Learning to Stop without Learning to Stop: Self-Supervised Confidence Training Improves Reasoning Efficiency

Paper recorded by Signals 4 on 2026-09-25 in cs.AI. Abstract reproduced from arXiv; link to the original below.

Published 2026-09-25 on arXiv · recorded by Signals 4 on 2026-09-28

Category: cs.AI · 人工智能 · first seen 2026-09-28

Abstract

Reasoning models often generate very long reasoning traces, making inference computationally expensive. Existing approaches typically improve efficiency either through inference-time early-stopping mechanisms or by explicitly encouraging shorter reasoning during training, for example through reinforcement learning with length penalties. We show that substantial efficiency gains can instead emerge

Read on arXiv →