[benreesman]

https://chat.lmsys.org/

is tons of fun for playing with this. I ran the Jon Rahm Masters thing through a dozen or so times:

- a couple of the claude models get it - a couple of the llama2 70b derivative models get it - the tulo DP 70b model explain how it got it

GPT-4 is very big and very heavily subsidized, but the other big ones can do all this stuff. I'm aware the authors of the papers know that "GPT-4" is a seller, and so the title is technically accurate, but if we as a community aren't going to push back against Open Philanthropy's dirty-money K-Street-to-Wall-St-to-University-Ave routine, who the hell is?