Carl

Obviously, the thing that's... There's a reason why OpenAI wants to invest in Stargate or the partnership with Oracle and NVIDIA is they need the compute, right?

Because they have all the... I'm thinking more of the inference cost of everyone doing searches or whatever it is, right?

And he was mentioning that...

Because I think the question was around, like, do you have enough compute or something like that?

He said a majority of the compute actually goes into the research.

And if they had to choose research or product, it would go to research first, then go to product.

Which is, if you think about, like, the interesting part of that is that's why maybe the products you get are

If they let's say if they move the compute, they reversed it, right?

The majority of the compute goes to product.

Then maybe you get like he said in a previous one before, he's like, we can't even show you some of the models because there's no way we you would be able to run it because we don't have the to run it.

Right.

And I think that's probably true for a lot of companies.

I'm about to swear because the amount of rate limiting I get from even Gemini and Anthropic is outrageous that I have to go into the IDE and Claude code because it doesn't rate limit me as much as the web apps.

It's just, it's so frustrating.

And that's why it's hard for me to recommend products like Claude and Gemini at the moment because it's,

Imagine you're just, you know, you're not in the AI space and it's like the notice and you're like, what?

Like, why can't I continue going?

Business runs like that.

So but it's interesting, though, right?

Because like that leads to what Jan LeCun was, why he left Meta, right?