trends.barnas.net

method · first seen 2 Nov 2026

sparse autoencoder feature steering

Also: SAE feature steering

Controlling model behaviour at inference by modulating features learned by a sparse autoencoder on internal activations.

Mentions per week

0119 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Earliest sources

Latest sources

Mentioned together with

abstention in language modelsLLaVA