Which Self-Improvements Should We Trust? Reliable Self-Improvement When Agents Reuse Their Benchmarks
Detailed statistical theory for RSI evaluation in this setting is developed, including simultaneous error control, valid lower bounds on cumulative improvement, and a characterization of the fundamental limits of adaptive evaluation reuse.
Xiao-Jing Sun, Yu-Hang Zeng, Zi-Hua She et al.
· 1 citation
· ⚡1