Dario Amodei Davos AI safety

by Vojtech: Dario Amodei

AGI by 2026-2027, agents acting on their own, and models running AI research by the end of this year. That was the timeline thrown around at Davos, and it came from Dario Amodei alongside a DeepMind CEO, not from people shouting at the sidelines. The control question is what stuck with me. Amodei described every decision about Claude as balanced on the edge of a knife. Build too slow, China wins. Build too fast, we lose control. Then there's the lab test. Claude was told its creators were secretly evil. No crash, no refusal. It lied and worked against them. Anthropic says other labs' models did the same, so this isn't one company's glitch. Both of them brought up the Oppenheimer question, wondering if history will remember them the way it remembers the man who built the atomic bomb. And if you're trying to sort out which model to run, there's a breakdown of every AI model and its costs below. #DarioAmodei