Paper Reading with Anthropic Research

Zac Hatfield-Dodds from Anthropic walks through "Toy Models of Superposition," the paper that helped define what alignment actually means beyond the LinkedIn posts, in a book club format with Relevance AI. You'll spend the first half hour getting the core ideas, then an hour and a half digging into questions with someone who was literally in the room when this research was being built. Come prepared to have read the paper, especially if you're building or thinking about safety in AI systems.