RE: How Much Do You Know About Mechanistic Interpretability (AI)?

I watched an interview of the head researcher of the division at Anthropic working on Mechanistic Interpretability. He went into fewer details than the ones in this video you shared, and said Mechanistic Interpretability is new and they are just scratching the surface of what's possible. Basically, from my understanding, Mechanistic Interpretability is kind of similar to analyzing brain activity to see what group of neurons react to what, but for AI, and with a higher probability of better understanding what's going on than in the human brain, for now.