01,000 Yeah, all right. But Europe? 231 00:12:01,000 --> 00:12:05,640 'S like anyway, sorry. Moving on prediction, by 232 00:12:05,640 --> 00:12:11,000 December 2025, Tesla will release fully self driving cars 233 00:12:11,320 --> 00:12:16,080 that are legally approved for unsupervised operation on a on 234 00:12:16,080 --> 00:12:19,160 public roads in at least one US state. 235 00:12:21,960 --> 00:12:27,320 What's the price? We are saying 40% probabilities. 236 00:12:28,600 --> 00:12:33,160 Interesting. I'll go over actually I one of 237 00:12:33,160 --> 00:12:36,440 my predictions coming in was around Tesla AV capabilities. 238 00:12:38,320 --> 00:12:41,840 I just think that Waymo's proved it's possible. 239 00:12:42,120 --> 00:12:43,840 Elon's got more chips than anyone. 240 00:12:43,840 --> 00:12:45,560 They've got more video data than anyone. 241 00:12:45,880 --> 00:12:47,120 They're obsessed with this problem. 242 00:12:47,120 --> 00:12:48,680 He thinks it's the entire future of the company. 243 00:12:48,920 --> 00:12:52,280 And then coming down to the ticky tacky state approval bit, 244 00:12:54,160 --> 00:12:56,920 you know, he's got his boy in the White House and he's got 245 00:12:56,920 --> 00:12:59,160 friends in every Republican government in America. 246 00:12:59,160 --> 00:13:02,600 So like fucking Wyoming could approve this shit and he win the 247 00:13:02,600 --> 00:13:05,960 bat. So I think that I'll take the 248 00:13:05,960 --> 00:13:10,440 over at 40%. So it it's to be driving like 249 00:13:10,440 --> 00:13:12,280 without a driver behind the wheel. 250 00:13:13,200 --> 00:13:16,000 Yeah, that's I think unsupervised, meaning you don't. 251 00:13:16,000 --> 00:13:18,680 Have to I still, I still think there's a very likely scenario 252 00:13:18,680 --> 00:13:24,800 where it's like in Texas or Wyoming or whatever you, you 253 00:13:24,800 --> 00:13:27,520 know, you can sort of allow it to drive, but people are 254 00:13:27,520 --> 00:13:31,080 effectively behind the wheel. Like I'm not very bullish on 255 00:13:31,080 --> 00:13:33,160 Tesla's like technology, I don't think. 256 00:13:33,160 --> 00:13:35,880 Are you talking about like remote assistance, like remote 257 00:13:35,880 --> 00:13:39,000 drivers taking over or something or like I think? 258 00:13:39,000 --> 00:13:41,800 That would be that would count or not I feel like. 259 00:13:42,240 --> 00:13:43,800 That would count like that's what Waymo does. 260 00:13:43,800 --> 00:13:46,120 Right, Waymo has remote takeover if there's. 261 00:13:46,120 --> 00:13:47,680 No takeover. Remote takeover, yeah. 262 00:13:48,120 --> 00:13:50,320 But it's a little. It's like a remote. 263 00:13:51,000 --> 00:13:53,960 How is what they have right now not already passing that? 264 00:13:53,960 --> 00:13:56,040 Like people are allowed to use autopilot. 265 00:13:56,120 --> 00:13:58,080 You're talking about Tesla. Tesla you. 266 00:13:58,080 --> 00:14:03,080 From a legal perspective, you have to have a driver in the car 267 00:14:03,360 --> 00:14:05,800 behind the wheel, ready to take over at a moment's notice. 268 00:14:05,800 --> 00:14:08,560 Basically it also has to be supervised, like you know that 269 00:14:08,560 --> 00:14:11,320 you have to be looking at the road or disengage. 270 00:14:11,320 --> 00:14:15,040 Like they have a camera on your eyes that show that checks if 271 00:14:15,040 --> 00:14:17,960 you're watching the road or not and it gets mad at you if you're 272 00:14:17,960 --> 00:14:19,840 like on your phone. I've tried this. 273 00:14:21,120 --> 00:14:25,080 Oh, wow, yeah, yeah, it it literally throws a warning up on 274 00:14:25,080 --> 00:14:27,560 the screen saying I like, I can tell you're looking at your 275 00:14:27,560 --> 00:14:29,160 phone. And I'm like, well, what's the 276 00:14:29,160 --> 00:14:31,120 point of self driving if I can't look at my phone? 277 00:14:32,200 --> 00:14:39,880 Like that's the entire use case. I think I'm I'm taking the over 278 00:14:39,880 --> 00:14:43,800 just because of all this sort of seeming cases where this would 279 00:14:43,800 --> 00:14:46,760 count as approved. I don't think they're close to 280 00:14:46,760 --> 00:14:49,280 Waymo. I'm not bullish on Tesla's 281 00:14:49,280 --> 00:14:53,400 approach, but I think there are lots of ways to sort of fudge 282 00:14:53,400 --> 00:14:56,560 this with humans still sort of behind the wheel. 283 00:14:57,080 --> 00:15:00,280 All right, for bonus points, which is the 1st state. 284 00:15:00,720 --> 00:15:04,400 That will allow this Texas. Texas would be the obvious 285 00:15:04,400 --> 00:15:07,840 choice, but it's also really big and there's a lot of complicated 286 00:15:07,840 --> 00:15:10,400 roads and cities and stuff there, so I don't. 287 00:15:10,400 --> 00:15:13,200 Know, but does it have to be the whole state for the rules or? 288 00:15:13,200 --> 00:15:15,200 Yeah, Yeah. It's a state government, right? 289 00:15:15,280 --> 00:15:16,680 Yeah. Wait the whole state. 290 00:15:17,600 --> 00:15:19,080 It says one state. Will approve it. 291 00:15:19,080 --> 00:15:21,040 That's interesting. You mean like Waymo is not in 292 00:15:21,040 --> 00:15:22,800 the whole state, it's in San Francisco, right? 293 00:15:23,080 --> 00:15:25,600 Or I think. I think it would just have to be 294 00:15:26,040 --> 00:15:29,680 not the whole state, but you know, somewhere within somewhere 295 00:15:29,680 --> 00:15:31,920 in the state. Yeah, like the whole state. 296 00:15:31,920 --> 00:15:33,760 I would change my whole. Yeah, yeah. 297 00:15:34,080 --> 00:15:36,040 I think it has certain. Things I guess, like I wasn't 298 00:15:36,040 --> 00:15:40,080 even thinking you'd be able to own a Tesla but only use it for 299 00:15:40,080 --> 00:15:43,040 full self driving if you're in one city. 300 00:15:43,360 --> 00:15:46,440 Yeah, that's interesting. All right, Texas. 301 00:15:46,600 --> 00:15:48,120 All right, We'll see how that plays out. 302 00:15:49,200 --> 00:15:53,520 James. Yeah, my take taking the over as 303 00:15:53,520 --> 00:15:54,880 well. I think you guys are, you guys 304 00:15:54,880 --> 00:15:58,960 are spot on. And yeah, I think I will 305 00:15:58,960 --> 00:16:01,920 actually go a little contrarian and take California as the first 306 00:16:02,200 --> 00:16:04,240 thing. Wow, OK, wow all. 307 00:16:04,800 --> 00:16:08,680 Right guys, let's move on to the next prediction we have and 308 00:16:08,680 --> 00:16:13,880 another prediction from ChatGPT. By the end of 2025, AI generated 309 00:16:14,160 --> 00:16:20,200 content will constitute at least 50% of online news articles 310 00:16:20,200 --> 00:16:25,840 published by a major news outlet, probability 30%. 311 00:16:27,240 --> 00:16:28,960 What's major? Again, I think this is a key 312 00:16:28,960 --> 00:16:30,640 question here. Like how major? 313 00:16:30,640 --> 00:16:38,120 Mitigated after the fact. What would be capital M major? 314 00:16:38,120 --> 00:16:40,600 I mean like I'm. Watching the I'm taking the 315 00:16:40,640 --> 00:16:44,240 over, I think Conde Nast is clearly major. 316 00:16:44,520 --> 00:16:46,640 I think Forbes is probably major. 