arxiv
PublishedNovember 11, 2026 at 5:00 AM
Train on prefill, monitor at decode: Using sparse autoencoders for topic monitoring
Publisher summary· verbatim
arXiv:2611.12935v1 Announce Type: new Abstract: Topic monitoring is the task of labelling the topic of each token in a model's response during generation. Activation-based monitors that run at generation time have so far been trained to detect harmful content and are limited to two or three classes.
Stay posted· Newsletter
Three short emails a week: Monday, Wednesday and Friday. Top movers, new models and the most-referenced story.
Your email is used only to send The Bubble Brief. Privacy policy
Discussion
No replies yet. Be first.
Related coverage
More from ARXIV
arxivMach2: Scientific Discovery Platform with Deterministic Execution Constraints1darxivOverServe: Resource Amplification Attacks on SaaS Agents via Tool--Prompt Co-Design1darxivCFGComposer: Compositional Audio-Text-to-Motion Generation via Copula Score Correction1darxivMulti-View 3D Reconstruction Using V-JEPA Features1dThe Bubble Brief
MON · WED · FRIRead AI insights every Monday, Wednesday and Friday — top movers, new models, story of the week.
Your email is used only to send The Bubble Brief. Privacy policy
Originally published on arxiv ↗