Sharpening Tax: Post-Training Sharpens Old Skills Rather Than Teaching New Ones

Researchers including Meta propose the "Sharpening Tax," showing that RL post-training mainly sharpens existing skills: it boosts pass@1 but sacrifices pass@K coverage compared to pretrained models with a lightweight inference harness.

2026-10-02 ~ 2026-10-02 · 2 related posts

• Sharpening Tax: Does post-training teach LLMs new agentic skills or just sharpen old ones? — SharonYixuanLi · 2026-10-02
• Meta Quantifies the 'Sharpening Tax': RL Post-Training Trades pass@K Coverage for Accuracy — meta · 2026-10-02