CROCODIL Curbs LLMs' Over-Editing of Other Models' Code

The CROCODIL paper finds LLMs over-edit code written by other models due to differing style preferences, and proposes a post-training framework that halves such unnecessary edits.

2026-09-06 ~ 2026-09-07 · 3 related posts

• Models over-edit code written by other models; CROCODIL training framework fixes it — omarsar0 · 2026-09-06
• Models over-edit code written by other models, researchers find — pbaylies · 2026-09-06
• CROCODIL paper: training for minimal diffs halves cross-model edit distance on Olmo3 7B — rohanpaul_ai · 2026-09-07