Jason Lampkin

It's easy to say it's more than an order of magnitude by the end of the year potentially.

You know, the cost decline that we're also seeing, despite the anthropic thing, it's hard.

The token consumption is increasing the deflation in the per token cost.

So we've seen that improvement, I think, in anthropic to Rory's point.

And I might be getting it wrong, but we haven't seen this break point where we're catching a break.

Maybe the margins are getting better, but we're going to keep burning more.

So I'm waiting for the moment.

But, you know, we've talked about memory, but ChatGBT and Claude don't have that much memory.

If you try it, don't remember much, does it?

It remembers this little bit.

So I built a version of Claude over the weekend called Wren.

You can try it at Wren, W-R-E-N, chat.ai.

It remembers everything.

So it never forgets anything.

And I learned a lot of things, but one of them is it's going to burn a lot of tokens.

You want to save every chat you've ever had, every discussion over all time and reference it for years.

And what if that runs 24-7?

Maybe I'm rambling and not answering your question, but I don't see, I just see it accelerating.

And again, my biggest concern for founders out there, especially for folks that are not quite in the top 0.01%, right?

Especially folks in ops on their team or folks that aren't close to AI that they're mismodeling this.