Dario Amodei — "We are near the end of the exponential"

Anthropic's Dario Amodei argues that training AI on many kinds of tasks can teach abilities that transfer to unfamiliar situations. In the interview, he describes reinforcement learning, which uses feedback from attempted tasks, as another opportunity to scale this process. The aim is to provide sufficiently varied practice that the model can handle tasks it was never explicitly taught. He says Anthropic sees improvements as this training expands. That is the basis of his optimism: broader practice could produce more general competence. It remains his account of the research trajectory, rather than a demonstration that every missing capability follows from more training.

“What has been the most surprising thing is the lack of public recognition of how close we are to the end of the exponential.”

Dario Amodei

“We would know it if you had the “country of geniuses in a data center”. Everyone in this room would know it. Everyone in Washington would know it.”

Dario Amodei

“There is zero time for bullshit. There is zero time for feeling like we’re productive when we’re not. These tools make us a lot more productive.”

Dario Amodei

“Each model makes money, but the company loses money.”

Dario Amodei

“If I’m just off by a year in that rate of growth, or if the growth rate is 5x a year instead of 10x a year, then you go bankrupt.”

Dario Amodei