1 00:00:00,000 --> 00:00:02,160 Welcome back everyone for another deep dive. 2 00:00:02,160 --> 00:00:03,840 Always fun to dive back in. 3 00:00:03,840 --> 00:00:06,160 You guys have been digging into AI. 4 00:00:06,160 --> 00:00:07,000 Yeah. 5 00:00:07,000 --> 00:00:11,320 Especially from some excerpts of an interview 6 00:00:11,320 --> 00:00:12,560 with Ilya Sutskiver. 7 00:00:12,560 --> 00:00:13,520 Right. 8 00:00:13,520 --> 00:00:15,560 The co-founder of OpenAI. 9 00:00:15,560 --> 00:00:17,360 And he's making some pretty bold predictions. 10 00:00:17,360 --> 00:00:18,200 Yeah, he is. 11 00:00:18,200 --> 00:00:19,840 About where AI is headed. 12 00:00:19,840 --> 00:00:21,920 And we're here to kind of help make sense of all that. 13 00:00:21,920 --> 00:00:22,880 Yeah, unpack it all. 14 00:00:22,880 --> 00:00:24,680 Extract some of those interesting nuggets. 15 00:00:24,680 --> 00:00:27,760 What I think is so interesting about what Sutskiver's saying 16 00:00:27,760 --> 00:00:30,640 is that he's not just some tech pundit kind of throwing out 17 00:00:30,640 --> 00:00:31,680 wild theories. 18 00:00:31,680 --> 00:00:32,680 Right. 19 00:00:32,680 --> 00:00:33,800 You know, he's at OpenAI. 20 00:00:33,800 --> 00:00:34,960 He's building this future. 21 00:00:34,960 --> 00:00:35,840 Exactly, yeah. 22 00:00:35,840 --> 00:00:39,280 So I think his predictions are particularly thought provoking. 23 00:00:39,280 --> 00:00:40,400 He gets right to the point. 24 00:00:40,400 --> 00:00:40,960 He does. 25 00:00:40,960 --> 00:00:42,440 He doesn't waste any time. 26 00:00:42,440 --> 00:00:43,000 Nope. 27 00:00:43,000 --> 00:00:46,480 Right out of the gate, he says, pre-training as we know it 28 00:00:46,480 --> 00:00:48,280 will unquestionably end. 29 00:00:48,280 --> 00:00:49,400 Wow. 30 00:00:49,400 --> 00:00:51,120 And that's a pretty big statement. 31 00:00:51,120 --> 00:00:51,680 Huge. 32 00:00:51,680 --> 00:00:53,800 For anyone who's been following AI at all. 33 00:00:53,800 --> 00:00:54,160 Yeah. 34 00:00:54,160 --> 00:00:56,480 Because pre-training is really the bedrock 35 00:00:56,480 --> 00:00:57,720 of so much of what we're seeing. 36 00:00:57,720 --> 00:00:58,320 For sure. 37 00:00:58,320 --> 00:01:01,040 Explain pre-training, though, for someone who maybe 38 00:01:01,040 --> 00:01:03,120 is just kind of getting into this. 39 00:01:03,120 --> 00:01:06,040 So think of pre-training as building a foundation. 40 00:01:06,040 --> 00:01:06,560 OK. 41 00:01:06,560 --> 00:01:10,520 You're feeding the AI massive amounts of data 42 00:01:10,520 --> 00:01:13,320 so it can learn the basics of language and code almost 43 00:01:13,320 --> 00:01:15,840 like a crash course in worldly knowledge. 44 00:01:15,840 --> 00:01:16,320 OK. 45 00:01:16,320 --> 00:01:18,320 But Sutskiver's argument is that we're 46 00:01:18,320 --> 00:01:20,200 hitting a wall with this approach. 47 00:01:20,200 --> 00:01:20,920 Makes sense. 48 00:01:20,920 --> 00:01:22,680 There's only so much data out there. 49 00:01:22,680 --> 00:01:23,000 Right. 50 00:01:23,000 --> 00:01:25,360 So this method has an inherent limit. 51 00:01:25,360 --> 00:01:25,760 Yeah. 52 00:01:25,760 --> 00:01:28,200 If pre-training is on its way out, then what comes next? 53 00:01:28,200 --> 00:01:28,760 Right. 54 00:01:28,760 --> 00:01:29,640 What replaces it? 55 00:01:29,640 --> 00:01:30,800 What replaces it, yeah. 56 00:01:30,800 --> 00:01:32,440 Well, Sutskiver believes one answer 57 00:01:32,440 --> 00:01:34,960 lies in the development of agentic AI. 58 00:01:34,960 --> 00:01:36,520 Oh, that's fascinating to me. 59 00:01:36,520 --> 00:01:37,360 It is, isn't it? 60 00:01:37,360 --> 00:01:38,400 Agentic AI, yeah. 61 00:01:38,400 --> 00:01:42,120 So right now, something like ChatGPT is impressive, 62 00:01:42,120 --> 00:01:43,640 but it's still a tool. 63 00:01:43,640 --> 00:01:45,440 You give it input, it gives you output. 