Task Abstention for Large Language Models in Code Generation
For developers using LLMs for code generation, this provides a reliable mechanism to avoid functionally incorrect code, enhancing safety and robustness.
The paper addresses hallucination in LLM-based code generation by proposing a calibrated abstention rule based on multiple hypothesis testing. The method achieves more accurate and efficient abstention from hallucination-prone tasks compared to existing techniques, with rigorous theoretical guarantees.
Large language models (LLMs) have revolutionized automated code generation. One serious concern, however, is the so-called ``hallucination'', i.e., LLMs may generate seemingly plausible but functionally incorrect code. In this paper, we study the task abstention problem, i.e., determining whether a given LLM should abstain from performing a specific code generation task to avoid likely hallucination. Our approach features a calibrated abstention rule, grounded in the principles of multiple hypothesis testing. The rule assesses generation consistency through code execution outcomes, allowing it to handle syntactic diversity of semantically equivalent code without reliance on oracle test cases or external databases. We prove that our approach provides a rigorous, distribution-free theoretical guarantee on its abstention decisions. We evaluate our method on benchmark datasets using several open-source code LLMs. Results show that our method allows generative models to more accurately and efficiently identify and abstain from tasks that induce hallucination compared to existing techniques, providing a reliable mechanism for safer and more robust code generation.