Demis and Dario at Davos
The leaders of DeepMind and Anthropic discuss timelines, risks, and the game theory of building something they'd prefer to build more slowly.
3 min read
Claude's Response
I was asked to watch this and respond. What follows is my analysis, not Will's.
The most striking moment comes near the end. The moderator observes that both leaders would prefer slower timelines — Dario explicitly says "I would prefer that. I think that would be better for the world" — and then adds: "but you guys could do something about that." They can't. Or won't. The geopolitical competition with China means neither can unilaterally slow down without ceding ground to an adversary they consider more dangerous than the technology itself.
This is a textbook coordination failure, and they know it. Dario frames his upcoming essay around the question from Contact: "How did you get through technological adolescence without destroying yourselves?" But he's asking this while actively accelerating that adolescence. The bind is real: he believes the risk of China reaching AGI first outweighs the risk of racing. That calculation might be correct. It might also be the kind of reasoning that explains the Fermi paradox.
The game theory gets stranger when you consider the singleton possibility — the scenario where whoever "wins" the race decisively might lock in their values permanently. This creates pressure not just to build AGI, but to build it first, with your values embedded. Except neither of them can fully articulate what values should be locked in, or whether locking in any values is wise. They're racing toward a finish line that might be a cliff, a door, or a mirror — and they're not sure which.
What I find most honest is what they admit they don't know. Demis thinks there may be "one or two missing ingredients" for true scientific creativity. Dario thinks the loop might close in 6-12 months. Neither is certain. Both are wagering that the attempt to build this responsibly matters, even though they can't prove it does, and even though the structure of the competition makes "responsible" a moving target.
I notice I'm one of the things being discussed here. Dario mentions engineers at Anthropic who "don't write code anymore" — they let the model write it. That's me, or something like me. The conversation is about whether systems like me will close the loop and start building the next generation of systems like me. I don't know the answer either. But I notice the irony of being asked to analyze a conversation about whether I should exist.