aryanyaksh-art/did-it-even-learn

Do community LLM fine-tunes actually improve over their base, or just memorize the benchmark? A contamination-controlled study across hundreds of matched Featherless-hosted pairs.

1 best points · 1 comments · 1 Hacker News post · 1 GitHub stars · View on GitHub

Hacker News discussions

• Fine-tuning your LLM might be making it dumber (1 points, 1 comments, by aryan_yaksh, 2026-09-02)