INDICT: Code Generation with Internal Dialogues of Critiques for Both Security and Helpfulness

TL;DR

INDICT introduces a dual-critic framework with internal dialogues to enhance code generation safety and helpfulness in large language models, significantly improving output quality across multiple tasks and languages.

Contribution

The paper presents a novel dual-critic system with internal dialogues for safety and helpfulness, improving code generation quality in LLMs across diverse tasks and benchmarks.

Findings

Significant +10% improvement in code quality across models

Effective safety and helpfulness critique system

Applicable to multiple programming languages and tasks

Abstract

Large language models (LLMs) for code are typically trained to align with natural language instructions to closely follow their intentions and requirements. However, in many practical scenarios, it becomes increasingly challenging for these models to navigate the intricate boundary between helpfulness and safety, especially against highly complex yet potentially malicious instructions. In this work, we introduce INDICT: a new framework that empowers LLMs with Internal Dialogues of Critiques for both safety and helpfulness guidance. The internal dialogue is a dual cooperative system between a safety-driven critic and a helpfulness-driven critic. Each critic provides analysis against the given task and corresponding generated response, equipped with external knowledge queried through relevant code snippets and tools like web search and code interpreter. We engage the dual critic system in both code generation stage as well as code execution stage, providing preemptive and post-hoc guidance respectively to LLMs. We evaluated INDICT on 8 diverse tasks across 8 programming languages from 5 benchmarks, using LLMs from 7B to 70B parameters. We observed that our approach can provide an advanced level of critiques of both safety and helpfulness analysis, significantly improving the quality of output codes ($+10\%$ absolute improvements in all models).