Stable Answers, Unfinished Reasoning: Why Self-Consensus Is Not a Safe Early-Exit Signal

2026-09-09 · Yunxiang Mo, Donghao Zhao, Hejia Geng arxiv

A natural way to cut reasoning-model inference cost is to repeatedly probe a single partial trajectory for its current answer and stop once probes agree -- self-consensus. We ask whether any such rule is both safe and token-saving, and whether one can be selected once and reused. A preregistered sweep of 3,520 consensus rules, replayed on frozen trajectories from two models and three benchmarks, clears none of three acceptance gates fixed in advance; the frontier reproduces on a held-out split and on two unseen models -- while a boundary-confidence control (DEER) swept through the same pipeline clears all three. The reason lies in the signal: agreement establishes that the current answer persists under a fixed probing procedure, not that the reasoning has terminated -- a consensus-termination gap. Stopping on it commits non-terminal answers. At a rule still saving 32% of the tokens, one stop in nine fires on an answer the trajectory itself later abandons, and most of those stops cut off a correction it would otherwise have made. Widening the agreement window does not remove them: the share levels off near 7%, and by then the saving has fallen to 8%. Probe re-wording and a hand-labelled error taxonomy show the agreed answer is often a placeholder the model had not settled on. Used on its own as the stop signal, agreement fails not because it is insufficiently strict, but because it repeatedly measures the wrong object.

📄 PDF Abstract BibTeX arXiv:2609.09989

Code (0)

등록된 구현이 없습니다.

Shorter Reasoning, Earlier Answers? An Evaluation of Reasoning Interfaces

2026-08-04 · Francesca Carlon, Vincent Ginis, Andres Algaba arxiv

Large language models often reason at length before answering, increasing cost and latency. Prompts and trained settings can shorten this reasoning, but a shorter trace may only show that the model stopped sooner. Here, …

GATES: Self-Distillation under Privileged Context with Consensus Gating

2026-02-24 · Alex Stein, Furong Huang, Tom Goldstein arxiv

We study self-distillation in settings where supervision is unreliable: there are no ground truth labels, verifiable rewards, or external graders to evaluate answers. We focus on document-grounded question answering with…

Question Answering

Dual Consensus: Escaping from Spurious Majority in Unsupervised RLVR via Two-Stage Vote Mechanism

2026-03-17 · Kaixuan Du, Meng Cao, Hang Zhang, Yukun Wang 외 arxiv

Current label-free RLVR approaches for large language models (LLMs), such as TTRL and Self-reward, have demonstrated effectiveness in improving the performance of LLMs on complex reasoning tasks. However, these methods r…

Reinforcement Learning

DiSCTT: Consensus-Guided Self-Curriculum for Efficient Test-Time Adaptation in Reasoning

2026-03-05 · Mohammad Mahdi Moradi, Sudhir Mudur arxiv

Test-time adaptation offers a promising avenue for improving reasoning performance in large language models without additional supervision, but existing approaches often apply a uniform optimization objective across all …

Reinforcement Learning Test-time Adaptation

LLMs as a Jury: Cross-Model Consensus Can Outperform Process Reward Models for LLM Reasoning

2026-07-11 · Ning Liu arxiv

Selecting the correct answer from a pool of candidate reasoning chains is the engine of test-time scaling, yet the standard selectors each carry a cost: self-consistency inherits the errors of the single model it resampl…