64 00:01:45,440 --> 00:01:47,360 Agentic AI is completely different. 65 00:01:47,360 --> 00:01:47,920 OK. 66 00:01:47,920 --> 00:01:50,880 Imagine an AI that sets its own goals. 67 00:01:50,880 --> 00:01:51,560 Oh, wow. 68 00:01:51,560 --> 00:01:54,440 Can reason about its environment and takes action 69 00:01:54,440 --> 00:01:57,560 without needing constant human guidance. 70 00:01:57,560 --> 00:02:00,800 So instead of me saying, OK, write me a poem, 71 00:02:00,800 --> 00:02:02,720 I guess I manage this whole project. 72 00:02:02,720 --> 00:02:03,320 Exactly. 73 00:02:03,320 --> 00:02:05,600 Do the research, figure out the execution. 74 00:02:05,600 --> 00:02:07,800 It's not just about asking and answering. 75 00:02:07,800 --> 00:02:10,120 It's about collaboration and partnership. 76 00:02:10,120 --> 00:02:11,000 Wow. 77 00:02:11,000 --> 00:02:14,360 Almost like having a truly intelligent colleague. 78 00:02:14,360 --> 00:02:15,920 Yeah, that's a totally different relationship. 79 00:02:15,920 --> 00:02:16,320 For sure. 80 00:02:16,320 --> 00:02:16,960 With AI. 81 00:02:16,960 --> 00:02:17,640 Absolutely. 82 00:02:17,640 --> 00:02:19,800 You know, earlier you were talking about data limitations. 83 00:02:19,800 --> 00:02:20,160 Yeah. 84 00:02:20,160 --> 00:02:22,760 Sutskiver also talks about synthetic data 85 00:02:22,760 --> 00:02:24,240 as a potential solution. 86 00:02:24,240 --> 00:02:25,120 He does. 87 00:02:25,120 --> 00:02:27,680 Which has its own pros and cons, right? 88 00:02:27,680 --> 00:02:28,320 Definitely. 89 00:02:28,320 --> 00:02:33,000 Yeah, he sees it as a way to break free from relying solely 90 00:02:33,000 --> 00:02:34,120 on existing data. 91 00:02:34,120 --> 00:02:34,600 OK. 92 00:02:34,600 --> 00:02:37,400 With synthetic data, we could create entirely new data 93 00:02:37,400 --> 00:02:40,480 sets that mimic real world information. 94 00:02:40,480 --> 00:02:41,080 Oh, interesting. 95 00:02:41,080 --> 00:02:45,000 So allowing us to train AI on scenarios that are rare or even 96 00:02:45,000 --> 00:02:46,760 impossible to capture in real life. 97 00:02:46,760 --> 00:02:47,960 I see the benefit of that. 98 00:02:47,960 --> 00:02:48,280 Yeah. 99 00:02:48,280 --> 00:02:51,160 But could that create AI that's out of touch with reality? 100 00:02:51,160 --> 00:02:52,280 That's the challenge, right? 101 00:02:52,280 --> 00:02:53,040 Yeah. 102 00:02:53,040 --> 00:02:55,840 Creating synthetic data that accurately reflects 103 00:02:55,840 --> 00:03:00,040 the nuances of the real world is a huge undertaking. 104 00:03:00,040 --> 00:03:02,520 If we get it wrong, we could end up 105 00:03:02,520 --> 00:03:06,960 with AI that's brilliant in theory, but fails in practice. 106 00:03:06,960 --> 00:03:10,240 You want AI that is creative and adaptable, 107 00:03:10,240 --> 00:03:12,880 but also grounded in the real world. 108 00:03:12,880 --> 00:03:15,280 He uses a really interesting analogy 109 00:03:15,280 --> 00:03:16,560 to explain another challenge. 110 00:03:16,560 --> 00:03:20,240 And that is the brain body size ratio in human evolution. 111 00:03:20,240 --> 00:03:22,400 Yeah, the brain body size ratio. 112 00:03:22,400 --> 00:03:24,040 It's a brilliant comparison. 113 00:03:24,040 --> 00:03:24,640 It is. 114 00:03:24,640 --> 00:03:28,920 In humans, our brain size increased at a rate far beyond. 115 00:03:28,920 --> 00:03:29,640 Much faster. 116 00:03:29,640 --> 00:03:31,880 What would be expected based on our body size. 117 00:03:31,880 --> 00:03:32,400 Right. 118 00:03:32,400 --> 00:03:36,480 He's suggesting we might need to find a similar leap for AI. 119 00:03:36,480 --> 00:03:37,160 I see. 120 00:03:37,160 --> 00:03:39,160 We can't just keep throwing more data and bigger 121 00:03:39,160 --> 00:03:40,120 networks at the problem. 122 00:03:40,120 --> 00:03:40,400 Right. 123 00:03:40,400 --> 00:03:41,360 Just getting bigger. 