Skip to content
Research

CLT Features Support Manifold Steering, but Hold Only Partial Steering Signal.

LessWrongAnalysis or commentary·
AI brief

An investigation extended cyclical manifold methods from the Goodfire lab to Anthropic's pre-trained CLT feature space on Gemma-2-2B, finding that manifold steering produces the expected cyclical transition between days of the week in activation space.

Written by AI from LessWrong's published text.

Background

From our own funding, model and news trackers, not from the source.

Source

“tl;dr I continue work to extend the Goodfire lab's cyclical manifold methods to Anthropic's pre-trained CLT feature space on Gemma-2-2B.”
LessWrong, 21 Sept 2026, 21:49.

Interpretation or community commentary rather than straight reporting.

Read original story ↗