Benjamin Mann
Co-founder of Anthropic; tech lead for product engineering. One of the architects of GPT-3 at OpenAI; one of the first authors on the GPT-3 paper. Founded the Labs/Frontiers team at Anthropic from which Claude Code and MCP emerged.
Work covered in this wiki
- Benjamin Mann on Anthropic and AGI — Lenny’s Podcast, 2025. Why Anthropic was founded; Constitutional AI; ASL framework; AGI timeline; RLAIF; alignment theory.
Key ideas
- Constitutional AI — natural-language principles + model self-critique loop; alignment technique that makes safety and capability convex.
- AI Safety Levels — ASL-3/4/5 risk framework embedded in Anthropic’s responsible scaling policy.
- Economic Turing Test — his preferred AGI measure: 50% of money-weighted jobs passable by an agent.
- Three worlds theory — pessimistic / optimistic / pivotal middle; human alignment effort is critical.
- “Build for the model 6 months from now” — the design principle behind Claude Code’s success; same principle articulated independently by Amjad Masad.