Tag
WorldReward introduces a vision-language reward model for camera-conditioned world models that unifies action-consistency and visual-quality evaluation through chunk decomposition and preference aggregation, outperforming existing methods like GPT-5.5 on benchmarks.
This paper evaluates whether clinician pairwise preferences reliably indicate clinical safety in LLMs, using 26,804 judgments from 736+ clinicians across 13 models. It finds that preference rankings poorly track safety-critical failures and proposes a clinically adjusted ranking that better incorporates rubric-based safety signals.
This paper formalizes pairwise reference alignment as a model-level ordinal observable, defining a statistic to measure agreement between a model's scoring and a reference preference distribution, with finite-sample estimators and an empirical study on Qwen2.5 models and RewardBench.