AI demand outpaced forecasts for compute, chips and power

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 1:08:39

   Five years ago, it was like... Five years ago, whoever is at the top most in AI, I don't even consider myself like that, but whoever is at the most frontier level of decision-making in AI, five years ago, I don't think they predicted the exact state we are in today. Nobody did. If they did, they would have already procured all the compute and, like, you know, manufactured all the chips, bought out all the fabs.

2. 1:09:02

   They have done all that, right? Just this is counterfactual. Everyone's, like, bottlenecked by not having enough compute and, like, we don't have enough chips, we don't have enough power. These are all the problems that if you invite anybody in AI and ask what is a bottleneck in AI today and everybody would say power. I think Jensen was here and he said the same thing, right? Yeah. But, okay, like, if you predicted this exact state five years before, Or wouldn't you have secured enough power and started building more power plans yourself and started getting permits and started planning out capacity?

3. 1:09:34

   Nobody did that. Everything is reactive to the demand that we're having today. And that's just five years? Yeah, that's just five years. So when you ask me to predict 250 years, I just have to honestly say I don't know. Do you ever sit back and think about it, though? I don't think about it. What it could be? I don't think about it. So there are a lot of fun. I use perplexity a lot for these kind of things, especially this new feature computer inside it.

4. 1:09:59

   And this is just for hypothetical scenarios. Let's say there is an AGI, right? I've seen you ask a lot of people about this. And a lot of conventional answers is like, oh, like we'll just become managers of the AIs. Don't worry. But if the price of cognition is the price of compute, managing an AI is also pretty much doable by the AI itself. Because the bottleneck is not like unique cognition capability there.

5. 1:10:32

   so the value of the society will automatically shift to what is scarce and uh fundamentally what has been scarce is like asking like high quality questions about things okay like what if like we just completely spend all our time understanding the past like that's an interesting endeavor it was not cool before but it'll it'll become cool again um and like we usually use the view like archaeology or history is not something that's like worth having a career in because it doesn't pay well but what if it actually starts paying you a lot more now that like actual knowledge works being done by AIs and like it's all mundane and all the price of that is basically zero right and archaeology would be one of the few things that it wouldn't have access to because it doesn't have the actual ground it can't get into the ground and do the scans and well let's say

6. 1:11:21

   we have like robots to go do that but but you're still going to be the one probing right because you have incomplete information all the time even the idea of like okay let's go explore this particular area let's go understand better let's let's go try to reverse engineer let's go try to build this again how would it be if we wanted to do the same thing on the moon there are like so many interesting projects to work on for us as long as we are we stay curious and we stay interested in like a lot of things that we've done before and trying to understand like civilization that I'm not really concerned about what things we get to do. We might be doing a lot more cool things for what it's worth.

7. 1:11:59

   I don't know if anybody will be coming and telling you that, oh, it's so cool to open an Excel sheet every day and make financial models. There's got to be somebody out there that likes that. I mean, there's something about the task you do and what you get paid for, what is the job title, blah, blah, blah. And some people associate their personal worth with where they work at and how much they get paid.

8. 1:12:23

   And I think that thing is going to collapse in a world where, like, the price of all that cognition is going to be the price of compute. What do you think happens to people if a large percentage of jobs get replaced by AI? I think they'll find new things. We've always gravitated towards things that are scarce because that's where the value lies. And so if, you know, one interesting analogy is the Gulf states where there's an abundance of resources and they export their resources to other states, and that pays for the whole state.

9. 1:13:08

   You know how, like, they offer everybody free electricity, subsidized health, subsidized education, and, like, no taxes? when i first went to dubai like in um almost like 20 years ago um they told me like people don't pay taxes here and nobody pays for electricity here and uh education is like super cheap and i was like wait how is that real and and um And the way that's real is that, I mean, of course, Texas also has no taxes and, you know, any well-run state can do this.

10. 1:13:42

    But the way it's happening is that because the government provides you all these things, it becomes a rentier state. Like, you offer political acquiescence to the state. And what ended up happening is citizens there expect the state to find them jobs, expect the state to take care of, like, job displacement for them. So they don't worry. so it made them a little more lazy so that's not a good future to have where some people talk about AI subsidies and AI dividends that get paid to everybody I think we need to do some form of that but that in entirety won't solve the problem Right, well the thing about human nature is sort of undeniable and if you give people the ability to be lazy a large percentage of people will take that That's right A large percentage won't though There's going to be enough people that are inspired to do something and they say, okay, well, now my basic needs are taken care of.

11. 1:14:37

    Let me pursue my actual interests and find purpose in that. Because a lot of people find purpose in whatever their occupation is. And if we can shift that to finding purpose in what your actual interests are and then really pursuing something, whatever it is in that, then you'll still have meaning in your life. It keeps coming back to staying curious. Yes. and and and finding value in your relationships your your the family uh caring for each other um if you ask a lot of retired people actually retired people is a good demographic to understand what would happen what what are things people find meaning in after like work's taken off them and all majority of the answers are always like family caring you know personal like like like relationships and community.

12. 1:15:26

    Like these are the things that retired people keep doing to like, you know, keep themselves active and wake up every day and have something to look for. So all those things will become even more important at a time when like work itself doesn't mean much. Doesn't mean humans won't be status seeking. I think we'll still be. But status is not gonna come from whether you're working at, you know, like a particular famous bank or a tech company or whatever.

13. 1:15:53

    It'll be driven by, like, how interesting you are. Are you interesting to talk to? When I can talk to an AI, like, despite that, are you still interesting to talk to? Are there certain things I get out of talking to you that completely change my perspective about a bunch of things? Or is it just fun to hang around you? Can we have a compounding relationship together? And I think, again, it goes back to, like, you know, being curious about things.

14. 1:16:19

    Well, this is best case scenario, right?