Transcript of ARE AI VALUATIONS A BUBBLE?
Nikhil Kamath Clips
0:00I sit far away from0:02what's happening in the world of AI,0:04but I interview a lot of these people0:06like Sam, Elon, Dario, and all of them.0:09What do you make of it? Like me, the0:11stock trader investor,0:14I'm starting to feel at this point that0:18if I were to take every private company0:20in AI0:21and short0:23their stock today, in 5 years I might0:26make money.0:28>> Okay. Why why do you feel that?0:30Do you think it's a bubble or what?0:33>> It feels a bit like0:35the internet bubble.0:36Uh I'm not saying AI will not have a use0:39case, but0:41again, this is a very like immature0:43understanding of somebody who does not0:46who's not an expert at the science. I'm0:48not an engineer or0:50in that domain,0:52but I feel like all of these models will0:54get to a point where they're very0:55similar to each other.0:57>> Mhm.0:57>> Uh0:58either by copying or reverse engineering1:00or whatever, they will come a point1:03where1:04these models become so similar to each1:07other that1:09more countries will go the China way1:11where1:12I think the world was getting1:14globalized. The next phase is the world1:16gets fragmented. India will have its own1:19copy of the model. Another country will1:21have its own copy.1:23>> Mhm.1:23>> Uh the tokens, the energy, all of that1:25will sit domestically within our1:27country. It might not be the at the1:30cutting edge, but it'll do the job.1:32If the world goes in that direction, I1:35don't see the reason to pay1:37the multiples that these private1:39companies have today.1:41>> Yeah, I think I see what you mean. I I1:43think1:44I probably somewhat agree in the sense1:46that1:48I mean, right now the1:49the demand for intelligence seems to be1:51unlimited. So, I think that's going to1:53continue,1:54but1:56you know, the open-source models are1:57really like 6 months behind and they're1:59like 99% cheaper or more sometimes for2:02inference. So, I think it's entirely2:05possible that um a larger percentage of2:08the workload goes to these models that2:10are like 99% cheaper. It's kind of like2:13what Moore's law happened with chips,2:14right?2:15Even more dramatic.2:17And2:19you know, the chip industry didn't like2:21it's a it's a very important industry.2:22It's had a good run recently cuz of this2:24new wave, but it's not um2:26just going to grow infinitely forever,2:27right? So, there is going to be some2:30world we get to where um okay, there's2:33like these very powerful frontier models2:35that are useful for discovering new2:36physics or something, but 99% of people2:39are just running on um2:41like they become price sensitive at a2:43certain point and the models get good2:44enough and cheap enough where you can2:46run it on commodity hardware. Um2:50so yeah, I don't know what the terminal2:51value of these AI company and and then2:53it's just like any technology2:54revolution. Like where is the value2:55going to be captured? Is it in the2:58is it in the frontier models or is it3:00really in the people building chips and3:01energy?3:03I mean the the the coding harnesses to3:05me don't feel like that defensible. Um3:08you know, they're that you could some of3:10them have actually been leaked on the3:11open source. I mean, you could actually3:13like I think software's basically free,3:14so you could recreate a coding harness3:16like the Anthropic coding harness, I3:17think.3:18Um so, it's like3:21yeah, I you know, it makes me a little3:22nervous when I see these valuations3:24growing this fast as well. Like I've3:26I've seen things like this happen before3:27in crypto and they you know, they3:28correct and then they there's real value3:30under it, so you then they grow later.3:33All markets seem to repeat these cycles3:35with the internet and everything. Even3:37going back to railroads and3:39>> Yeah.3:39>> canals and everything.3:40>> I agree with that. In a way, I'm rooting3:42for it.3:43>> Yeah.3:44>> Like a correction? The one thing I do3:46hope3:49you know, especially in India, this is3:51again personal to me, but since I am3:53from there, I'm batting for that region.3:56We had this phase where3:59there was Microsoft available and there4:01was Linux available. They were both4:03almost equally the same thing, but4:07getting stuck with a Microsoft4:12for a few years when they don't charge4:14much takes you to the point where they4:16begin to charge and it becomes harder to4:18move away from it.4:20>> Mhm.4:20>> I hope people building in India pick4:23open-sourced4:24over these models which are4:27charging a fee because the last thing4:28India needs is, you know, a kid paying4:30$20 a month to another American company,4:33especially if you're building4:35a business which is a application layer4:38on any of these models,4:39>> Mhm.4:40>> you don't want to be held hostage4:42to these guys4:45when tomorrow the4:47the4:48thing that was 6 months behind has4:49caught up, but you're stuck because they4:52can get you stuck.4:54>> Yeah.4:55>> Yeah. I don't know how to I don't know4:57how to articulate this better.4:59>> Yeah, I mean, I'm a fan of open source.5:00I think um5:02I'd like to see the open source models5:04continue to evolve and build and get5:05better and better.5:07Um5:08yeah, so and then I also appreciate5:10companies who put in all the time and5:12there was like a thousand details to get5:14right to make something good, you know,5:15it's like5:16um5:18having Mac OS or something like that,5:20you know, versus an open source that5:22didn't it the5:24I'm glad Mac OS exists, you know, these5:26kinds of things. So, it's a good it's a5:28good healthy balance, you know.5:30>> Yeah.5:31>> What are the are there any companies in5:33India focusing on open source models5:35that are particularly good?5:37>> Building on building an application on5:39top of open source?5:40>> Or or yeah, or making models themselves?5:44>> There's Sarvam, which is5:47India's answer to a large language5:49model.5:50>> Mhm.5:51>> I think I hope they do well.5:53>> Yeah.5:54>> Yeah. I hope they do well for all of our5:56sakes.5:57Uh5:58But I think that's the thing. If you're5:59building open-source model, I don't I6:01don't know if it matters who built it.6:04>> Mhm.6:04>> As long as you can6:07get the most productivity out of it.6:09>> Yeah.6:10>> Right? I think that's brilliant. And6:14that's one when I when I think about6:16these things, that's one way I think of6:17it.6:19The other path it might take is if these6:22companies keep getting any bigger, they6:23get nationalized and we go into some6:26form of socialism where the productivity6:29gains of this are redistributed in some6:32UBI kind of way.6:34>> Yeah.6:35I'm very I'm very skeptical of that,6:36too. Whenever6:38whenever somebody says, um "This is so6:41scary.6:42You You Everybody should have to come in6:43and get a license from the government."6:45I'm like,6:46"Well,6:47the government does a lot of things very6:48well, but the the one thing they're not6:49great at is like understanding the very6:51frontier of technology, and that usually6:53just slows things down, and it makes it6:55better for the incumbents, and it kills6:56a bunch of startups. So, I I agree with6:58you on that."
1,248 words · 220 lines