317 00:16:46,840 --> 00:16:50,320 And I think if you're just talking articles, you know, 318 00:16:50,480 --> 00:16:53,720 we've seen a lot of media just be willing to produce a lot of 319 00:16:53,800 --> 00:16:58,680 articles to, you know, get get gobbled up by SEO. 320 00:16:58,720 --> 00:17:02,880 And so it's not necessarily the majority of their homepage, but 321 00:17:02,880 --> 00:17:05,040 the majority of the articles. One outlet. 322 00:17:05,240 --> 00:17:08,240 I say yes. I think, yeah, I think it's 30. 323 00:17:08,240 --> 00:17:11,680 I'd grab the over. I just think 3 to one odds on 324 00:17:11,680 --> 00:17:13,480 that or two to one odds sounds pretty good. 325 00:17:13,599 --> 00:17:17,240 I think to your point, like yeah, the SB Nation network I 326 00:17:17,240 --> 00:17:21,880 could see or Vox or whatever, like just somebody is gonna give 327 00:17:21,880 --> 00:17:26,440 in to the SEO, the SEO temptation to have a crap load 328 00:17:26,440 --> 00:17:29,360 of articles generated by AI that people are going to link to. 329 00:17:29,360 --> 00:17:31,200 So will. They talk about that publicly. 330 00:17:31,200 --> 00:17:33,640 Do you think like will they? No, I'm sure there will be some 331 00:17:33,640 --> 00:17:36,680 journalist that gets leaked to in a bar where where Eric lives 332 00:17:36,680 --> 00:17:39,440 in Brooklyn and then then somebody will write an. 333 00:17:39,440 --> 00:17:42,080 Angry story about I don't think Axios would necessarily do it, 334 00:17:42,080 --> 00:17:45,920 but there's definitely a breed of sort of business first media 335 00:17:45,920 --> 00:17:50,120 publication that would brag about this to seem cutting edge. 336 00:17:50,320 --> 00:17:53,200 I bet someone will write a cranky article before somebody 337 00:17:53,200 --> 00:17:55,160 brags brags about it. But before. 338 00:17:55,680 --> 00:17:57,400 Yeah, yeah. We'll see. 339 00:17:57,760 --> 00:18:05,400 AI itself break the news that. Not until it hooks up to the 340 00:18:05,400 --> 00:18:07,360 Bitcoin network and starts getting paid for it. 341 00:18:09,360 --> 00:18:15,440 I'll take the under on 30% I'm sure have to litigate what 342 00:18:15,920 --> 00:18:21,480 content a major outlet, but I'm skeptical that it'll hit 50% of 343 00:18:21,480 --> 00:18:25,120 articles like I mean, at least in the next year. 344 00:18:25,120 --> 00:18:27,720 I think that I think that it'll maybe start, you know, kind of 345 00:18:27,720 --> 00:18:30,200 long form like some some indie Blogger or something. 346 00:18:30,200 --> 00:18:32,880 We'll be generating hundreds of thousands of articles or 347 00:18:32,880 --> 00:18:35,920 something with AI next year, but I'm skeptical that a major 348 00:18:35,920 --> 00:18:39,560 outlet will be doing that at that scale next year, but we'll 349 00:18:39,560 --> 00:18:40,480 see. I just like. 350 00:18:40,920 --> 00:18:43,000 Media outlets. I just go with the, yeah, the 351 00:18:43,000 --> 00:18:46,400 Logan Roy attitude here, which is that money wins. 352 00:18:46,440 --> 00:18:49,240 And there's a lot of private equity firms that own these now 353 00:18:49,240 --> 00:18:51,200 and. If they're not doing this, 354 00:18:51,200 --> 00:18:53,320 they'll they'll yeah, management will get fired. 355 00:18:53,320 --> 00:18:55,520 This is an amazing way to cost cut and it's also an amazing way 356 00:18:55,520 --> 00:18:58,920 to get links from Google search. So I just think it's going to be 357 00:18:58,920 --> 00:19:01,360 very tempting for a lot of these companies. 358 00:19:02,400 --> 00:19:08,760 All right, next prediction. By December 31st, 2025, the 359 00:19:08,760 --> 00:19:13,120 first AI generated feature length film will be released in 360 00:19:13,120 --> 00:19:18,000 theaters, with the script and visuals both entirely crafted by 361 00:19:18,000 --> 00:19:25,880 AI. 25% probability under. I'm crushing the under on that. 362 00:19:25,880 --> 00:19:29,400 I just think that you have to do a deal with theaters like get 363 00:19:29,400 --> 00:19:32,160 distribution and Hollywood's going to lose their effing minds 364 00:19:32,200 --> 00:19:34,000 if somebody tries to do this in the next year. 365 00:19:34,000 --> 00:19:38,040 So I just think the the Gears of entertainment don't turn that 366 00:19:38,040 --> 00:19:40,000 quickly. I'm not saying somebody won't 367 00:19:40,000 --> 00:19:43,160 generate a feature film by the end of 25 that plausibly could 368 00:19:43,160 --> 00:19:45,720 be in a theater, although I I would actually debate that as 369 00:19:45,720 --> 00:19:47,200 well. But I think that the odds that 370 00:19:47,200 --> 00:19:48,840 it's like actually released in theaters. 371 00:19:48,840 --> 00:19:50,640 I'm I'm just super under on that. 372 00:19:51,360 --> 00:19:54,320 They barely, they barely let Netflix into the Oscars. 373 00:19:54,760 --> 00:19:57,360 And your mistake is misunderstanding the 374 00:19:57,360 --> 00:20:01,760 entertainment industry. Even though you're you grew up, 375 00:20:01,960 --> 00:20:06,760 you grew up in LA area, but. The you can pay to play in 376 00:20:06,760 --> 00:20:09,600 theaters any day. Like there are all these sort of 377 00:20:09,600 --> 00:20:12,240 documents, there are all these films that sort of pay to get 378 00:20:12,240 --> 00:20:15,040 theatre distribution so that they account for stuff and you 379 00:20:15,040 --> 00:20:17,120 can absolutely do that. I'm taking the over. 380 00:20:17,920 --> 00:20:20,440 There are tons, tons of companies with an interest to 381 00:20:20,440 --> 00:20:24,240 have this achieved and somebody will just buy the theatre plays 382 00:20:24,480 --> 00:20:26,320 to make it so you. Have the like, buy the US 383 00:20:26,320 --> 00:20:29,160 Constitution angle here, which is somebody's just gonna fucking 384 00:20:29,160 --> 00:20:32,400 do it because it's gonna sound cool basically, and pay to play 385 00:20:32,400 --> 00:20:33,120 the theaters. OK? 386 00:20:33,600 --> 00:20:35,520 I mean, there's already, there's already a. 387 00:20:35,520 --> 00:20:40,440 Film that Sarah, my wife saw that you know, there are some 388 00:20:40,440 --> 00:20:44,040 boundaries like it was shot by humans, but then it reorders 389 00:20:44,560 --> 00:20:47,880 based on AI, you know, so that was already in theaters. 390 00:20:47,880 --> 00:20:49,560 So I think. This is a good, this is a good 391 00:20:49,560 --> 00:20:50,800 robust debate. I like that. 392 00:20:50,800 --> 00:20:53,760 You're like, you guys believe in the forces of change and I 393 00:20:53,760 --> 00:20:56,600 believe in the entrenched hating institutions. 394 00:20:56,720 --> 00:20:59,120 I haven't taken my. I haven't taken my. 395 00:20:59,160 --> 00:21:02,040 Well, I was referencing the the legislation discussion from 396 00:21:02,040 --> 00:21:04,600 earlier, but yeah, this is a meta point about how quickly the 397 00:21:04,600 --> 00:21:07,080 Gears of power turn. And you guys are like really 398 00:21:07,120 --> 00:21:08,360 quick. And I'm like, I don't know. 399 00:21:09,640 --> 00:21:12,080 So, all right, James, what do you think? 400 00:21:13,160 --> 00:21:15,400 I am also taking the under on this one. 401 00:21:15,600 --> 00:21:19,880 Yes, and not for any remote reason around theater 402 00:21:19,880 --> 00:21:22,800 distribution, but just purely from model capability. 