Tag
This paper finds that 82% of the reported out-of-domain multiple-choice gains from RLVR on four 7B SimpleRL model pairs are artifacts of answer extractors that only read \boxed{} answers, meaning RL mostly teaches models to place answers in the box rather than improving actual reasoning.
This paper reveals that much of the reported progress in LLM hallucination detection is due to benchmark construction artifacts, where ground-truth answers are embedded in prompts, allowing a simple text-similarity baseline to achieve near-perfect scores. Through a large-scale controlled evaluation, the authors show that most methods perform near chance under proper controls, except for supervised probes on upper-layer hidden states such as SAPLMA and their proposed DRIFT.