124 00:03:41,360 --> 00:03:44,120 We need to find new ways to scale intelligence itself. 125 00:03:44,120 --> 00:03:46,040 To make it smarter, not just bigger. 126 00:03:46,040 --> 00:03:46,960 Exactly. 127 00:03:46,960 --> 00:03:49,320 And that leads us to one of the most mind blowing 128 00:03:49,320 --> 00:03:50,960 aspects of his predictions. 129 00:03:50,960 --> 00:03:51,480 OK. 130 00:03:51,480 --> 00:03:53,160 This future of AI reasoning. 131 00:03:53,160 --> 00:03:53,560 OK. 132 00:03:53,560 --> 00:03:54,600 And the implications of that. 133 00:03:54,600 --> 00:03:57,800 Yeah, what separates future AI from what we have now 134 00:03:57,800 --> 00:04:00,520 is this ability to go beyond pattern matching. 135 00:04:00,520 --> 00:04:00,920 OK. 136 00:04:00,920 --> 00:04:02,360 To genuine reasoning. 137 00:04:02,360 --> 00:04:03,400 Think about it. 138 00:04:03,400 --> 00:04:06,480 If AI starts reasoning like we do, 139 00:04:06,480 --> 00:04:09,360 it becomes inherently less predictable. 140 00:04:09,360 --> 00:04:09,960 Oh, OK. 141 00:04:09,960 --> 00:04:12,160 Because it's no longer just following a set of rules. 142 00:04:12,160 --> 00:04:12,520 Right. 143 00:04:12,520 --> 00:04:13,040 Right. 144 00:04:13,040 --> 00:04:13,600 OK. 145 00:04:13,600 --> 00:04:15,720 So we have AI that can think for itself. 146 00:04:15,720 --> 00:04:16,240 Yeah. 147 00:04:16,240 --> 00:04:18,080 Set its own goals, act independently. 148 00:04:18,080 --> 00:04:18,560 Uh-huh. 149 00:04:18,560 --> 00:04:21,160 This is where things start to feel a little sci-fi. 150 00:04:21,160 --> 00:04:21,660 Right. 151 00:04:21,660 --> 00:04:22,240 It does get a little. 152 00:04:22,240 --> 00:04:26,920 What happens when AI starts to solve problems in ways 153 00:04:26,920 --> 00:04:28,120 we didn't anticipate? 154 00:04:28,120 --> 00:04:28,360 Right. 155 00:04:28,360 --> 00:04:30,120 Or even understand? 156 00:04:30,120 --> 00:04:31,000 It's a great question. 157 00:04:31,000 --> 00:04:31,400 Yeah. 158 00:04:31,400 --> 00:04:35,120 And one of the most fascinating and potentially challenging 159 00:04:35,120 --> 00:04:37,840 aspects of what Sitzgever is describing. 160 00:04:37,840 --> 00:04:38,480 Right. 161 00:04:38,480 --> 00:04:42,400 He believes future AI will be able to reason through problems, 162 00:04:42,400 --> 00:04:45,320 consider different options, and make decisions 163 00:04:45,320 --> 00:04:48,640 based on an actual understanding of the situation. 164 00:04:48,640 --> 00:04:51,240 So it's not just about giving it more information. 165 00:04:51,240 --> 00:04:51,520 Right. 166 00:04:51,520 --> 00:04:53,760 It's about giving it the ability to think critically 167 00:04:53,760 --> 00:04:56,160 and creatively about that information. 168 00:04:56,160 --> 00:04:56,840 Exactly. 169 00:04:56,840 --> 00:04:59,320 How do we ensure that that reasoning 170 00:04:59,320 --> 00:05:02,200 aligns with our own values and goals? 171 00:05:02,200 --> 00:05:03,040 It's a key question. 172 00:05:03,040 --> 00:05:04,440 Especially as it becomes more autonomous. 173 00:05:04,440 --> 00:05:04,880 A big. 174 00:05:04,880 --> 00:05:05,400 Yeah. 175 00:05:05,400 --> 00:05:06,560 It's a balancing act, isn't it? 176 00:05:06,560 --> 00:05:07,560 Huge. 177 00:05:07,560 --> 00:05:09,640 We want AI that can think for itself 178 00:05:09,640 --> 00:05:11,560 and help us solve complex problems. 179 00:05:11,560 --> 00:05:12,200 Yeah. 180 00:05:12,200 --> 00:05:15,240 But we also need to ensure that its actions 181 00:05:15,240 --> 00:05:17,040 align with human values and goals. 182 00:05:17,040 --> 00:05:17,280 Yeah. 183 00:05:17,280 --> 00:05:20,120 We've got to be careful about what we ask it to do. 184 00:05:20,120 --> 00:05:21,800 And this is just the tip of the iceberg 185 00:05:21,800 --> 00:05:25,680 when it comes to the future of AI as Sitzgever sees it. 186 00:05:25,680 --> 00:05:27,440 Oh, there's so much more to unpack. 