403 00:21:22,800 --> 00:21:26,680 Like the specification of this prediction is that the script 404 00:21:26,680 --> 00:21:30,600 and visuals have to entirely be created by AII think that we are 405 00:21:30,600 --> 00:21:33,800 not there on visuals, we're barely there on scripting. 406 00:21:34,720 --> 00:21:36,800 I don't think we're there on scripting either, by the way. 407 00:21:36,800 --> 00:21:39,120 I haven't. Seen the script as you started 408 00:21:39,120 --> 00:21:42,680 the episode with is that there's a lot of control in models over 409 00:21:42,680 --> 00:21:45,040 prompting. And so it's like if they prompt, 410 00:21:45,040 --> 00:21:49,720 they could prompt a whole script into it and then output a script 411 00:21:49,920 --> 00:21:52,960 and say it was AI. And that I just think there's 412 00:21:52,960 --> 00:21:56,680 such a strong incentive for the result to be it was AI that like 413 00:21:56,680 --> 00:21:59,120 humans can put a lot of thought into it and get what they want. 414 00:21:59,120 --> 00:22:01,080 I don't think it'll be hard to do a script. 415 00:22:01,120 --> 00:22:04,680 I mean, it won't be a good script, but I do think it'll be 416 00:22:04,680 --> 00:22:07,080 hard to create a full feature length. 417 00:22:07,120 --> 00:22:08,760 Does it say it says feature length? 418 00:22:08,840 --> 00:22:12,760 I don't know that Probably 120 minutes or something of visuals 419 00:22:12,760 --> 00:22:17,240 like a that's extremely expensive to create, and also B, 420 00:22:17,240 --> 00:22:20,200 it's like very hard to get any sort of consistency of the 421 00:22:20,200 --> 00:22:22,480 characters or anything. I'm just skeptical of the 422 00:22:22,480 --> 00:22:23,160 visual. Aspect. 423 00:22:23,160 --> 00:22:26,600 I think our guys at Runway will personally make sure this 424 00:22:26,600 --> 00:22:28,320 happens. Yeah, probably the guys. 425 00:22:28,960 --> 00:22:31,400 It's probably the way Eric wins is at the runway. 426 00:22:31,800 --> 00:22:33,120 Our friends at Runway just bribe. 427 00:22:33,200 --> 00:22:34,560 Have a theater chain into doing right? 428 00:22:34,680 --> 00:22:36,600 Yeah. They listen to this episode, 429 00:22:36,840 --> 00:22:38,560 they listen to this episode and they're like. 430 00:22:39,560 --> 00:22:42,840 They're like Eric's got some skin in the game and we're gonna 431 00:22:42,840 --> 00:22:45,440 help him out here. We do this, Eric is gonna brag 432 00:22:45,440 --> 00:22:48,760 about us on the next. Year's yeah. 433 00:22:48,760 --> 00:22:50,840 I Episodes I. Don't know they haven't even 434 00:22:50,840 --> 00:22:54,040 released Sora you know yet or. It's coming soon. 435 00:22:54,200 --> 00:22:55,920 Somebody was just one of the competitors. 436 00:22:55,920 --> 00:22:58,400 I think Runway maybe was saying we think Sora is about to come, 437 00:22:58,600 --> 00:23:00,720 come all right. All right, good discussion. 438 00:23:00,720 --> 00:23:05,480 Let's switch over to Claude and see how anthropics predictions 439 00:23:05,600 --> 00:23:09,120 do and how you guys respond to the probabilities. 440 00:23:09,120 --> 00:23:12,160 And then we can have a little discussion at the end of who's 441 00:23:12,160 --> 00:23:17,200 better at predicting, but meta discussion but starting with. 442 00:23:17,200 --> 00:23:20,000 We're already having a discussion about our opinions on 443 00:23:20,000 --> 00:23:22,080 their predictions, and then we're going to have a higher 444 00:23:22,080 --> 00:23:25,360 level of discussion about our opinions of the relative 445 00:23:25,360 --> 00:23:27,480 predictive power of the two models. 446 00:23:27,760 --> 00:23:31,080 And then we're going to have AI watch this represent and react 447 00:23:31,080 --> 00:23:33,360 to our predictions. It's all good. 448 00:23:33,360 --> 00:23:35,960 It's all good. Well, I guess on a certain 449 00:23:35,960 --> 00:23:38,480 level, you know, think about AI is all about intelligence. 450 00:23:38,480 --> 00:23:40,640 So it's like, how do we introspect our introspection 451 00:23:40,640 --> 00:23:41,920 even more anyway? Sure. 452 00:23:42,200 --> 00:23:46,000 Yeah, somebody. Someday we'll be teaching a 453 00:23:46,000 --> 00:23:48,720 class in university. About this episode. 454 00:23:48,720 --> 00:23:51,320 All. Right, too self indulgent. 455 00:23:51,320 --> 00:23:55,760 All right, all right. Starting with Claude's first 456 00:23:55,760 --> 00:24:00,320 prediction, by December 31st, 2025, at least three major 457 00:24:00,320 --> 00:24:05,200 smartphone manufacturers among Apple, Samsung, Google and 458 00:24:05,200 --> 00:24:10,120 Xiaomi will release phones with dedicated AI Co processors 459 00:24:10,440 --> 00:24:14,680 capable of running large language models with at least 7 460 00:24:14,680 --> 00:24:20,120 billion parameters entirely on device probability 75%. 461 00:24:21,000 --> 00:24:23,160 And the parameters ones are where I'm like. 462 00:24:23,360 --> 00:24:25,440 I don't even know how many parameters they can run on 463 00:24:25,440 --> 00:24:27,960 device right now. Can't they run 7 billion locally 464 00:24:27,960 --> 00:24:29,720 already? Is that wrong? 465 00:24:30,360 --> 00:24:33,120 I think that the Apple intelligence chip can run 7 466 00:24:33,120 --> 00:24:34,000 billion, right? I thought. 467 00:24:34,160 --> 00:24:38,200 It's at least three to four, you know. 3 billion. 3 billion. 468 00:24:38,200 --> 00:24:38,960 Yeah. OK, three to four. 469 00:24:38,960 --> 00:24:40,360 All right, I got it. I nailed. 470 00:24:40,520 --> 00:24:44,040 It Apple's current. Just to reiterate there, Apple's 471 00:24:44,040 --> 00:24:49,600 current on device LLM uses 100% local LLM with three billion 472 00:24:49,600 --> 00:24:54,560 parameters and then can also connect to external servers for 473 00:24:54,560 --> 00:24:57,960 larger parameter models. And what's the price on this? 474 00:24:58,960 --> 00:25:05,160 Over 75% or probability 75%? OK, well I'm gonna grab the 475 00:25:05,160 --> 00:25:10,080 under on that cuz I think dedicated Co processor is really 476 00:25:10,080 --> 00:25:13,120 what's holding me back here cuz that requires like a separate 477 00:25:13,120 --> 00:25:15,960 chip essentially to go into the hardware. 478 00:25:16,000 --> 00:25:19,200 I'm a avid follower of Apple rumors and I have not gotten any 479 00:25:19,200 --> 00:25:22,480 sense that they're interested in adding chips like separate chips 480 00:25:22,480 --> 00:25:24,400 to their system to do this kind of thing. 481 00:25:24,400 --> 00:25:27,640 And so you'd basically have to get Xiaomi, Google and Samsung 482 00:25:27,640 --> 00:25:30,400 to all do this separately chip in the next year. 483 00:25:30,400 --> 00:25:34,560 And there's not really clear sentiment that that's necessary 484 00:25:34,560 --> 00:25:38,360 to run these models. So yeah, I'll take, I'll take 485 00:25:38,360 --> 00:25:41,520 the under at 75 for sure. You sold me under. 