Mustafa Suleyman

Yeah, I think we're obviously, there's massive market incentives to make this as cheap as possible, to make it efficient.

There's a lot of competition in the market.

So I think that inference costs, the cost of providing access to these models for an individual user are just coming down like crazy.

I think the cost of serving a token has already come down 100x in the last two years or two and a half years.

I think it's a remarkable transition to see that these are getting made widely available to everybody at such pace.

And I think that's a good thing because everybody's going to get access to amazing productivity gains roughly all at the same time or within two or three years.

So that's going to be very democratizing.

I mean, Copilot is obviously my number one tool.

Obviously, I use it on Windows.

Vision is an incredible experience.

I also use all of the other models and try out everything else.

But Copilot is the model I'm building.

So that's my go-to.

Thank you, Grant.

Thank you, Corey.

This was super fun.