Andy Halliday

By the way, there's a great write-up about that, which is why it makes me laugh because the newsletter literally came out yesterday.

There's a great write-up about that in the newsletter that Andy and I worked on over the weekend.

But anyway, the reason I bring it up is not for one, any particular story.

It's that I really saw like two or three over the last couple of days.

I just don't have them up on my screen where it was just more stories, not at that level, but they were like,

software enhancements, it would be like the art of the blog title would be something like new advancements cuts inference rate by 1.8%, right?

Or whatever it was, right?

You know, or one to 1.8 times more, more effective work inference, whatever.

And I just see this coming up more and more.

Maybe it's because you would just talk about that Google story.

So it's in the back of my mind.

But then if you couple those stories of basically not needing a new NVIDIA chip necessarily, the latest and greatest, what is it, N100 or T100 or whatever, to get more out of it, because that's kind of what you were saying with Google, then we are seeing these other improvements in inference and other areas that are more software-related

And then you couple that with stories like we talked about it last week, but Cursor has Composer 2.

And one of the things that came out with that, but it was sort of brief, was that it was basically self-improving.

Well, they had a post on X over March 26.

Well, I guess it was a few days ago.

I just happened to catch it.

But it's like, we're sharing additional research on how we train new checkpoints with real-time reinforcement learning.

We can ship improved versions of the model every five hours.

And I just think this is Cursor 2, right?