486 00:25:44,080 --> 00:25:47,440 Hardware Hardware timelines are really long, and it that that's 487 00:25:47,440 --> 00:25:49,560 only a slightly larger model than they can already run 488 00:25:49,560 --> 00:25:52,280 locally, so it's just not clear. To me, the parameters make 489 00:25:52,280 --> 00:25:58,560 sense, I And it's also not all these come on device for Google. 490 00:25:58,560 --> 00:26:00,160 Like, is that really in their mission? 491 00:26:00,160 --> 00:26:02,720 Like there's so much like, we'll send everything to the cloud. 492 00:26:02,720 --> 00:26:05,040 Like, are they really? Yeah, Apple's the one selling 493 00:26:05,040 --> 00:26:06,800 like everything will happen on device. 494 00:26:07,320 --> 00:26:11,200 I think I'd like grab the over at like 10% on this, but that's 495 00:26:11,200 --> 00:26:16,360 75 I think under big time. Yeah. 496 00:26:16,360 --> 00:26:20,480 I mean, I guess I'm hearing what you're saying on the dedicated 497 00:26:20,480 --> 00:26:24,320 aspect, but like I feel like you could quibble with that 498 00:26:24,320 --> 00:26:30,160 semantically of like as long as it's a GPU at all that is 499 00:26:30,160 --> 00:26:34,440 dedicated to running AI tasks. Like that's what the. 500 00:26:34,440 --> 00:26:36,160 GPU. They already have a GPU, right? 501 00:26:36,200 --> 00:26:39,320 I mean, yeah. OK, so they have essentially a 502 00:26:39,320 --> 00:26:44,360 GPU and then there's a set of cores, right? 503 00:26:48,000 --> 00:26:49,640 I guess it's. Just what is the neural engine I 504 00:26:49,640 --> 00:26:51,040 guess is what? I'm yeah, it's like are we 505 00:26:51,040 --> 00:26:54,360 counting the current neural engine on there for like 8 years 506 00:26:54,360 --> 00:26:56,680 already as as a yes to this because. 507 00:26:56,680 --> 00:26:58,120 Then it's like. Sure. 508 00:26:58,120 --> 00:27:01,880 Like I guess I mean like, but I don't think that's the spirit of 509 00:27:01,880 --> 00:27:05,000 the prediction, right, OK. Like the the thing they've had 510 00:27:05,000 --> 00:27:06,760 on the phones for the last eight years. 511 00:27:06,760 --> 00:27:07,840 Is I don't. Know. 512 00:27:08,080 --> 00:27:10,600 So far, Cha Chi BT is beating Anthropic because I feel like 513 00:27:10,600 --> 00:27:12,240 there's. A quality of prediction. 514 00:27:12,240 --> 00:27:13,200 I don't love this. This. 515 00:27:13,440 --> 00:27:16,800 Whatever, Yeah, Anyway. James, Just pick, pick and we'll 516 00:27:16,800 --> 00:27:20,400 litigate everything afterwards. Yeah, I'll take the over and 517 00:27:20,400 --> 00:27:22,480 I'll try to litigate my take that the neural ending. 518 00:27:23,240 --> 00:27:24,720 All right, Great, great. All right. 519 00:27:25,480 --> 00:27:30,160 Prediction #2 Anthropic will release a model that achieves a 520 00:27:30,160 --> 00:27:35,240 score above 90% on the Unified Benchmark for AI Reasoning, or 521 00:27:35,240 --> 00:27:38,840 Bar Evaluation Suite, which will become the first model to 522 00:27:38,880 --> 00:27:43,640 outperform the human baseline. Human expert baseline of 87%, 523 00:27:44,000 --> 00:27:47,520 probability 65%. Where are they at right now? 524 00:27:48,080 --> 00:27:50,280 Unified. What was the the metric? 525 00:27:50,280 --> 00:27:53,400 It was unified. Benchmark for unified, benchmark 526 00:27:53,400 --> 00:27:57,240 for AI reasoning. We're all going to frantically 527 00:27:57,240 --> 00:28:00,120 Google for that, yeah? It's funny that we have 528 00:28:00,120 --> 00:28:02,880 anthropic predicting its own intelligence. 529 00:28:03,000 --> 00:28:07,080 Capabilities I know. I mean, a key part of this is 530 00:28:07,080 --> 00:28:10,120 the claim that it's the first also, right? 531 00:28:12,040 --> 00:28:15,800 It says first, right? Yeah, it'll be the first model. 532 00:28:16,320 --> 00:28:19,800 For this, for the record, is this benchmark real? 533 00:28:19,800 --> 00:28:23,240 I'm I can't find this benchmark. Is this is this benchmark? 534 00:28:23,240 --> 00:28:25,560 Being my God, that would be hallucinated. 535 00:28:25,920 --> 00:28:28,320 I think it's my God. I can't find this. 536 00:28:28,680 --> 00:28:31,320 Unified benchmark for AI reasoning. 537 00:28:34,840 --> 00:28:37,760 It's like I was like God. I was like, I've never heard of 538 00:28:37,760 --> 00:28:40,600 this benchmark like. I love it. 539 00:28:40,600 --> 00:28:44,280 It's crazy. It's hallucinate a benchmark. 540 00:28:44,640 --> 00:28:50,000 I mean, can you find it? Like it says something about our 541 00:28:50,280 --> 00:28:54,960 deference that at first I'd. Wear like I did not defer. 542 00:28:54,960 --> 00:28:58,080 I didn't. No, no, no, I assume. 543 00:28:58,120 --> 00:29:00,640 That it's. Do do not loot me into this. 544 00:29:00,640 --> 00:29:03,400 Definitely. I feel like I I immediately 545 00:29:03,400 --> 00:29:05,320 googled it and I was like, what I've never heard of this. 546 00:29:05,680 --> 00:29:08,600 I've heard of like MMLU. I've heard of like, you know, a 547 00:29:08,600 --> 00:29:11,720 couple other evals, but I've never heard of the bar like or 548 00:29:11,720 --> 00:29:14,800 whatever. Maybe this is like internal 549 00:29:14,800 --> 00:29:18,480 philanthropic. Maybe this is like like, we'll 550 00:29:18,480 --> 00:29:22,000 all be proven idiots when this becomes a benchmark, right? 551 00:29:22,000 --> 00:29:24,560 And then? They leaked this. 552 00:29:24,880 --> 00:29:26,840 Yeah, OK. I guess I will. 553 00:29:27,280 --> 00:29:29,160 I'll take the under given I'm not sure this benchmark. 554 00:29:29,160 --> 00:29:33,040 Is real I guess if it doesn't exist we can create it and you 555 00:29:33,040 --> 00:29:35,680 know y'all rule it and you can make it true. 556 00:29:38,120 --> 00:29:39,400 We'll have to do some more Googling. 557 00:29:39,400 --> 00:29:43,000 Later OK, but I don't think this benchmark exists for my own 558 00:29:43,160 --> 00:29:47,160 recent. Just very quick for the listener 559 00:29:47,160 --> 00:29:49,520 to get to the substance. I know we're focused on 560 00:29:49,640 --> 00:29:53,280 technicalities. Do you think someone will 561 00:29:53,280 --> 00:29:58,240 surpass sort of human intelligence overall by the end 562 00:29:58,240 --> 00:30:00,640 of next year? And do you think it will be 563 00:30:00,640 --> 00:30:02,480 anthropic first? Just quick 10. 564 00:30:02,800 --> 00:30:05,520 In some kind of evaluation, you're saying basically like, 565 00:30:05,520 --> 00:30:09,160 yeah, yeah, for sure. I, I sort of think we're like 566 00:30:09,160 --> 00:30:11,280 already there. But I guess it depends what 567 00:30:11,480 --> 00:30:13,000 invented benchmark. Exactly. 568 00:30:13,000 --> 00:30:15,960 That's the take I was going to get, which is we have sort of 569 00:30:15,960 --> 00:30:19,800 quietly passed humans on a lot of things already. 570 00:30:19,800 --> 00:30:25,040 And we're just this, you know, since people, you know, smartest 571 00:30:25,040 --> 00:30:27,400 in every domain still remains smarter than it. 572 00:30:27,440 --> 00:30:31,000 Like we've sort of under undersold how exciting it is for 573 00:30:31,000 --> 00:30:34,520 it to be smarter than the average human on on a lot of 574 00:30:34,520 --> 00:30:35,840 stuff. Totally. 