Teaching Large Language Models to Self-Debug
First page
Paper summary
Teaches LLMs to debug their own code via few-shot demonstrations.
Ask this paper
01
Self-debugging via explanation: LLMs identify mistakes by explaining their generated code in natural language, then iteratively fix errors.
02
Few-shot teaching: Requires only a handful of debugging demonstrations to enable the capability across tasks.
03
Text-to-SQL SOTA: Achieves state-of-the-art on several code generation tasks including text-to-SQL generation.
04
Self-correction research: Influential paper establishing self-debugging as a distinct capability, informing 2024 reasoning + self-correction agents.