It seems like the big improvements in the current AI flavor are done. It happened very quickly, and so did the diminishing returns. It's amazing compared to 2 years ago, it's great compared to a year ago, it's not that different to 6 months ago.

Lots of room in optimization and figuring out how to actually use it usefully. But I don't think any one country is now poised to take a leap ahead on this stuff. And Chinese researchers kind of showed that once the technique was out of the bag, catching up wasn't hard.