Gary Sutton

And again, that's why you've got these massive data centers being built because

OpenAI demonstrated this is how you get better performing models.

And so there's a power law relationship between the amount of data that you train a model on and the kind of performance uplift you get from training that model or any model, quite frankly, on additional data and then having the hardware to churn through all of that.

You know, I think I've noticed that different language models have different strengths and weaknesses.

So I often play them off against one another.

Like I'll use one as my primary, but if I'm not really sure of the response that I'm getting, I might share that with another language model.

Oh, yeah, because they're because they're trained not necessarily in the same data.

That's right.

Oh, yeah.

And that's why when you log into ChatGPT, for instance, there's a disclaimer in a small font on the bottom of the screen.

ChatGPT can make mistakes.

Check the results or whatever.

Right, right.

So, yeah, they're covering themselves.

And I'm sure the other language models do the same thing.

Yeah.

So, yeah, it's a little bit – is that any – I mean, you got – you interacted with five different language models.

You could have interacted with five different doctors.

You also could have gotten five different responses from five doctors, right?

Very good point.