作者位次 #3/6ACL 2026Findings· Findings of ACL
Rethinking RL Evaluation: Can Benchmarks Truly Reveal Failures of RL Methods?
Zihan Chen · Yiming Zhang · Hengguang Zhou · Zenghui Ding · Yining Sun · Cho-Jui Hsieh
摘要
Current benchmarks are inadequate for evaluating progress in reinforcement learning (RL) for large language models (LLMs). Despite recent benchmark gains reported for RL, we find that training on these benchmarks’ training sets achieves nearly the same performance as training directly on the test...