Retrieval-augmented large language models frequently face contexts interleaving useful evidence with misleading statements or instruction-like content. Blanket refusal discards valid evidence, while uncritical adoption yields incorrect or unsafe answers.
Therefore, the ability to selectively adopt relevant information while rejecting deceptive or harmful content is critical for reliable deployment in real-world retrieval settings. We introduce SelectBench, a controlled benchmark and training set for selective evidence adoption, and post-train Qwen3.5-4B directly with DAPO using either deterministic rule rewards or a frozen semantic judge.
On the corrected 325-example SelectBench-v2 test set, strict success rises from 22.46% for the original checkpoint to 25.54% with DAPO-Rule and 26.46% with DAPO-DeepSeek. Both trained policies reduce forbidden-content adoption and produce shorter, more focused responses, yet prompt-injection following does not improve.
The paired gains are modest and fail to survive Holm correction, suggesting that stronger reward shaping or additional training iterations may be needed for more robust gains. DAPO-DeepSeek exhibits no material degradation on MMLU or clean HotpotQA, indicating that the post-training procedure preserves general capabilities.
These results demonstrate a directional improvement in selective evidence use, while identifying injection resistance and statistical robustness as important remaining challenges for future work.
Blogger's Review: This paper highlights the potential of using reinforcement learning to enhance large language models' selective evidence adoption capabilities in complex retrieval environments. Although progress is modest, it points to future research directions, particularly in reward design and training strategies. Looking forward to more significant advancements in subsequent work.