works on
autoresearch 1budgeted training 1confidence cliff 1llm judging 1selective accuracy 1self-improving code 1

From the 1 of 13 linked papers with an AI index.

activity
20242026
collaborators
Showing cs.AIShow all

1 paper · 1 filter

cs.AI2026

Rehearse: Stepping Back from the Confidence Cliff in Self-Improving Autoresearch

Jiazhen Ji, Shouhong Ding

The paper studies how self‑improving autoresearch systems become less reliable at judging proposed code changes over time (the "confidence cliff") and introduces Rehearse, a lightw…