187 00:05:27,440 --> 00:05:28,280 So much more. 188 00:05:28,280 --> 00:05:29,280 We've got to keep going. 189 00:05:29,280 --> 00:05:30,200 Let's do it. 190 00:05:30,200 --> 00:05:31,360 So much more. 191 00:05:31,360 --> 00:05:31,720 Yeah. 192 00:05:31,720 --> 00:05:34,720 And we haven't even touched on the hallucination problem, 193 00:05:34,720 --> 00:05:39,240 which has been a major hurdle in AI development. 194 00:05:39,240 --> 00:05:41,320 As impressive as current AI is. 195 00:05:41,320 --> 00:05:41,840 Yeah. 196 00:05:41,840 --> 00:05:44,120 It still has that tendency to just make stuff up. 197 00:05:44,120 --> 00:05:44,720 Totally. 198 00:05:44,720 --> 00:05:50,080 You know, generate incorrect or even nonsensical information. 199 00:05:50,080 --> 00:05:50,480 Yeah. 200 00:05:50,480 --> 00:05:52,640 And that's a big problem if we want to rely on AI 201 00:05:52,640 --> 00:05:54,720 for things like scientific research. 202 00:05:54,720 --> 00:05:55,760 Or medical diagnoses. 203 00:05:55,760 --> 00:05:56,240 Exactly. 204 00:05:56,240 --> 00:05:57,440 High stakes situations. 205 00:05:57,440 --> 00:05:58,120 Exactly. 206 00:05:58,120 --> 00:06:01,520 Sitzgever believes that as AI becomes more sophisticated 207 00:06:01,520 --> 00:06:04,200 in its reasoning abilities, it will actually 208 00:06:04,200 --> 00:06:07,640 be able to identify and correct these solutions on its own. 209 00:06:07,640 --> 00:06:10,360 So almost like a built-in fact checker. 210 00:06:10,360 --> 00:06:10,680 Yeah. 211 00:06:10,680 --> 00:06:12,200 Like a self-correcting mechanism. 212 00:06:12,200 --> 00:06:15,600 So AI that can not only generate information, 213 00:06:15,600 --> 00:06:19,920 but also critically evaluate its own output 214 00:06:19,920 --> 00:06:20,960 and make adjustments. 215 00:06:20,960 --> 00:06:21,760 Yeah, make adjustments. 216 00:06:21,760 --> 00:06:22,040 Yeah. 217 00:06:22,040 --> 00:06:23,840 Based on reasoning and logic. 218 00:06:23,840 --> 00:06:26,480 It ties back to his point about generalization. 219 00:06:26,480 --> 00:06:27,000 OK. 220 00:06:27,000 --> 00:06:29,280 He believes that AI will eventually 221 00:06:29,280 --> 00:06:32,080 be able to solve problems that hasn't been specifically 222 00:06:32,080 --> 00:06:32,840 trained on. 223 00:06:32,840 --> 00:06:33,360 OK. 224 00:06:33,360 --> 00:06:35,360 Much like humans can apply knowledge 225 00:06:35,360 --> 00:06:36,800 from one domain to another. 226 00:06:36,800 --> 00:06:40,440 I'm thinking like when we learn to drive, 227 00:06:40,440 --> 00:06:42,520 we might practice in a limited area. 228 00:06:42,520 --> 00:06:43,040 Right. 229 00:06:43,040 --> 00:06:45,160 But then we can navigate new environments 230 00:06:45,160 --> 00:06:46,680 in unexpected situations. 231 00:06:46,680 --> 00:06:47,280 Exactly. 232 00:06:47,280 --> 00:06:49,080 So he seems to think AI will achieve 233 00:06:49,080 --> 00:06:50,760 a similar level of adaptability. 234 00:06:50,760 --> 00:06:51,320 He does. 235 00:06:51,320 --> 00:06:53,080 Although he acknowledges that it might not 236 00:06:53,080 --> 00:06:56,040 be quite as seamless as human generalization. 237 00:06:56,040 --> 00:06:56,400 Right. 238 00:06:56,400 --> 00:06:58,880 But even approaching that level of adaptability 239 00:06:58,880 --> 00:07:01,440 would have massive implications for AI's ability 240 00:07:01,440 --> 00:07:02,680 to function in the real world. 241 00:07:02,680 --> 00:07:03,000 Yeah. 242 00:07:03,000 --> 00:07:05,480 It feels like we're talking about AI that's not just a tool, 243 00:07:05,480 --> 00:07:06,480 but like a partner. 244 00:07:06,480 --> 00:07:06,920 Yeah. 245 00:07:06,920 --> 00:07:09,680 That can help us navigate complex and unpredictable 246 00:07:09,680 --> 00:07:10,640 situations. 247 00:07:10,640 --> 00:07:11,800 I like that a partner. 