Models over-edit code written by other models; CROCODIL training framework fixes it
omarsar0 · x · 2026-09-06
Researchers found a weird behavior: coding models over-edit code written by other models. Different training data produces different stylistic preferences, so models make more—often excessive—edits on foreign code than on their own, and multi-model commit history now changes how each model edits.
CROCODIL is a post-training framework that reduces this: a similarity reward penalizes large changes, an execution reward scores build/test success, and the two are multiplied rather than added, preventing the policy from shrinking edits by simply failing the task.
Related event: CROCODIL Curbs LLMs' Over-Editing of Other Models' Code(3 posts)→
More from coding & agent
- 105 Real Bugs Tested: GPT-6.1 Sol Over 10x Cheaper Than GPT-5.6 Sol — PawelHuryn · 2026-10-23
- Dev builds SkiFree-style sled game in one prompt, praises DeepSeek Flash — victormustar · 2026-10-23
- Every ditched personal agents for a shared Claude Managed Agents coworker in one week — every · 2026-10-23
- WorkOS shows how to let AI agents safely sign up for your app with auth.md and ID-JAG — AI Engineer · 2026-10-23
- Devs on Reddit swap tactics for stopping prompt regressions in AI agents — SleepFlimsy6963 · 2026-10-23
- Claude Code + ComfyUI MCP workflow for client work: paid models or open weights? — newaccount47 · 2026-10-23