Toward Latent Language Model Skills Steering and Optimization: An Empirical Study
This empirical study investigates whether procedural LLM skills can be represented as directions in activation space and whether vector-space operations over these directions can express skill-level behaviors and finds that procedural skills admit a vector-space representation.
Xun-Yi Jiang, Jun-Da Wu, Yuxin Xiong et al.
· 0 citations