RL Training Boosts pass@1 on Easy Puzzles but May Reinforce Wrong Modes on Hard Ones

burny_tech · x · 2026-07-22

In a thread, evangelinejy99 analyzes that RL training amplifies correct moves SFT already preferred on easy puzzles, improving pass@1, but pass@16 gains are smaller or negative. On hard puzzles, RL surfaces correct moves from the tail but also reinforces wrong modes.

Original post →

More from Research

• Researchers identify "Forking": train loss falls while validation loss turns up — ZimingLiu11 · 2026-10-26
• Google DeepMind pilots world's first double-blind evaluations for frontier models — scychan_brains · 2026-10-26
• Rewriting scaling law study: optimal synthetic data share rises from 10% to 32% with compute — heghbalz · 2026-10-26
• FLASH: GPU-native deformable sim learns cloth folding in minutes on one RTX 5090 — ChongZzZhang · 2026-10-26
• LEMON: zero-shot LLM misbehavior monitor flags hallucinations from attention maps — yoavartzi · 2026-10-26
• Behind the EMNLP keynote: LMs as simulated learners to generate child language acquisition hypotheses — najoungkim · 2026-10-26

Research channel →