An AI pretraining researcher resigned from Anthropic today, and his exit thread is interesting.

His claim: the people building frontier AI genuinely believe it could kill us all by 2030 and they're racing to build it anyway. At Anthropic, everyone understands the stakes. Their answer to "why?" is: someone will do it, so it has to be us.

When the risk is everyone's, the choice can't belong to a handful of companies who each believe they're the only responsible ones, which is exactly what every one of them believes.

The answer isn't better behaved centralized systems. I believe the answer is open, verifiable systems.

https://x.com/hilbertspaess/status/2097476196791709843