Kevin Roose: about 50 people decide where AI goes

In an interview with Sam Harris, published on 20 October, journalist Kevin Roose, author of The AGI Chronicles, argues that those who said language models were "stochastic parrots" or would plateau have been wrong for years. In his account, capabilities have improved steadily as models got bigger and used more computing power, as the authors of the scaling laws, Dario Amodei among them, had predicted.

Roose acknowledges that today's models are no longer just language models, because they also use reinforcement learning, and that some skeptics have given ground.

He also estimates that about 50 people, perhaps fewer, can decide whether to slow down or carry on. That is his estimate, not a measured figure. No skeptic appears in the video.