Arvind Narayanan
Senior author of science-of-agent-reliability (ICML 2026), at Princeton, with sayash-kapoor. Discovered while ingesting that paper on 2026-08-23.
The paper's stance is the thing worth recording under this node: reliability is treated as "an empirically measurable property of agent behavior" and deliberately not as an algorithms problem — the authors propose no method for improving it, only twelve metrics for seeing it. Its central finding cuts against the field's default reading of its own progress: eighteen months of accuracy gains bought only modest reliability gains, and almost none on the open-ended benchmark.
Nothing further about this person is evidenced by the sources held here.
Connections
- co-author of science-of-agent-reliability, with sayash-kapoor
- affiliated with princeton-university