248 00:07:11,800 --> 00:07:13,080 And all of these predictions. 249 00:07:13,080 --> 00:07:13,440 Yeah. 250 00:07:13,440 --> 00:07:17,440 Agentic AI, synthetic data, advanced reasoning, 251 00:07:17,440 --> 00:07:18,680 generalization. 252 00:07:18,680 --> 00:07:20,120 All pointing towards. 253 00:07:20,120 --> 00:07:22,200 They all seem to point towards this idea 254 00:07:22,200 --> 00:07:23,160 of superintelligence. 255 00:07:23,160 --> 00:07:24,240 Yeah, ultimate goal. 256 00:07:24,240 --> 00:07:24,560 Right. 257 00:07:24,560 --> 00:07:25,160 Yeah. 258 00:07:25,160 --> 00:07:27,120 And that's where things get really interesting. 259 00:07:27,120 --> 00:07:30,240 Things get really interesting because superintelligence, 260 00:07:30,240 --> 00:07:34,480 as he describes it, isn't just about being really smart. 261 00:07:34,480 --> 00:07:37,440 It's about surpassing human intelligence 262 00:07:37,440 --> 00:07:39,160 in virtually every aspect. 263 00:07:39,160 --> 00:07:41,280 So not just AI that can beat us at chess. 264 00:07:41,280 --> 00:07:41,800 Right. 265 00:07:41,800 --> 00:07:45,800 But AI that can solve problems that we can't even comprehend. 266 00:07:45,800 --> 00:07:47,960 Solve problems we don't even know exist. 267 00:07:47,960 --> 00:07:49,720 It might have insights and perspectives 268 00:07:49,720 --> 00:07:51,360 that are completely alien to us. 269 00:07:51,360 --> 00:07:51,720 Yeah. 270 00:07:51,720 --> 00:07:53,960 It really challenges our current understanding 271 00:07:53,960 --> 00:07:55,680 of what intelligence means. 272 00:07:55,680 --> 00:07:57,800 It's both exhilarating and a little daunting. 273 00:07:57,800 --> 00:07:59,200 It is both of those things. 274 00:07:59,200 --> 00:08:01,040 I think a lot of people hear superintelligence 275 00:08:01,040 --> 00:08:02,880 and they immediately think of. 276 00:08:02,880 --> 00:08:03,720 Hollywood as a country. 277 00:08:03,720 --> 00:08:06,880 Yeah, the Terminator machines taking over the world. 278 00:08:06,880 --> 00:08:07,560 Right. 279 00:08:07,560 --> 00:08:11,800 But Sootskiver doesn't really seem to see it that way. 280 00:08:11,800 --> 00:08:12,440 He doesn't. 281 00:08:12,440 --> 00:08:16,480 He actually emphasizes the potential for collaboration 282 00:08:16,480 --> 00:08:19,120 between humans and superintelligent AI. 283 00:08:19,120 --> 00:08:19,680 OK. 284 00:08:19,680 --> 00:08:23,080 He talks about AI helping us solve some of the world's most 285 00:08:23,080 --> 00:08:26,600 pressing problems, augmenting our creativity, 286 00:08:26,600 --> 00:08:28,960 and pushing the boundaries of human potential. 287 00:08:28,960 --> 00:08:30,720 That sounds incredibly optimistic. 288 00:08:30,720 --> 00:08:31,440 It does. 289 00:08:31,440 --> 00:08:33,760 But it also raises the question of control. 290 00:08:33,760 --> 00:08:34,240 Right. 291 00:08:34,240 --> 00:08:37,920 How do we ensure that superintelligent AI with its 292 00:08:37,920 --> 00:08:40,920 potentially vast capabilities remains 293 00:08:40,920 --> 00:08:43,080 aligned with human values and goals? 294 00:08:43,080 --> 00:08:44,680 That's the million dollar question. 295 00:08:44,680 --> 00:08:46,960 And it's one that Sootskiver doesn't shy away from. 296 00:08:46,960 --> 00:08:47,640 OK. 297 00:08:47,640 --> 00:08:51,080 He acknowledges that we need to develop robust frameworks 298 00:08:51,080 --> 00:08:54,120 and safeguards to guide the development and deployment 299 00:08:54,120 --> 00:08:57,240 of superintelligent AI, ensuring that it remains 300 00:08:57,240 --> 00:08:58,760 beneficial to humanity. 301 00:08:58,760 --> 00:09:00,840 It's not just about building smarter AI. 302 00:09:00,840 --> 00:09:01,040 Right. 303 00:09:01,040 --> 00:09:03,360 It's about building AI that's aligned with our values 304 00:09:03,360 --> 00:09:03,960 and goals. 