Skip to content

Wire Briefings Search

Fetching from the wire…

01

02

03

04

05

06

07

08

Wire Briefings Search

Entity trail

RetireOPD

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

Briefing refs

1

Findings

1

Edges

3

Sources

1

Corpus findings

2026-09-20 / sources-researcherRetireOPD lets the student fire its teacher when the gap stops closing, beating the teacher every timeRetireOPD (arXiv 2609.20784, Sept 17) attacks two failures in self on-policy distillation for multi-turn agents: privileged information alone does not make a teacher reliable, and teacher supervision helps only at certain training stages. Instead of a fixed distillation schedule, the student drops the teacher on its own once their discrepancy stops shrinking and it reaches a target fraction of the teacher's success rate, then continues on RL alone. Across Qwen2.5 models from 1.5B to 7B it improves ALFWorld success by 14.1-18.8% and WebShop accuracy by 11.8-19.0% over the RL baseline, and surpasses its own skill-conditioned teacher in every setting.

Graph relationships

BENCHMARKED AGAINST
WebShop
RetireOPD improved WebShop accuracy by 12-19 percent
Source finding
BENCHMARKED AGAINST
ALFWorld
RetireOPD improved ALFWorld success by 14-19 percent
Source finding
BENCHMARKED AGAINST
Qwen2.5
RetireOPD tested self-taught distillation on Qwen2.5 models from 1.5B to 7B parameters
Source finding

Source trail

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues