What Makes a Good Unmasking Policy? A Controlled Study of Learned Unmasking in Diffusion Language Models

No. 236 of 245 · arXiv · Jinya Sakurai, Xun Xu, Jaehong Yoon · published 23 Oct 2026, 04:00 UTC · 45 min read

AI & machine learning
Paper

2 judges. Scored by two of the three judges. The third reads only the day’s strongest candidates. How far stories are checked

This study revisits learned unmasking policies for masked diffusion language models, which commit multiple tokens in parallel. Treating supervised policies as distillation of a decoding procedure, the authors separately vary what is labeled committable, the training states, and the policy inputs.

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

0.55total under rubric v2.0

Significance0.47
Reader interest0.47
Audience breadth0.35
Practical relevance0.56
Substantiated0.89
Technical depth0.91
New information0.85
Counts against: Promotional0.04
Counts against: Overclaim0.25

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	0.54	0.40	—	0.47	0.14	+0.25	+0.118
Reader interest	0.54	0.40	—	0.47	0.14	+0.20	+0.094
Audience breadth	0.46	0.25	—	0.35	0.21	+0.10	+0.035
Practical relevance	0.63	0.50	—	0.56	0.13	+0.15	+0.085
Substantiated	0.97	0.80	—	0.89	0.17	+0.10	+0.089
Technical depth	0.97	0.85	—	0.91	0.12	+0.10	+0.091
New information	0.91	0.80	—	0.85	0.11	+0.10	+0.086
Promotional	0.04	0.05	—	0.04	0.01	−0.15	−0.007
Overclaim	0.25	0.25	—	0.25	0.00	−0.15	−0.037
Sum of contributions							+0.553
Claim-check penalty							0.000
Total, as published							0.552

Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested. Judge 3 did not score this story; its mean uses the other two.

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.