OPEN PROBLEMS IN MECHANISTIC INTERPRETABILITY

Summary from the TLDR AI newsletter

A great paper that talks through some of the challenges and unsolved problems facing the field when trying to understand how knowledge and skills are represented internally in a language model.