OVERCLOCK.news
developer

Agentic On-Policy Distillation Implicitly Reshapes LLMs' ReAct Patterns

On-policy distillation (OPD) transfers specialized capabilities to language-model agents, but its cross-domain effects on interaction remain unclear.

BlueskyXRedditMail
Why am I seeing this Ranked on source trust — arXiv

It ranks mainly on source trust: arXiv is the most reliable outlet we track on this subject, and is the only one on the story so far.

Link-outLink-out, because it did not clear the bar for a write-up. Link-out means we point at the publisher and say nothing of our own.

Blended score 0.546 — every figure below is computed, none of it is editorial.
FactorWeightScore ContributionWhere it came from
Corroboration 0.35 0.39 +0.135 25% 1 independent org on the story. Tier-3 aggregators never corroborate — they can show something is circulating, never that it is true.
Source trustleads 0.25 0.85 +0.212 39% arXiv is the highest-trust source on this story and is first-party — the organisation announcing its own news. Trust is taken from the best source, not averaged.
Pickup rate 0.20 0.00 +0.000 0% One counted organisation, so there is no spread to measure — nothing has picked this up to set a rate.
Freshness 0.20 0.99 +0.198 36% Halves every 10 hours from the newest item on the story. This is the only factor that rewards a story for nothing more than being recent.

Corroboration counts distinct organisations, once each, and only from tiers 1 and 2. Freshness halves every 10 hours, so this ranking is a snapshot and will differ at the next build.

Read the full article at Arxiv →

What happened

On-policy distillation (OPD) transfers specialized capabilities to language-model agents, but its cross-domain effects on interaction remain unclear. We investigate how OPD reshapes ReAct interaction patterns across mathematics, ALFWorld, and $\tau^2$-bench, focusing on reasoning, actions, and observation acquisition. General-reasoning supervision reduces mathematical tool attempts from 70.4% to 14.4% while increasing accuracy from 16.7% to 38.0%; code-interpreter supervision improves household actions and retail queries. In contrast, ALFWorld supervision reduces $\tau^2$-bench accuracy from 41.5% to 6.6%, accompanied by frequent tool calls that violate user-information dependencies.

1independent orgs
25story score
0velocity
85source trust
1passes seen

How this story arrived

Ordered by when each source was first observed, which is what the velocity figure is computed from. Publishers backdate; observed order does not.

  1. 01 Arxivfirst-party first seen Agentic On-Policy Distillation Implicitly Reshapes LLMs' ReAct Patterns

Overclock clusters coverage from independent sources and grades it automatically. The figures above are computed, not editorial. This page summarises and links to reporting by the outlets named — follow the links for the original work.