Outcome-Confounded Local Supervision in On-Policy Distillation
On-policy distillation (OPD) trains a student on its own trajectories while a teacher supplies dense token-level likelihoods at student-visited prefixes. These likelihoods are often read locally: agreement appears safe to imitate, whereas disagreement appears to identify an error. We show that both readings are confounded by the outcome of the completed trajectory. We introduce an outcome-resolved diagnostic that crosses pointwise teacher-student divergence with final-answer correctness, separat
Record details
Published: 26 July 2026
Source: arXiv cs.LG
Category: Research
Topics: Regulation · Healthcare · Children & education
Retrieved: 29 July 2026
Related evidence
These records share source-supplied organisations, an exact publisher byline, automatic topics or regions. The reason is shown on every link; related does not mean supporting, agreeing with or verifying this record.
Balancing public health and individual autonomy: a study of Chinas vaccination policy
Journal of Medical Ethics (BMJ) · 22 July 2026
Building and Governing AI Systems: Advancing Social Workers' Roles across the Technology Industry, Human Service Organizations, and Policy Institutions
arXiv · 4 August 2026
Building and Governing AI Systems: Advancing Social Workers' Roles across the Technology Industry, Human Service Organizations, and Policy Institutions
arXiv cs.CY · 6 August 2026
Limited awareness of AI regulations among developers raises potential concerns for healthcare rollout
NTU Singapore AI · 15 July 2026
Decolonizing Linguistic Policies in Automated Speech Recognition: A Framework for Cross-Culturally Competent Speech AI
arXiv · 6 August 2026
Decolonizing Linguistic Policies in Automated Speech Recognition: A Framework for Cross-Culturally Competent Speech AI
arXiv cs.CY · 7 August 2026
How to cite this record
ethics.ai (26 July 2026), “Outcome-Confounded Local Supervision in On-Policy Distillation,” evidence record 14493, https://ethics.ai/record/14493 (originally published by arXiv cs.LG).
Use and limitations
This page is a stable index and citation surface for a source record. ethics.ai did not author the underlying report and has not independently verified every claim. Automatic topics may be imperfect. For consequential use, quote and cite the original publisher.