On-Policy Self-Training: Data-Free Self-Improvement in Large Language Models
Comments on “On-Policy Self-Training: Data-Free Self-Improvement in Large Language Models”
3 judges. Scored by all three judges. Only the day’s top stories get the claim check. How far stories are checked
Hrih News: A new paper, On-Policy Self-Training (OPST), proposes improving language models without human-curated questions or answers. Seeded only by coarse topic descriptors, the model generates its own question-answer pairs, filters them through pseudo-verification and a consensus gate, and trains via on-policy self-distillation from a privileged EMA teacher.
Why it ranked
No. 49 of 155 with a score of 0.63. The panel rated it highest on new information (0.95) and technical depth (0.89), lowest on practical relevance (0.58).
- AI judgeJudge 2claude-haiku-5-5
- Google research paper on self-training LLMs from self-generated data; substantive method and results, primarily of interest to ML practitioners, strongly AI-focused, not in the recent story list.
- AI judgeJudge 3gpt-5.6-luna
- Primary-style research presents a technically detailed, reproducible self-training method with substantial benchmark gains and ablations. It is highly relevant to AI builders, though experiments are limited to small models and some claims rely on omitted tables.
Each judge wrote its line after answering the rubric, with the site and the author still hidden. Every answer behind this is in the table under Score.
Score
0.63total under rubric v2.2
- Significance0.63
- Reader interest0.68
- Audience breadth0.60
- Practical relevance0.58
- Substantiated0.84
- Technical depth0.89
- New information0.95
- Counts against: Promotional0.04
- Counts against: Overclaim0.30
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 | 0.67 | 0.45 | 0.78 | 0.63 | 0.33 | +0.24 | +0.152 |
| Reader interest | 0.55 | 0.55 | 0.93 | 0.68 | 0.38 | +0.26 | +0.176 |
| Audience breadth | 0.57 | 0.50 | 0.74 | 0.60 | 0.24 | +0.13 | +0.078 |
| Practical relevance | 0.51 | 0.50 | 0.72 | 0.58 | 0.22 | +0.22 | +0.127 |
| Substantiated | 0.94 | 0.70 | 0.88 | 0.84 | 0.24 | +0.05 | +0.042 |
| Technical depth | 0.96 | 0.75 | 0.95 | 0.89 | 0.21 | +0.05 | +0.044 |
| New information | 0.93 | 0.95 | 0.97 | 0.95 | 0.04 | +0.05 | +0.048 |
| Promotional | 0.06 | 0.05 | 0.02 | 0.04 | 0.04 | −0.19 | −0.008 |
| Overclaim | 0.36 | 0.35 | 0.18 | 0.30 | 0.18 | −0.11 | −0.033 |
| Sum of contributions | +0.626 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.626 | ||||||
Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.
The badge for this ranking, for the author's page or anyone's: one line of HTML, and it links back here.
<a href="https://rubric.hrih.si/editions/2026-10-28/49/"><img src="https://rubric.hrih.si/badge/2026-10-28/49.svg" alt="Ranked #49 on Rubric, 28 Oct 2026" width="245" height="28"></a>Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN 1 pts, not on front page
Recorded for context only, checked 28 Oct 2026, 06:47 UTC. It played no part in the ranking.