Sayash Kapoor
Co-author of science-of-agent-reliability (ICML 2026) at Princeton, with arvind-narayanan and four colleagues. Discovered while ingesting that paper on 2026-08-23.
What the source evidences is a position on evaluation rather than on models: that a benchmark reporting one accuracy number from one run in a fixed environment "reveals nothing about whether an agent would succeed on the same task tomorrow", and that reliability has to be measured as a separate axis with repeated runs and perturbed conditions. The work runs on Princeton's HAL (Holistic Agent Leaderboard) harness, which the same group maintains.
Nothing else about this person is evidenced by what the wiki holds, and nothing is assumed here.
Connections
- co-author of science-of-agent-reliability, with arvind-narayanan
- affiliated with princeton-university