CROCODIL Framework Reduces Code Over-editing
Developers using multiple AI coding assistants will benefit from CROCODIL's reduced over-editing behavior that maintains code quality across different models.
Researchers discovered that coding models over-edit code written by other models due to different training data producing stylistic preferences.
CROCODIL is a post-training framework that addresses this behavior using a similarity reward and an execution reward.
The framework multiplies rather than adds these rewards to prevent the policy from shrinking edits by simply failing the task.
The paper is available at academy.dair.
ai/papers/crocodi..
.
This is a weird behavior in coding models and something worth looking into. It turns that some models over-edit code that another models wrote. There is a high chance that your repo now has commits from more than one model, and that changes how each of them edits. Researchers measured what happens when one model edits code another model wrote. Different training data produces different stylistic preferences, and models make more edits, often excessive ones, on foreign code than on their own. CROCODIL is a post-training framework that reduces that behavior. A similarity reward penalizes large changes and an execution reward scores build and test success, and the two are multiplied rather than added. That product stops the policy from shrinking edits by simply failing the task. Paper: academy.dair.ai/papers/crocodi…📊 18 ⚡
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include CROCODIL, coding models, code editing, post-training framework.
Source attribution: curated from elvis (@omarsar0).
