Study Finds Preference-Based Model Rankings Vulnerable to Manipulation
A new arXiv paper reports that arena-style leaderboards ranking language models by human preference can be manipulated, raising concerns about the reliability of such evaluations.
Read the full story at arXiv cs.AI ↗
Timeline · 1 report
- 2026-11-13 05:00 · arXiv cs.AI
Leaderboard Hacking: Preference-Based Model Evaluations are Vulnerable to Manipulation