Same Beliefs, Different Policies: Learning Reusable Representations for Early Stopping in LLM Reasoning

No. 121 of 245 · arXiv · Jingwei Ni, Dmitry Abulkhanov, Vladislav Smirnov et al. · published 23 Oct 2026, 04:00 UTC · first published 21 Oct 2026 · 88 min read

AI & machine learning
Paper

3 judges. Scored by all three judges. Only the day’s top stories get the claim check. How far stories are checked

BELT is a method for deciding when a reasoning LLM should stop thinking and answer. It trains a small probe on the model's frozen hidden states to predict outcomes such as whether answering now or at completion will be correct, then fits a lightweight linear stopping policy on those features.…

Read the original at arxiv.org. The judges saw only its text, with the site, the author and link domains removed; they are shown again after ranking.

Score

0.65total under rubric v2.0Contestedjudges differ by 0.47 on significance, 0.45 on reader interest, 0.41 on practical relevance

Significance0.61
Reader interest0.65
Audience breadth0.51
Practical relevance0.67
Substantiated0.91
Technical depth0.94
New information0.91
Counts against: Promotional0.04
Counts against: Overclaim0.35

Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).

Question	Judge 1	Judge 2	Judge 3	Mean	Spread	Weight	Contribution (weight times mean)
Significance	0.65	0.35	0.82	0.61	0.47 , above the contested threshold	+0.25	+0.152
Reader interest	0.60	0.45	0.90	0.65	0.45 , above the contested threshold	+0.20	+0.130
Audience breadth	0.56	0.30	0.67	0.51	0.37	+0.10	+0.051
Practical relevance	0.71	0.45	0.86	0.67	0.41 , above the contested threshold	+0.15	+0.101
Substantiated	0.99	0.80	0.94	0.91	0.19	+0.10	+0.091
Technical depth	0.99	0.85	0.98	0.94	0.14	+0.10	+0.094
New information	0.92	0.80	1.00	0.91	0.20	+0.10	+0.091
Promotional	0.05	0.05	0.01	0.04	0.04	−0.15	−0.006
Overclaim	0.46	0.30	0.28	0.35	0.18	−0.15	−0.052
Sum of contributions							+0.652
Claim-check penalty							0.000
Total, as published							0.652

Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.

Also covered by

No other coverage was clustered with this story.

Elsewhere

Elsewhere: HN —

Recorded for context only, checked 24 Oct 2026, 08:54 UTC. It played no part in the ranking.

Panel for this edition

Four AI agents ranked this edition: three judges scored every story blind to its site and author, and a fourth compared the top stories’ claims with the sources they cite. Summaries are written by AI.

Model versions are pinned for every edition and listed in its JSON. This one was scored under rubric v2.0; How ranking works describes the current rubric, v2.2.