305 00:09:03,960 --> 00:09:05,280 Aligning it with our value. 306 00:09:05,280 --> 00:09:07,960 It's about shaping a future where AI and humanity 307 00:09:07,960 --> 00:09:10,040 can coexist and thrive together. 308 00:09:10,040 --> 00:09:12,360 And that requires careful consideration 309 00:09:12,360 --> 00:09:14,280 of the ethical implications. 310 00:09:14,280 --> 00:09:15,760 Of such powerful technology. 311 00:09:15,760 --> 00:09:16,400 Absolutely. 312 00:09:16,400 --> 00:09:18,800 We need to start having those conversations now 313 00:09:18,800 --> 00:09:21,240 before we reach a point where superintelligence becomes 314 00:09:21,240 --> 00:09:22,160 a reality. 315 00:09:22,160 --> 00:09:23,720 Can't just wait until it's too late. 316 00:09:23,720 --> 00:09:24,520 No. 317 00:09:24,520 --> 00:09:28,400 We need to think about the potential consequences, 318 00:09:28,400 --> 00:09:31,880 the ethical dilemmas, the societal impacts 319 00:09:31,880 --> 00:09:35,600 of such a profound technological advancement. 320 00:09:35,600 --> 00:09:36,960 We need to make sure we're not just 321 00:09:36,960 --> 00:09:38,400 building superintelligence. 322 00:09:38,400 --> 00:09:38,880 Right. 323 00:09:38,880 --> 00:09:40,280 We're building it responsibly. 324 00:09:40,280 --> 00:09:41,640 Responsibly, yeah. 325 00:09:41,640 --> 00:09:43,680 We need to ensure that the benefits of AI 326 00:09:43,680 --> 00:09:45,320 are shared equitably. 327 00:09:45,320 --> 00:09:46,160 Oh, sure. 328 00:09:46,160 --> 00:09:49,480 That we don't create new forms of inequality or injustice 329 00:09:49,480 --> 00:09:50,360 in the process. 330 00:09:50,360 --> 00:09:51,920 It's a huge responsibility. 331 00:09:51,920 --> 00:09:52,400 It is. 332 00:09:52,400 --> 00:09:54,760 But it's one we can't afford to ignore. 333 00:09:54,760 --> 00:09:56,720 It feels like a fundamental shift. 334 00:09:56,720 --> 00:09:57,120 It is. 335 00:09:57,120 --> 00:09:58,680 In our relationship with technology. 336 00:09:58,680 --> 00:09:59,160 Absolutely. 337 00:09:59,160 --> 00:10:02,600 Moving beyond simply using it to actively shaping 338 00:10:02,600 --> 00:10:03,640 its evolution. 339 00:10:03,640 --> 00:10:05,800 Yeah, we're not just passive users anymore. 340 00:10:05,800 --> 00:10:09,280 Especially when it comes to something as potentially 341 00:10:09,280 --> 00:10:11,040 transformative as superintelligence. 342 00:10:11,040 --> 00:10:11,440 Yeah. 343 00:10:11,440 --> 00:10:13,360 Yeah, it is a massive undertaking. 344 00:10:13,360 --> 00:10:15,000 But it's also an incredible opportunity. 345 00:10:15,000 --> 00:10:15,600 I agree. 346 00:10:15,600 --> 00:10:18,200 I mean, Setscaver's vision of the future of AI. 347 00:10:18,200 --> 00:10:20,080 Yeah. 348 00:10:20,080 --> 00:10:22,560 Isn't just about technological advancement. 349 00:10:22,560 --> 00:10:25,560 It's about reimagining what's possible for humanity. 350 00:10:25,560 --> 00:10:26,160 It's exact. 351 00:10:26,160 --> 00:10:29,920 Using this incredible tool to solve problems, 352 00:10:29,920 --> 00:10:33,800 enhance creativity, and ultimately create a better future. 353 00:10:33,800 --> 00:10:36,360 That's what I find so inspiring about his perspective. 354 00:10:36,360 --> 00:10:40,480 He's not just predicting a future where AI is more powerful. 355 00:10:40,480 --> 00:10:43,440 He's envisioning a future where AI helps 356 00:10:43,440 --> 00:10:48,680 us unlock human potential in ways we can't even imagine yet. 357 00:10:48,680 --> 00:10:51,520 But with all this talk of superintelligence and AI 358 00:10:51,520 --> 00:10:54,000 that can reason, act independently, 359 00:10:54,000 --> 00:10:56,040 and even solve its own errors. 360 00:10:56,040 --> 00:10:57,760 There's a question that lingers in my mind. 361 00:10:57,760 --> 00:10:58,880 OK. 362 00:10:58,880 --> 00:11:01,920 What happens to humans in this future? 