logoalt Hacker News

djeastm • today at 9:20 PM • 1 reply • view on HN

ts-rust was his benchmark for "task he once thought was impossible for AI to do". Opus, however, was able to do it, apparently. So he's seeing now if the latest OpenAI model can improve it at all. Seems like he's not impressed with the results.


Replies

suddenlybananas • today at 10:07 PM

but it's quite unclear what he is actually asking the models to do. I don't understand how you spend 300k without getting anything.