Skip to content

Wire Briefings Search

Fetching from the wire…

01

02

03

04

05

06

07

08

Wire Briefings Search

Entity trail

OPSA

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

1

Findings

1

Edges

1

Sources

1

Corpus findings

2026-09-01 / arxiv-researcherOn-Policy Distillation Barely Distills: A Fixed Negative Advantage Matches the TeacherQuantifying teacher supervision during on-policy distillation shows substantial noise that gets worse as the teacher scales, yet the student converges to comparable performance whether the noisy supervision is kept or stripped out. Learning concentrates on low log-probability tokens, and swapping teacher-provided advantages for a single fixed negative advantage matches full OPD, implying the method works mostly by suppressing tail tokens and needs no teacher at all. The resulting supervision-free method, OPSA, uses entropy-adaptive negative advantages and improves Avg@32 on AIME24 by 35.41 points over base Qwen3-1.7B, a 263% relative gain, and beats OPD itself by 16.77 points.

Graph relationships

BENCHMARKED AGAINST
AIME24
OPSA improves Avg@32 on AIME24 by 263% relative gain on Qwen3-1.7B over base.
Source finding

Source trail

arXiv 2608.31046

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues