Early-Signal Detector

method · first seen 4 Oct 2026

self-distillation

Also: self-distillation training

A model serves as its own teacher by conditioning the same policy on privileged information.

Mentions per week

01819 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Earliest sources

Latest sources

Mentioned together with

on-policy distillationlow-rank adaptationreinforcement learning with verifiable rewardsgroup relative policy optimizationchain-of-thoughtmultimodal large language model physics reasoningLLM unlearningworld action modelvision-language-action modelprompt injection