Research map / Large language models / Agents, code and math
Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| augment HumanEval tests 80x with LLM seeds + type-aware mutation | GPT-4 evaluation · HumanEval | pass@1 greedy: 88.4 -> 76.2 | Hurt |
| augment HumanEval tests with EvalPlus | ChatGPT evaluation · HumanEval | pass@1 greedy: 73.2 -> 63.4 | Hurt |
| augment HumanEval tests with EvalPlus | 26 LLMs · HumanEval | pass@k: up-to 19.3-28.9% | Hurt |
| augment HumanEval tests with EvalPlus | model ranking · HumanEval | ranking vs ChatGPT: Phind 71.3->67.1 vs ChatGPT 73.2->63.4 | Mixed |
| reduce test suite using only branch coverage (ablation) | HumanEval+-mini · HumanEval+ (GPT-4) | pass@1 greedy (lower=stricter): 78.0 -> 86.0 | Hurt |
| reduce using only mutant killings (ablation) | HumanEval+-mini · HumanEval+ (GPT-4) | pass@1 greedy (lower=stricter): 78.0 -> 82.9 | Hurt |
| reduce using only LLM sample killings (ablation) | HumanEval+-mini · HumanEval+ (ChatGPT) | pass@1 greedy: 65.2 vs 65.2 | No clear effect |
| minimize test suite 47x via set cover | HumanEval+ · HumanEval+ (ChatGPT) | pass@1 greedy: 63.4 -> 65.2 | No clear effect |
Closest papers by meaning
- How efficient is LLM-generated code? A rigorous & high-standard benchmark (ICLR 2025 · similarity 0.92)
- Can LLM Replace Stack Overflow? A Study on Robustness and Reliability of Large Language Model Code Generation (AAAI 2024 · similarity 0.90)
- Rethinking Verification for LLM Code Generation: From Generation to Testing (NeurIPS 2025 · similarity 0.90)
- LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code (ICLR 2025 · similarity 0.89)
- EffiBench-X: A Multi-Language Benchmark for Measuring Efficiency of LLM-Generated Code (NeurIPS 2025 · similarity 0.89)
- ConvCodeWorld: Benchmarking Conversational Code Generation in Reproducible Feedback Environments (ICLR 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.