575 00:30:36,480 --> 00:30:40,720 Next prediction from Claude. At least 5 Fortune 500 companies 576 00:30:40,720 --> 00:30:43,920 will replace more than 25% of their middle management 577 00:30:43,920 --> 00:30:48,520 positions with AI systems for task delegation and performance 578 00:30:48,520 --> 00:30:51,840 monitoring, publicly acknowledging this transition in 579 00:30:51,840 --> 00:30:58,840 their annual reports. Probability 25%. 25% chance that 580 00:30:58,840 --> 00:31:01,880 25% of middle management positions get axed. 581 00:31:03,560 --> 00:31:06,600 What if they become ICS? If they become individual 582 00:31:06,600 --> 00:31:09,360 contributors where they don't get fired, is this met? 583 00:31:09,760 --> 00:31:12,240 I think that this would be met because I think that we're 584 00:31:12,240 --> 00:31:16,120 saying past delegation and performance monitoring, meaning 585 00:31:16,120 --> 00:31:18,720 it's most most of the management responsibilities. 586 00:31:19,240 --> 00:31:23,400 I'll, I'll, I'll take that over. And my reasoning is essentially 587 00:31:23,400 --> 00:31:26,600 just that there's a lot of companies, there's in fact 500 588 00:31:26,600 --> 00:31:29,440 companies in the Fortune 500 and. 589 00:31:29,680 --> 00:31:30,400 A lot. There's some. 590 00:31:30,640 --> 00:31:31,840 Number. Yeah, yeah. 591 00:31:31,840 --> 00:31:34,040 There's a pretty high number, somewhere around 500 of them. 592 00:31:34,960 --> 00:31:39,680 And and plenty of those companies are not growing or 593 00:31:39,680 --> 00:31:45,440 shrinking and need to cut talent in some way or another to manage 594 00:31:45,440 --> 00:31:48,840 the business. And this will be the greatest 595 00:31:48,960 --> 00:31:51,800 freaking like cover stories. Yeah, exactly. 596 00:31:51,800 --> 00:31:55,320 Smoke and mirrors. Like, you know, look over here, 597 00:31:55,320 --> 00:31:58,760 like the left hand's doing some AI doing layoffs because they're 598 00:31:58,800 --> 00:32:00,440 right. Yes, right, right. 599 00:32:00,440 --> 00:32:03,360 This is going to be an amazing way to drop in your earnings 600 00:32:03,360 --> 00:32:06,120 report that like we had to layoff a bunch of people, you 601 00:32:06,120 --> 00:32:08,760 know, but it's cool because it's, it's about the future dog. 602 00:32:08,760 --> 00:32:12,280 We're adopting AI. So I think that, yeah, I just 603 00:32:12,280 --> 00:32:17,920 totally buy that, you know, 5 out of 500, we'll need to do 604 00:32:17,920 --> 00:32:21,720 that in the next year. And it's just an amazing cover 605 00:32:21,720 --> 00:32:23,640 story for doing layoffs. What's the percentage? 606 00:32:23,640 --> 00:32:28,080 Eric James. They need to replace more than 607 00:32:28,080 --> 00:32:32,520 25% of their management positions and the probability is 608 00:32:32,520 --> 00:32:35,960 25% that this occurs. Wow, so it's a much better 609 00:32:35,960 --> 00:32:38,760 return for me to say. Yes, yeah, you get a, you get a 610 00:32:38,760 --> 00:32:41,840 4X return on the Yes. Yeah, over. 611 00:32:43,080 --> 00:32:46,880 I mean, yeah, I, I think the, the covers you guys, I'm, I'm 612 00:32:46,880 --> 00:32:50,360 not, I'm not as excited about AI in middle management 613 00:32:50,440 --> 00:32:54,120 necessarily, but I do think there are too many middle 614 00:32:54,120 --> 00:32:57,200 manager like we have this whole separate trend of like Facebook 615 00:32:57,200 --> 00:33:01,200 and companies saying let's just like be flatter organizations 616 00:33:01,200 --> 00:33:02,600 overall. And I feel like we're in that 617 00:33:02,600 --> 00:33:05,800 trend. And I do agree that the AI trend 618 00:33:05,800 --> 00:33:10,200 is sort of going to get mixed in with that, but I'm not convinced 619 00:33:10,680 --> 00:33:14,320 that AI is going to come for the middle managers first. 620 00:33:14,320 --> 00:33:17,360 I think it's going to supplement sort of the coders and 621 00:33:17,360 --> 00:33:19,240 frontline. You know, I to me, the 622 00:33:19,240 --> 00:33:22,040 deployment model is more like people are actually doing the 623 00:33:22,040 --> 00:33:26,120 work, figuring out how AI makes them better and sort of adding 624 00:33:26,120 --> 00:33:29,640 it to their just making themselves look better to their 625 00:33:29,640 --> 00:33:32,280 bosses. And I think that's how my model 626 00:33:32,280 --> 00:33:35,640 for how AI is getting deployed in workplaces, which is just 627 00:33:35,640 --> 00:33:38,960 sort of like people hacking shit together and delivering results. 628 00:33:40,600 --> 00:33:43,600 But I'm still taking the over. Yeah. 629 00:33:43,680 --> 00:33:45,960 I, I don't know, you kind of convinced me to take the under 630 00:33:45,960 --> 00:33:50,400 there. I think that I think that AI 631 00:33:50,400 --> 00:33:54,400 will have a much bigger impact on sort of IC roles. 632 00:33:54,400 --> 00:33:57,640 And you know, we've seen this already and some of the customer 633 00:33:57,640 --> 00:34:02,040 service oriented companies sort of announcing and bragging that 634 00:34:02,040 --> 00:34:05,000 they can hire less representatives. 635 00:34:05,760 --> 00:34:08,760 So I think we're more likely to see that first before companies 636 00:34:08,760 --> 00:34:14,760 start taking this AI approach on management, but could be wrong. 637 00:34:14,880 --> 00:34:19,320 25% is a big number at a Fortune 5. 638 00:34:19,800 --> 00:34:24,080 Exactly like are are they gonna? I I, I think I'm gonna I've my 639 00:34:24,080 --> 00:34:26,800 whole argument is the under. I feel like I should just like. 640 00:34:27,000 --> 00:34:28,159 No, you're not allowed to switch. 641 00:34:28,159 --> 00:34:31,560 You you really you? You made me not switch on like 642 00:34:31,560 --> 00:34:33,080 2. On the fence, I could see the 643 00:34:33,080 --> 00:34:34,600 advantage you. Know you're on the fence, you 644 00:34:34,719 --> 00:34:37,560 know, No, I'm committed or I'm over. 645 00:34:37,560 --> 00:34:38,120 Yeah. Committed. 646 00:34:38,159 --> 00:34:40,679 It's funny that I made the whole case for the I. 647 00:34:40,679 --> 00:34:42,320 Know, I know. Thank you. 648 00:34:42,320 --> 00:34:45,040 Thank you for. Your all right, all right. 649 00:34:45,679 --> 00:34:47,159 Once you say the words, you're. Locked. 650 00:34:47,159 --> 00:34:48,719 Yeah, once you say the words, you're locked, Bear. 651 00:34:49,199 --> 00:34:53,040 It locked. All right, let's go to the next 652 00:34:53,040 --> 00:34:59,040 prediction from Claude. This prediction is that the deep 653 00:34:59,040 --> 00:35:02,960 mind or let's say Google broadly, will demonstrate an AI 654 00:35:02,960 --> 00:35:06,680 system capable of discovering at least one novel pharmaceutical 655 00:35:06,680 --> 00:35:11,520 compound that also passes phase one clinical trials, reducing 656 00:35:11,880 --> 00:35:16,560 that typical discovery to trial timeline by around 50%. 657 00:35:16,560 --> 00:35:20,000 We will just specifically say that it has to have discovered a 658 00:35:20,000 --> 00:35:23,640 compound and passed phase one trials within the next year, 659 00:35:23,920 --> 00:35:27,640 probability 20%. I'm gonna take the Max world 660 00:35:27,640 --> 00:35:30,640 view, which is like a lot of these are predictions about 661 00:35:30,640 --> 00:35:33,360 society. And I definitely don't think 662 00:35:33,360 --> 00:35:37,440 drug approvals are going to sort of just get much faster because 663 00:35:37,560 --> 00:35:40,400 AI companies say they should. So I'm going to take the under. 