Bullshit benchmark for LLMs(twitter.com)

1 ポイント投稿者 gpvos5 か月前1 コメント

https://twitter.com/petergostev/status/2026396163637731794

1 コメント

noemit5 か月前

The underlying data looks scarce. If there's only a few questions per "category" of bullshit they can easily be gamed to favor one model over another.