Get Started
Research questionHow can LLMs edit another model’s code while preserving functionality and minimizing unnecessary changes?Models trained on different data can have different coding styles, causing an LLM to rewrite more of foreign code than the requested change requires. The challenge is to limit these unnecessary edits without causing build or test failures.
AI
Code Generation & Program Synthesis
LLM Pretraining & Post-training
Latest papersRecent research connected to this question, newest first.CROCODIL: Cross-Model Code Editing with LLMsThe source reports excessive edits on code written by another LLM and presents CROCODIL, a post-training framework that combines similarity and execution rewards. Its evidence concerns LLM-based code editing, with functional success assessed through builds and tests.research paper · Sep 3, 2026
Related questions
How can LLMs generate functionally correct code without introducing security vulnerabilities?How can LLMs synthesize executable machines and improve their designs through physical simulation feedback?How can LLMs generate correct routines when correctness depends on their joint runtime behavior?How can LLMs remain faithful to provided context without doubling inference cost?
Home
Topics
Search
Library