ExecuCritic: Calibrated Critic Shaping for Code Generation with Verifiable Rewards
Source
Junjie Cao, Yingjie He
Author
Junjie Cao, Yingjie He
Date
Terms in this piece · Glossary
calibration — How well a model's confidence matches reality — a calibrated model saying "90% sure" is right about 90% of the time.
sandbox — An isolated environment where AI-generated code or agent actions run without being able to touch anything real.
Why it matters
ExecuCritic jointly trains a code-generation policy and a critic on the same execution rollouts, crediting the coder only when the critic's verdict matches the executor; this beats GRPO and prompted-reviewer baselines with fewer sandboxAn isolated environment where AI-generated code or agent actions run without being able to touch anything real.Full definition → runs.
articleCriticGen: Generation-Aware Evaluation as Actionable FeedbackHuifang Du, Zecheng Zuo, Sen Wang, Chenghao Fan, Haofen Wang, Yehui Yang
articleGraphAlignCoder: Aligning Program and Proof Graphs for Code GenerationYueke Zhang, Zihan Fang, Kevin Leach, Yu Huang
articleTIPCODER: Reinforcement Learning Boosted Test-time Instruction Proposer for Code GenerationMinyu Chen, Sihao Wu, Ling-I Wu, Song Qin, Jingyang Li, Lei Ning, Jianxin Xue, Guoqiang Li