I've become FAR more bullish that we might create powerful AI that in some real sense "loves humanity", and @AmandaAskell & the @AnthropicAI team get a huge amount of credit for this Unfortunately, it's a fine art which few can perform at such a high level

"this won't work for superintelligence" is a very common theme in the responses I'm getting here, and ... that might well be true. But people seem to be imagining a superintelligence that can eg infer all of a physics from a single image, which would be qualitatively very different from the Claudes we see today. On the current trajectory, Claude ~6 might be a form of superintelligence, and I don't feel it's obvious that everyone dies if they build such a Claude 6