1 paper
Chaitanya Ravuri, Saman Amarasinghe
Modern code-generation LLMs can already solve a large fraction of programming problems, yet they still hallucinate subtle bugs that make their outputs unsafe for autonomous deploym…