ExplorationBench: Feedback and Self-Designed Probes Drive Agent Discovery
Source
arxiv.org
Author
Tencent Hy
Date
Why it matters
Shows that agents gain far more from environment feedback and self-chosen experiments than from extra reasoning turns (AlienCode: 0.5-11% without feedback vs up to 89% with it). Also shows run-to-run variance of tens of points, so single-run AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition →evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition → mislead.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
