Back to Blackwell: Closing the Loop on Intransitivity in Multi-Objective Preference Fine-Tuning.

Zhuohao Yu 0003 Carnegie Mellon University, Pittsburgh, PA, USA Zhuohao Yu Zhuohao Yu 0001 Peking University, Beijing, China Renmin University of China, School of Information, China https://zhuohaoyu.github.io/ https://scholar.google.com/citations?user=zVYE7-UAAAAJ https://github.com/zhuohaoyu https://orcid.org/0009-0000-8256-8588 Zhuohao Yu 0002 Chinese Academy of Sciences, Institute of Software, Beijing, China https://scholar.google.com/citations?user=mTbIqosAAAAJ https://orcid.org/0009-0006-0037-0396 https://www.wikidata.org/entity/Q130803334

Jiahao Zhang Lujing Zhang Keltin Grimes Zhuohao Yu 0003 Gokul Swamy 0001 Zhiwei Steven Wu Back to Blackwell: Closing the Loop on Intransitivity in Multi-Objective Preference Fine-Tuning. 2026 February abs/2602.19041 CoRR https://doi.org/10.48550/arXiv.2602.19041 db/journals/corr/corr2602.html#abs-2602-19041 streams/journals/corr

Zhuohao Yu 0003 Zhiwei Steven Wu Adam Block From Curiosity to Caution: Mitigating Reward Hacking for Best-of-N with Pessimism. 2026 April abs/2604.04648 CoRR https://doi.org/10.48550/arXiv.2604.04648 db/journals/corr/corr2604.html#abs-2604-04648 streams/journals/corr

Adam Block Keltin Grimes Gokul Swamy 0001 Zhiwei Steven Wu Jiahao Zhang Lujing Zhang