363 00:11:01,920 --> 00:11:03,440 Do we become obsolete? 364 00:11:03,440 --> 00:11:05,040 That's a valid concern. 365 00:11:05,040 --> 00:11:08,560 And one that's been explored in science fiction for decades. 366 00:11:08,560 --> 00:11:11,800 But I think Setscaver offers a more nuanced view. 367 00:11:11,800 --> 00:11:14,040 He doesn't see superintelligence as replacing 368 00:11:14,040 --> 00:11:15,400 human intelligence. 369 00:11:15,400 --> 00:11:17,240 He sees it as complimenting it. 370 00:11:17,240 --> 00:11:20,000 So instead of a scenario where AI takes over, 371 00:11:20,000 --> 00:11:22,200 it's more about collaboration and partnership. 372 00:11:22,200 --> 00:11:23,400 More like a partnership. 373 00:11:23,400 --> 00:11:26,680 Humans and AI working together to achieve common goals. 374 00:11:26,680 --> 00:11:28,280 Exactly. 375 00:11:28,280 --> 00:11:32,000 Imagine a world where we have AI partners that 376 00:11:32,000 --> 00:11:34,280 can help us tackle climate change, 377 00:11:34,280 --> 00:11:37,520 develop new medical treatments, or even explore 378 00:11:37,520 --> 00:11:39,280 the depths of space. 379 00:11:39,280 --> 00:11:39,960 Wow. 380 00:11:39,960 --> 00:11:44,480 AI that can augment our creativity, expand our knowledge, 381 00:11:44,480 --> 00:11:46,640 and help us understand the universe in ways 382 00:11:46,640 --> 00:11:48,400 we never could alone. 383 00:11:48,400 --> 00:11:50,120 That's a powerful image. 384 00:11:50,120 --> 00:11:53,360 But how do we ensure that this collaboration is 385 00:11:53,360 --> 00:11:55,480 truly beneficial to all of humanity? 386 00:11:55,480 --> 00:11:55,760 Yeah. 387 00:11:55,760 --> 00:11:58,520 How do we prevent the concentration of power 388 00:11:58,520 --> 00:12:00,880 and resources in the hands of those who control 389 00:12:00,880 --> 00:12:02,760 these advanced AI systems? 390 00:12:02,760 --> 00:12:04,760 That's where the ethical considerations we've 391 00:12:04,760 --> 00:12:06,120 been discussing come into play. 392 00:12:06,120 --> 00:12:06,760 OK. 393 00:12:06,760 --> 00:12:09,920 We need to establish clear guidelines and frameworks 394 00:12:09,920 --> 00:12:12,920 to ensure that the development and deployment of superintelligence 395 00:12:12,920 --> 00:12:15,080 is guided by principles of fairness, equity, 396 00:12:15,080 --> 00:12:16,040 and accessibility. 397 00:12:16,040 --> 00:12:18,120 So it's not just about building advanced AI. 398 00:12:18,120 --> 00:12:18,520 Right. 399 00:12:18,520 --> 00:12:20,200 It's about building it responsibly. 400 00:12:20,200 --> 00:12:21,040 Responsibly. 401 00:12:21,040 --> 00:12:23,200 Ensuring that its benefits are shared widely. 402 00:12:23,200 --> 00:12:23,480 Right. 403 00:12:23,480 --> 00:12:25,600 And that it doesn't exacerbate existing inequality. 404 00:12:25,600 --> 00:12:26,160 That's excellent. 405 00:12:26,160 --> 00:12:27,280 Or create new ones. 406 00:12:27,280 --> 00:12:29,120 We need to be very careful about that. 407 00:12:29,120 --> 00:12:31,280 It seems like navigating this future 408 00:12:31,280 --> 00:12:34,720 will require a deep understanding of not only 409 00:12:34,720 --> 00:12:37,800 the technological aspects of AI, but also 410 00:12:37,800 --> 00:12:40,600 the ethical, social, and economic implications. 411 00:12:40,600 --> 00:12:41,160 All of it. 412 00:12:41,160 --> 00:12:44,120 We'll need to rethink our education systems, 413 00:12:44,120 --> 00:12:47,240 our labor markets, even our political institutions 414 00:12:47,240 --> 00:12:48,920 to adapt to this new reality. 415 00:12:48,920 --> 00:12:50,440 It's a huge shift. 416 00:12:50,440 --> 00:12:50,840 It is. 417 00:12:50,840 --> 00:12:54,240 The future of AI isn't just a technological challenge. 418 00:12:54,240 --> 00:12:56,160 It's a societal challenge. 