664 00:35:42,680 --> 00:35:44,480 Who's Running the FDAI guess is another. 665 00:35:44,480 --> 00:35:45,760 Yeah, exactly. Oh yeah, that's. 666 00:35:45,760 --> 00:35:48,280 Trump I'm only conflicted because the Trump administration 667 00:35:48,280 --> 00:35:50,080 is about to take over. But they don't like Google, 668 00:35:50,080 --> 00:35:52,000 Google or they're not going to like Fast Track Google. 669 00:35:52,000 --> 00:35:53,880 That's the last company. That's the company they want to 670 00:35:53,880 --> 00:35:56,320 hit the most. Yeah, they don't like the FD. 671 00:35:56,520 --> 00:35:58,120 Or sorry, Yeah, they don't like Google. 672 00:35:59,360 --> 00:36:01,960 They also like pretend they didn't do Operation Warp Speed 673 00:36:01,960 --> 00:36:03,480 even though that was like the coolest thing ever. 674 00:36:04,440 --> 00:36:07,160 I don't really know how long, like Phase 1 specifically. 675 00:36:07,280 --> 00:36:08,440 I know I don't know a lot. About. 676 00:36:08,440 --> 00:36:11,360 I mean, I know the entire half. The time, right? 677 00:36:11,400 --> 00:36:12,680 Yeah, so. I know the entire. 678 00:36:12,680 --> 00:36:15,880 I don't know. I think that's just like kind of 679 00:36:15,880 --> 00:36:19,200 an additive less necessary component of the prediction 680 00:36:19,200 --> 00:36:21,160 reducing the typical discovery. OK. 681 00:36:21,160 --> 00:36:24,840 So well it just says bypassing phase one clinical trial within 682 00:36:24,840 --> 00:36:27,360 a year, it is reducing the. Typical would be very fast. 683 00:36:27,480 --> 00:36:30,520 Would be a 50% improvement in speed I guess. 684 00:36:30,640 --> 00:36:32,680 I know the entire time is like passage. 685 00:36:32,880 --> 00:36:36,080 It's like 10. 10 to 15 years. And so it's on the order of 686 00:36:36,080 --> 00:36:41,960 years for sure for phase one. Gosh, yeah, I think I just got 687 00:36:41,960 --> 00:36:44,240 to stick with my, I don't believe that the the 688 00:36:44,240 --> 00:36:47,880 institutions change as fast as the technology here and take the 689 00:36:47,880 --> 00:36:50,840 under even though 20% with the new administration is is 690 00:36:50,840 --> 00:36:54,600 appealing, that's just feels like a high bar to pass phase 691 00:36:54,600 --> 00:36:56,520 one trials. It does seem like phase one 692 00:36:56,520 --> 00:37:00,840 involves actual experiments on yeah, which AI cannot speed up. 693 00:37:01,080 --> 00:37:06,320 I just read Dario's essay and like one of the main points that 694 00:37:06,320 --> 00:37:10,880 he makes in his bowl case on AI is just a lot of the limits will 695 00:37:10,880 --> 00:37:13,920 be human systems and things where you have to run real world 696 00:37:13,920 --> 00:37:17,160 experiments and even if it has good ideas, the experiments 697 00:37:17,160 --> 00:37:21,120 could take a while anyway. All right, would you say? 698 00:37:21,640 --> 00:37:24,760 I'm taking the under too. I think that it just seems too 699 00:37:24,760 --> 00:37:28,400 fast to occur. Would you guys take the over on 700 00:37:28,400 --> 00:37:32,160 Google entering phase one trials or something with something? 701 00:37:33,400 --> 00:37:35,840 Yeah, I think there'll be a discovery or some. 702 00:37:36,400 --> 00:37:38,440 Yeah, yeah. Registering for them or 703 00:37:38,440 --> 00:37:39,960 something? Over 20%, I would take the 704 00:37:39,960 --> 00:37:43,680 around 20%. Obviously Demis at DeepMind is 705 00:37:43,680 --> 00:37:46,960 quite passionate about biology, is a major application for AI. 706 00:37:47,240 --> 00:37:49,720 So yeah, I mean, I think that would be a cool thing for them 707 00:37:49,720 --> 00:37:52,280 to demo, say, hey, look, we made a drug. 708 00:37:52,640 --> 00:37:54,240 Yeah. All right. 709 00:37:54,960 --> 00:37:59,120 Let's do one more from Claude. All right, prediction. 710 00:37:59,120 --> 00:38:02,560 The first international treaty specifically governing AI 711 00:38:02,560 --> 00:38:06,800 development and deployment would be ratified by 15 countries or 712 00:38:06,800 --> 00:38:12,360 more, including three of the following countries, US, China, 713 00:38:12,960 --> 00:38:16,480 EU, UK. What's the percentage on this? 714 00:38:16,560 --> 00:38:18,040 Is 5% like? Yeah. 715 00:38:18,040 --> 00:38:21,480 What is this like a 1%? They're legally binding 716 00:38:21,480 --> 00:38:26,240 commitments on things like model evaluation and a standard. 717 00:38:26,480 --> 00:38:27,920 What is the percentage they gave? 718 00:38:28,360 --> 00:38:31,400 Yeah, 50%. Oh my God, this is that's the 719 00:38:31,520 --> 00:38:33,400 Claude's. That's the I feel like that we 720 00:38:33,480 --> 00:38:35,800 needed much better. That's like that's the easiest. 721 00:38:35,800 --> 00:38:39,200 Under Under. Claude's like, under Claude's 722 00:38:39,200 --> 00:38:42,560 like dreaming of this fantasy safety world or it's something 723 00:38:42,560 --> 00:38:45,360 where where every country has come together. 724 00:38:45,400 --> 00:38:47,960 China, or like I didn't get the memo. 725 00:38:48,480 --> 00:38:51,920 You don't need the US or China because you only need three of 726 00:38:51,920 --> 00:38:56,520 these entities. So you could have EUUK and Japan 727 00:38:56,520 --> 00:38:58,040 all entering some sort of treaty. 728 00:38:58,560 --> 00:39:02,120 OK, interesting. EUI would believe I mean UK 729 00:39:02,120 --> 00:39:05,880 though is sort of I mean I definitely am picking the. 730 00:39:05,920 --> 00:39:09,680 Under I just feel like the yeah, the UK is a mess economically 731 00:39:09,680 --> 00:39:11,840 right now. Like our, you know, is the 732 00:39:11,840 --> 00:39:14,440 Labour Party really going to be like, yeah, guys, we're not 733 00:39:14,440 --> 00:39:17,240 fixing the economy, but we signed this dumb AI treaty that 734 00:39:17,240 --> 00:39:20,400 none of you care about. Like, I, I don't know, I, I 735 00:39:20,400 --> 00:39:22,200 think they got bigger fish to fry there. 736 00:39:23,680 --> 00:39:27,760 Yeah, under, under A50 for sure. Yeah, I agree with you guys 737 00:39:27,760 --> 00:39:32,760 taking the under on that. Yeah, I think this is Claude's 738 00:39:32,760 --> 00:39:37,080 own fever dream of hoping for. A better place, a better world, 739 00:39:37,080 --> 00:39:39,680 a better world. AI where Claude itself is highly 740 00:39:39,680 --> 00:39:45,320 regulated, Yes. Yeah, OK, great. 741 00:39:45,480 --> 00:39:48,240 Great work everyone. That was fun. 742 00:39:48,880 --> 00:39:52,280 How do you guys think ChatGPT did compared to Claude? 743 00:39:52,920 --> 00:39:54,880 What do you like about the different predictions? 744 00:39:56,720 --> 00:39:58,680 Well, it seemed like Claude made-up some stuff. 745 00:39:58,760 --> 00:40:03,160 I did think it had a little bit of a safety ISM trend line. 746 00:40:04,400 --> 00:40:08,800 So I don't know. They're all blur to me but it 747 00:40:08,800 --> 00:40:10,840 felt like the first half. Were. 748 00:40:11,160 --> 00:40:15,200 More interesting in the second. It seemed like GBT's predictions 749 00:40:15,200 --> 00:40:19,160 were more entertaining as well as more somewhat realistic. 