Fernando Diaz

Associate Professor, CMU. Researcher, Google. Evaluation and design of information retrieval and recommendation systems, including their societal impacts.

Articles & links

Please try to look deeper than success rate for agent evaluation. Thank you. arxiv.org/abs/2606.17541

Offline Preference-Based Trajectory Evaluation arxiv.org
AI Weekly's analysis
  • Fernando Diaz argues success-only metrics tie agent comparisons on roughly 75% of instances, gutting statistical power.
  • His preference-based trajectory evaluation compares progress and time-to-return profiles, cutting ties to roughly 35%.
  • The paper suggests apparent benchmark saturation may reflect the evaluation measure, not exhausted data or problems.
Read full analysis →
View on Bluesky · ♥ 2 ↻ 1 ↩ 1 · 2 from the directory shared this · 6d ago

In Fernando Diaz's orbit

Center = Fernando Diaz. Left = members they follow (green edges). Right = members who follow them (blue edges). Top = mutual follows (orange edges, slightly larger). Drag any node to reposition; click to open that profile.