What if Dario Amodei Is Right About A.I.?
About this episode
Klein asks Anthropic CEO Dario Amodei: is he right that AI capability is on an exponential curve, and what if he is? Amodei, who led GPT-2 and GPT-3 at OpenAI before co-founding Anthropic, argues growth follows predictable scaling laws — more compute and data, more power — and we're entering the steep part, with sci-fi-level systems in two to five years. He predicts agentic AI that acts in the world (coding, booking, operating computers) within 3 to 24 months. He rejects the AGI frame: no single threshold, just a smooth curve with societally meaningful moments — AlphaFold hints at a hundred future breakthroughs. On safety, Anthropic's Responsible Scaling Plan tiers risks: ASL-3 (bioweapon and cyber misuse, possibly this year or next) and ASL-4 (state-level military advantage, near-autonomy, 2025–2028). Amodei admits racing dynamics, investors, and geopolitics make pausing hard, and says he's uncomfortable with private actors wielding such power.