Zack Kass

And so you couldn't have lined up these things better.

ChowGBT, which was not a breakthrough on a research basis, it was GBT 3.5, slightly aligned in an application,

gave people the tool to be like, oh my God, I get it.

I understand why this matters.

Just in time for us to actually release a truly phenomenal underlying model, GPT-4.

And that GPT-4, once it comes into ChatGPT, then people go, whoa.

And that's when you see we go from 10 million to 100 million to a billion downloads.

It was certainly overwhelming.

It was probably much more overwhelming to the engineering team than almost anyone because there was incredible load dependency.

I don't think it was until GBT4 that things really went crazy.

It wasn't probably until March.

Keep in mind, obviously, a lot of people paid attention at that point, but it wasn't until GBT4 and the model got really good that

That people were like, whoa, because for a while still people were complaining about, oh, it's slow or the text isn't good enough, whatever.

And then we solved a lot of these issues.

And that's when it really went, you know, it got big.

The exact numbers, no, but it was like a million in the first two weeks, 10 million in the first

three months, and then it was a sprint to a hundred million and then a billion within the year.

There was a debate on whether there was a product

whether the model was a product.

And the API, of course, has turned out to be a great product, but there was some debate whether it was actually something you could sell.