Inference

Theories of Sense Perception.

Ideal, Inferential, Phenomenal, and Relative.

Natural Realism.

The Ideal Theory.

So I spent my first four or five years actually working on pre-training efficiency all the way from GPT 3.5 to our latest models.

And over the past few years, I've seen like in France become more and more and more important.

As you know, our products have hit the market and the industry has grown.

And so it's

except getting like a DGX station, that's like $100,000, right? So like the jump is pretty massive.

Now, why should we be thinking about this local inference capacity and people needing it? I think that it's really important for the tech industry in general, if we want AI to actually be a part of people's lives and for it to actually permeate everything people do, people are going to need local inference capacity.

The same way that we needed people to have powerful CPUs and then powerful GPUs to get just technology to work, Inference is an important part of that.

The reason that that's really being throttled right now and the costs are so high, it's mostly around RAM prices.

And I would go a bit further than that, which is I think the frontier labs have artificially limited what people can get and and artificially up the prices by spending so much of the investment burn rate that they're getting into pre buying all of the capacity anybody could ever get their fucking hands on for a couple of reasons a you know when it comes to the neo clouds and like the actual big gpu deployments i think it's an anti-competitive thing Right.

Theories of Sense Perception.

Ideal, Inferential, Phenomenal, and Relative.

Natural Realism.

The Ideal Theory.