AI Hub AI情报 26 天前

Manifold Steering from a Natural State is Effective and Generalizable, but Relies on Model Accuracy

tl;dr I continue work to apply Goodfire lab's manifold steering methods to the Gemma-2-2b model's natural state, both in raw residual space and in the space produced by Anthropic's pre-trained CLT features.

发布于 2026-10-13 02:06 阅读原文