Data for "The observability gradient predicts where AI benchmarks measure truth versus consensus" (NeurIPS 2026 submission)
收藏官方服务：
资源简介：
Anonymized dataset accompanying a position paper submitted to NeurIPS 2026 (Position Paper Track). Contains MMLU per-subject accuracy data, MMLU-Pro cross-model results, FActScore entity-level data, Prolific MMLU rater study responses, and computed observability scores. Access is restricted during the double-blind review period.
提供机构：
Zenodo创建时间：
2026-05-04
