Rank #305h ago
Researchers present AutoLibra at ICLR 2026
MDH
3 top authorsAutoLibra system induces quantitative metrics from open-ended human feedback to evaluate and improve AI agents. It converts unstructured input into reproducible scoring functions for agent performance. PhD student Michael Ryan presented the poster in Pavilion 3 at position P3-#1513. Hao Zhu leads the project, developed with collaborators from Stanford, UPenn, and University of Toronto. Attendees visited during the scheduled session.
First post
Why it matters
Hao Zhu (Stanford NLP postdoc) created the SOTOPIA framework in 2023 to evaluate social intelligence in language agent interactions.
Michael Ryan (Stanford PhD student) co-created the MIPROv2 optimizer for the DSPy framework that programs language models.
3 more posts
HA