World Wires · story 100688 · corroborated · 1 source(s)

Controlling sharpening in LLM post-training via reinforcement learning

This paper investigates the phenomenon of sharpening during LLM reinforcement learning post-training and explores mitigation strategies.

Open in the desk

Coverage

• Controlling sharpening in LLM post-training via reinforcement learning arXiv: new preprints (security, AI, physics) · tier C · clear net · 2026-11-02 13:54 UTC

What this site indexes