Skip to content
NeurIPS 2026 planner

Oral ยท Sydney

Leaderboard Hacking: Preference-Based Model Evaluations are Vulnerable to Manipulation

Time not published yetRoom not published

When
Time not published yet
Where
Room not published
Poster
Poster of this paper
Authors
Eve Fleisig, Joachim Baumann, Dylan Hadfield-Menell, Dirk Hovy
Organisations
Princeton University, University of Zurich, Massachusetts Institute of Technology, Bocconi University

Open in scheduleOfficial page