Measuring and Guiding Monosemanticity

Ruben Härle, Felix Friedrich, Manuel Brack +4

There is growing interest in leveraging mechanistic interpretability and controllability to better understand and influence the internal dynamics of large language models (LLMs). H…