Hacker News new | past | comments | ask | show | jobs | submit login

I think part of the problem was many/most previous models coming out of China were trained on eval data to cheat in the rankings. Quite an uphill battle for Deepseek.





I think part of the problem was many/most previous models coming out of the US were trained on eval data to cheat in the rankings.

Assuming that is true, which I don't have a reason to believe otherwise, only gave an excuse to the average reader to disregard without reading more than the title.

You can see a similar bias in academia with work originating outside EU/USA.

before someone thinks something strange regarding me, I can only tell you I'm not chinese, but Argentinian :)




Consider applying for YC's Spring batch! Applications are open till Feb 11.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: