Comment by nqnielsen

And how to use LLM interpretability research for applied evaluation