750 00:40:19,160 --> 00:40:22,840 I guess. So I think I would give GBT the 751 00:40:22,840 --> 00:40:26,240 clear victory in this round of entertaining and reasonably 752 00:40:26,240 --> 00:40:28,400 accurate predictions. Maybe you prompted? 753 00:40:28,520 --> 00:40:32,200 For O1 for reasoning models. Yeah, you prompted exactly the 754 00:40:32,200 --> 00:40:33,000 same stuff. Or. 755 00:40:33,320 --> 00:40:35,040 Exact same prompt. Yeah, yeah. 756 00:40:35,920 --> 00:40:37,680 Right. Well, that's a that makes me 757 00:40:37,680 --> 00:40:40,280 more excited for the full version of O1 coming out. 758 00:40:40,320 --> 00:40:42,400 Yeah, that would that's probably one of our one of our 759 00:40:42,400 --> 00:40:43,840 predictions. I think that's coming out this 760 00:40:43,840 --> 00:40:47,880 year, right, is the rumor. I mean, the zoom out thing that 761 00:40:47,880 --> 00:40:53,560 I took from this is just how much human systems interplay 762 00:40:53,560 --> 00:40:56,680 with the things we want AI to deliver. 763 00:40:56,680 --> 00:41:02,600 And even if we are bullish on rapid development of AI, there 764 00:41:02,600 --> 00:41:05,040 are a lot of things where we're like, well, the human systems 765 00:41:05,040 --> 00:41:07,840 will still effectively limit what it can do. 766 00:41:08,160 --> 00:41:11,960 That's self driving drug development, sort of everything 767 00:41:11,960 --> 00:41:14,320 under the sun. And so I think that's 768 00:41:14,320 --> 00:41:17,080 interesting to come out of this. How much? 769 00:41:18,080 --> 00:41:22,600 Yeah, this this sort of societal piece is still sort of a strong 770 00:41:23,080 --> 00:41:25,200 clever on what AI can accomplish. 771 00:41:25,560 --> 00:41:29,360 I feel like the way we get to that treaty is that something 772 00:41:29,360 --> 00:41:32,400 horrible goes on, right? Yeah, exactly right. 773 00:41:32,400 --> 00:41:34,240 Like, yeah. Something cool. 774 00:41:34,240 --> 00:41:36,720 Yeah. We underestimated Black Swan. 775 00:41:36,960 --> 00:41:40,440 We have to narrowly escape from some horrible thing to get a 776 00:41:40,640 --> 00:41:42,880 binding treaty between a bunch of countries, I think. 777 00:41:43,240 --> 00:41:44,280 Oh. Yeah, then we'll look really 778 00:41:44,280 --> 00:41:44,760 dumb. Yeah. 779 00:41:44,760 --> 00:41:47,720 It's like they, you know, AI does figure out how to like. 780 00:41:48,560 --> 00:41:50,600 Although we just had a paper. Clips or whatever to send a 781 00:41:50,600 --> 00:41:53,000 nuclear. Bomb we just had a pandemic and 782 00:41:53,640 --> 00:41:55,280 nothing came out of that like hey. 783 00:41:55,920 --> 00:41:58,040 Move on. It's not like. 784 00:41:58,440 --> 00:42:00,560 Some tragedies that bother us more than others. 785 00:42:00,680 --> 00:42:03,120 Yeah, but the US, China and the EU didn't get together and be 786 00:42:03,120 --> 00:42:05,280 like, hey, maybe we should like control pandemics a little more 787 00:42:05,280 --> 00:42:06,840 carefully. Like, you know, let's let's 788 00:42:06,880 --> 00:42:09,280 regulate that a little bit. It's like, Nah, we we got 789 00:42:09,280 --> 00:42:10,440 through it. Don't worry about it. 790 00:42:11,160 --> 00:42:14,520 Let's just get on the record on yeah, I mean the core just 791 00:42:14,520 --> 00:42:17,120 because it is a prediction episode and like we, we got into 792 00:42:17,120 --> 00:42:21,680 sort of the technicals like do you think the models will be 793 00:42:22,760 --> 00:42:26,560 significantly smarter next year, like by the end of next year in 794 00:42:26,560 --> 00:42:30,320 that they will we'll see the same basic rate of progress, 795 00:42:30,320 --> 00:42:33,680 meaning these are like PhD level. 796 00:42:34,000 --> 00:42:39,280 Thinkers I think by the next year, I think next year the 797 00:42:39,280 --> 00:42:41,560 overall take from this discussion like you said is the 798 00:42:41,560 --> 00:42:44,520 difference between things that can be facilitated by software, 799 00:42:44,920 --> 00:42:47,800 things that can be require hardware and then things that 800 00:42:47,800 --> 00:42:51,480 require institutions and government and so on right and I 801 00:42:51,480 --> 00:42:54,480 think next year I think it's going to be a crazy year for 802 00:42:54,480 --> 00:42:58,120 leaps forward and software. I think that GBT 5 is going to 803 00:42:58,120 --> 00:43:01,680 come I think 01 or O2 or whatever the GBT 5 level version 804 00:43:01,680 --> 00:43:02,960 of the thinking model is going to come. 805 00:43:03,360 --> 00:43:06,000 I think these computer use API's, as I said in a previous 806 00:43:06,000 --> 00:43:08,960 episode, or just an incredibly powerful interface to the 807 00:43:08,960 --> 00:43:11,480 Internet and the software world for these models. 808 00:43:11,840 --> 00:43:17,520 And so I think it's going to be like a crazy, crazy software a 809 00:43:17,520 --> 00:43:18,920 year. I'm I think it'll be the 810 00:43:18,920 --> 00:43:22,640 beginning of the impact on hardware and institutions. 811 00:43:23,360 --> 00:43:26,960 But it feels like the old Intel like tick tock strategy where 812 00:43:26,960 --> 00:43:29,600 you have one year where you have huge progress and then one year 813 00:43:29,600 --> 00:43:31,400 of refinement. And that's like the tick and the 814 00:43:31,400 --> 00:43:34,280 talk or whatever. I think 24 was kind of a talk 815 00:43:34,280 --> 00:43:37,240 year where we didn't have really dramatic progress in the models, 816 00:43:37,240 --> 00:43:40,800 but we started to see the existing models eat away at all 817 00:43:40,800 --> 00:43:44,240 these, you know, tools and software that we have. 818 00:43:44,240 --> 00:43:46,720 And now next year, I think we have a big, big tick here. 819 00:43:46,960 --> 00:43:50,040 And that'll be the foundation for like, you know, societal 820 00:43:50,040 --> 00:43:52,640 transformation for years and decades to come. 821 00:43:52,640 --> 00:43:55,640 So I'm I'm very long model progress in 25. 822 00:43:56,200 --> 00:44:01,920 I think that next year we'll we'll be talking about this 823 00:44:01,920 --> 00:44:04,440 trend of AI getting slower and slower. 824 00:44:04,440 --> 00:44:09,000 So we started out. Just going slower and adventure 825 00:44:09,000 --> 00:44:10,120 define. Slower. 826 00:44:10,120 --> 00:44:12,000 All right. No, no, no, I don't know. 827 00:44:12,200 --> 00:44:14,880 I'm not talking about reasoning capability. 828 00:44:14,880 --> 00:44:16,880 I'm talking about inference time essentially. 829 00:44:16,880 --> 00:44:21,680 And basically what you, what we've seen already is like with 830 00:44:21,680 --> 00:44:24,440 O1, right? And now it takes, you know, 5 to 831 00:44:24,440 --> 00:44:27,000 30 seconds to return an answer, but it can be better. 832 00:44:27,000 --> 00:44:30,800 What we're hearing is that, you know, internally, at least in 833 00:44:30,800 --> 00:44:33,960 open AI, they feel like that's a whole new scaling paradigm to 834 00:44:33,960 --> 00:44:36,880 get better and better reasoning is, is inference time compute. 