Information gain rewards improve reliable AI code generation tests
Information-Gain Rewards over Diversity-Pruned Tests: GT-Anchored Verifier Co-Training for Reliable Code Generation
Artificial Intelligence
Summary
Generating code using AI can be tricky because the tests that check if the code works are sometimes too simple or too similar, which leads to unreliable results. The authors designed a new method called CoVer that helps the AI create better, more varied tests by rewarding tests that provide useful information about whether the code is truly correct. They also use a smart way to pick diverse tests so the AI learns from many different examples. This approach improved code generation accuracy on various benchmarks compared to previous methods.
What this means in practice
- •For software development teams: Generate more reliable code by integrating AI that produces and selects diverse, informative tests to better assess code correctness.
- •For automated code review tools: Improve the accuracy of automated code assessment by using co-trained AI systems that generate meaningful tests during review.
Authors
Ana Nunez, Peyman Najafirad
Abstract
Self-play methods that co-train a single language model as both coder and test author promise to move code-generation RL beyond fixed test suites, but they suffer from two coupled pathologies: permissiveness collapse, where pass-rate rewards are maximised by trivial, non-discriminative tests, and concentration bias, where i.i.d. sampled tests cluster on modal inputs and inflate estimator variance. We introduce CoVer (Co-trained Coder and Verifier), a single-policy GRPO framework that addresses both failure modes. First, an information-gain (IG) reward scores each self-generated test by the mutual information between its pass/fail vector and a graded, ground-truth-anchored correctness signal y [0, 1] m, gated by the sign of their covariance so that only positively discriminative tests receive reward. Second, a three-stage diversity-aware selection step prunes a candidate pool to a behaviourally non-redundant suite (invalidity, input-string, execution-profile filtering), raising the effective sample size of the IG estimator at fixed execution budget. On five benchmarks (LiveBench, MBPP, LiveCodeBench, CodeContests, Code-Forces), CoVer raises one-shot pass@1 by +5.8 points at 7B and +7.1 points at 14B over the Qwen2.5-Instruct backbone, and achieves the highest macro-average among all compared methods at both scales. As a drop-in backbone inside the CodeT ranking pipeline, CoVer-7B adds +3.5 points, demonstrating the dual benefit of co-training for both generation and selection.