ML//code generation//CriticGPT

- OpenAI model trained to spot bugs in LLM-generated code.

OpenAI model trained to spot bugs in LLM-generated code.

LLMs produce code with subtle security issues, logic errors, and silent failures that pass basic tests.

CriticGPT catches errors human reviewers miss, especially in complex codebases.

Anthropic's parallel: use SAEs (mechanistic interpretability) to identify features that cause buggy generation.