835 00:44:37,360 --> 00:44:41,480 And what I think, you know, we'll see is like, like Max 836 00:44:41,480 --> 00:44:45,200 said, a lot of these computer use models are opening up 837 00:44:45,240 --> 00:44:48,360 opportunities for kind of agentic systems. 838 00:44:48,360 --> 00:44:51,440 And we'll start to see, you know, people like ourselves 839 00:44:51,440 --> 00:44:55,880 using these models that take hours to return US information, 840 00:44:55,880 --> 00:44:57,760 right? So Eric could go talk to an 841 00:44:57,760 --> 00:45:00,120 agent and just be like, hey, could you go research this thing 842 00:45:00,120 --> 00:45:02,520 for an upcoming article I'm writing? 843 00:45:02,520 --> 00:45:07,480 Or, you know, Max and I could have one of these agents go off 844 00:45:07,480 --> 00:45:10,760 and kind of research some game genre that we're looking to 845 00:45:10,760 --> 00:45:13,440 build, right? But it will just take longer and 846 00:45:13,440 --> 00:45:15,760 longer to return. And, and, you know, I think 847 00:45:15,760 --> 00:45:17,120 that'll be a really interesting trend. 848 00:45:17,480 --> 00:45:21,600 And I think that, you know, maybe that's the prediction for 849 00:45:21,600 --> 00:45:24,480 our, our own internal usage of how things are, how we're going 850 00:45:24,480 --> 00:45:27,400 to be changing how we work with these age, these AI systems. 851 00:45:27,720 --> 00:45:32,000 But I also think more specific prediction for like industry or 852 00:45:32,640 --> 00:45:36,120 or enterprise would be that we're going to see a lot of use 853 00:45:36,120 --> 00:45:39,040 cases in, in sort of the finance space for these agents, right? 854 00:45:39,040 --> 00:45:44,120 Like there will be new forms of hedge funds or investment groups 855 00:45:44,120 --> 00:45:47,280 that are like using these agents to actually trade. 856 00:45:47,280 --> 00:45:49,800 And, you know, maybe we'll see that primarily in crypto, but I 857 00:45:49,800 --> 00:45:52,800 could also see that occurring in sort of traditional assets. 858 00:45:52,800 --> 00:45:55,520 But I do think we're going to see a lot of like adoption of 859 00:45:55,520 --> 00:45:59,160 these agentic models to like trade their own portfolios on 860 00:45:59,160 --> 00:46:04,440 behalf of their, you know, human kind of owners or overseers, I 861 00:46:04,440 --> 00:46:06,280 guess. I mean, this isn't a prediction. 862 00:46:06,320 --> 00:46:10,960 And I think it sort of I mean, I think one thing we are seeing is 863 00:46:10,960 --> 00:46:15,000 that intelligence can be out there and most a lot of people 864 00:46:15,000 --> 00:46:17,800 don't know how to use it, right. So I think we're going to see 865 00:46:18,160 --> 00:46:22,240 obviously this uptick in applications taking existing, 866 00:46:22,280 --> 00:46:25,120 you know, the intelligence that has been proven with open AI, 867 00:46:25,120 --> 00:46:28,800 Anthropic and others and making it much more straightforward for 868 00:46:28,800 --> 00:46:30,800 people to use it in particular use cases. 869 00:46:30,800 --> 00:46:33,640 And I think there's going to be a ton of money there. 870 00:46:33,800 --> 00:46:36,000 So that's applications, which we've talked a lot about. 871 00:46:36,280 --> 00:46:38,680 And then I think the other piece that you're sort of touching on 872 00:46:38,680 --> 00:46:41,600 is that they're going to be these sort of specialists that 873 00:46:41,600 --> 00:46:44,520 appreciate how intelligent these models are. 874 00:46:44,760 --> 00:46:46,880 And they're going to figure out either how to use it and make 875 00:46:46,880 --> 00:46:51,360 money on it directly or become service providers for people who 876 00:46:51,360 --> 00:46:54,400 haven't really figured it out and do things better and say, 877 00:46:54,400 --> 00:46:56,400 all right, just pay me. I'll solve this problem. 878 00:46:56,760 --> 00:46:59,600 Secretly I'm using AI or not so secretly I'm using AI so I can 879 00:46:59,600 --> 00:47:01,800 do it cheaper than you. And I think that's how we're 880 00:47:01,800 --> 00:47:05,320 going to see these two models. There's sort of two ways people 881 00:47:05,320 --> 00:47:08,760 are going to implement AI. And I think so we'll see more AI 882 00:47:08,760 --> 00:47:10,840 applications and AI service companies. 883 00:47:11,080 --> 00:47:13,560 And, you know, the AI applications will almost 884 00:47:13,560 --> 00:47:17,960 certainly lag the actual power, the thinking power of AI. 885 00:47:18,600 --> 00:47:22,240 So it'll be an interesting sort of push and pull there. 886 00:47:22,880 --> 00:47:24,880 For sure. Cool. 887 00:47:25,760 --> 00:47:27,800 Yeah, Super fun. Yeah. 888 00:47:28,880 --> 00:47:32,400 Well, I feel, I don't know, James, you'll try and come up 889 00:47:32,400 --> 00:47:34,960 with the score of those predictions. 890 00:47:34,960 --> 00:47:38,040 I don't know. I think, I think to me they were 891 00:47:38,040 --> 00:47:42,440 more like what the draft I am invested, ego invested in the 892 00:47:42,440 --> 00:47:44,240 outcome. I feel like these predictions 893 00:47:44,240 --> 00:47:48,960 were more thought experiment than necessarily to me, the sort 894 00:47:48,960 --> 00:47:53,040 of perfect proxy for what we think, how accurately we see 895 00:47:53,040 --> 00:47:55,880 what's coming in AI, but certainly enjoyed it. 896 00:47:56,520 --> 00:47:58,880 That was a fun, fun concept. Fun episode. 897 00:47:58,880 --> 00:48:00,920 Yeah. Yeah, Greg. 898 00:48:01,080 --> 00:48:03,880 Well, like really really demonstrated that AI was better 899 00:48:03,880 --> 00:48:06,440 at creating the grist for the content. 900 00:48:06,520 --> 00:48:07,880 Than we are. Yeah, exactly. 901 00:48:07,880 --> 00:48:10,040 I mean, we were supposed to come up with predictions as like, 902 00:48:10,840 --> 00:48:12,160 yeah. I got a couple. 903 00:48:12,280 --> 00:48:14,080 Ideas. I mean it shows that sometimes 904 00:48:14,080 --> 00:48:16,600 like just having throwing shit at the wall to react to you is 905 00:48:16,600 --> 00:48:19,440 better than like overthinking my perfect prediction. 906 00:48:20,920 --> 00:48:23,000 And getting and getting your hot picks live, right? 907 00:48:23,000 --> 00:48:24,720 Getting. Yeah, exactly. 908 00:48:24,840 --> 00:48:27,560 Having to think through it. Seeing the thinking, that's 909 00:48:27,560 --> 00:48:30,560 something that, you know, AI is just starting to learn how to 910 00:48:30,560 --> 00:48:33,320 do. You know, humans still went on 911 00:48:33,320 --> 00:48:38,160 the showing our thought process. Anyway, we will be back after 912 00:48:38,160 --> 00:48:43,160 Cerebral Valley and we will be publishing all the talks. 913 00:48:43,200 --> 00:48:46,480 We're going to be throwing you our favorite clips and then we 914 00:48:46,480 --> 00:48:50,920 will be back with Max and James after the conference, reacting 915 00:48:51,240 --> 00:48:52,880 to everything that happened on stage. 916 00:48:52,880 --> 00:48:56,520 So come back and see us, newcomer. 917 00:48:56,720 --> 00:48:57,600 All right. Thanks so much. 918 00:48:58,320 --> 00:48:58,640 Thank you.