AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Debunking the Usefulness of Mechanistic Interpretability
In this chapter, the speakers discuss their disagreements with a previous post about the usefulness of mechanistic interpretability and argue that some people should still worry about usefulness or theory of change. They also explore the existence of induction heads and French neurons, debating whether our world has these phenomena or if alternative explanations exist.