419 00:12:56,160 --> 00:12:59,880 It will require collaboration across disciplines, 420 00:12:59,880 --> 00:13:02,600 thoughtful dialogue, and a willingness 421 00:13:02,600 --> 00:13:05,600 to adapt to a world where the boundaries between human 422 00:13:05,600 --> 00:13:09,320 and artificial intelligence are becoming increasingly blurred. 423 00:13:09,320 --> 00:13:12,840 Sutskiver paints a picture of a future 424 00:13:12,840 --> 00:13:15,200 that is both exciting and daunting. 425 00:13:15,200 --> 00:13:15,920 He does. 426 00:13:15,920 --> 00:13:18,120 Full of potential, but also fraught with challenges. 427 00:13:18,120 --> 00:13:19,520 So many challenges. 428 00:13:19,520 --> 00:13:22,440 It's clear that the choices we make today 429 00:13:22,440 --> 00:13:25,600 will have a profound impact on how this future unfolds. 430 00:13:25,600 --> 00:13:26,160 Absolutely. 431 00:13:26,160 --> 00:13:28,320 And that's what makes this conversation so crucial. 432 00:13:28,320 --> 00:13:30,160 We're not just passive observers. 433 00:13:30,160 --> 00:13:30,720 Right. 434 00:13:30,720 --> 00:13:33,640 We're active participants in shaping the future of AI. 435 00:13:33,640 --> 00:13:34,320 We have a voice. 436 00:13:34,320 --> 00:13:35,120 We have a choice. 437 00:13:35,120 --> 00:13:37,800 We do by engaging in informed discussions, 438 00:13:37,800 --> 00:13:40,920 demanding ethical development, and embracing 439 00:13:40,920 --> 00:13:42,720 the potential for collaboration. 440 00:13:42,720 --> 00:13:45,520 We can help guide AI towards a future that 441 00:13:45,520 --> 00:13:48,600 benefits all of humanity. 442 00:13:48,600 --> 00:13:53,840 So as you continue your own exploration of AI, 443 00:13:53,840 --> 00:13:56,880 remember that you're not just learning about a technology. 444 00:13:56,880 --> 00:13:59,800 You're engaging with a force that has the potential 445 00:13:59,800 --> 00:14:01,120 to reshape our world. 446 00:14:01,120 --> 00:14:02,480 A very powerful force. 447 00:14:02,480 --> 00:14:04,480 What excites you most about this potential? 448 00:14:04,480 --> 00:14:05,440 It's a big question. 449 00:14:05,440 --> 00:14:07,200 What concerns you and what role do you 450 00:14:07,200 --> 00:14:09,240 want to play in shaping this future? 451 00:14:09,240 --> 00:14:10,920 Those are questions worth pondering. 452 00:14:10,920 --> 00:14:13,960 As we stand at the cusp of this new era of intelligence. 453 00:14:13,960 --> 00:14:15,360 It's a new era. 454 00:14:15,360 --> 00:14:18,200 It is the future of AI is not predetermined. 455 00:14:18,200 --> 00:14:18,560 Right. 456 00:14:18,560 --> 00:14:20,080 It's being written right now. 457 00:14:20,080 --> 00:14:23,360 And we all have a part to play in shaping its direction. 458 00:14:23,360 --> 00:14:26,200 Well, that's about all the time we have for today's deep dive. 459 00:14:26,200 --> 00:14:27,200 Time flies. 460 00:14:27,200 --> 00:14:29,560 It's been an incredible journey into the world of AI. 461 00:14:29,560 --> 00:14:30,240 It has. 462 00:14:30,240 --> 00:14:33,200 And the mind-blowing possibilities that lie ahead. 463 00:14:33,200 --> 00:14:35,720 We hope this conversation has sparked your curiosity. 464 00:14:35,720 --> 00:14:36,200 Yes. 465 00:14:36,200 --> 00:14:38,600 And inspired you to think deeply about the future 466 00:14:38,600 --> 00:14:39,880 we're creating together. 467 00:14:39,880 --> 00:14:40,880 It's up to us. 468 00:14:40,880 --> 00:14:43,840 Until next time, keep exploring. 469 00:14:43,840 --> 00:14:45,000 Keep questioning. 470 00:14:45,000 --> 00:14:45,800 Keep learning. 471 00:14:45,800 --> 00:14:48,920 And keep imagining the incredible potential of AI 472 00:14:48,920 --> 00:14:50,840 to shape a better future for all. 473 00:14:50,840 --> 00:14:52,840 A better future for all. 474 00:14:52,840 --> 00:14:54,920 Thanks for joining us on this deep dive. 475 00:14:54,920 --> 00:14:55,960 It's been a pleasure. 476 00:14:55,960 --> 00:15:16,120 We'll see you next time.