Code for "The observability gradient predicts where AI benchmarks measure truth versus consensus" (NeurIPS 2026 submission)
收藏DataCite Commons2026-05-04 更新2026-05-07 收录
下载链接：
https://zenodo.org/doi/10.5281/zenodo.20029508
下载链接
链接失效反馈官方服务：
资源简介：
Anonymized code repository accompanying a position paper submitted to NeurIPS 2026 (Position Paper Track). This deposit contains the analysis pipeline, statistical scripts, and figure-generation code used to produce the empirical results in the manuscript, including MMLU gradient analysis, MMLU-Pro cross-model scaling, FActScore correlations, and the Prolific MMLU rater study analysis.
Access is restricted during the double-blind review period and will be opened upon acceptance or after the review period concludes.
提供机构：
Zenodo
创建时间：
2026-05-04
