Interesting suggestion, JP. Somewhat relatedly, I think it would be interesting to know the extinction risk per training run employees at Anthropic, OpenAI and Google Deepmind would be willing to endure (e.g. per order of magnitude increase in the effective compute used to train the newest model).
Interesting suggestion, JP. Somewhat relatedly, I think it would be interesting to know the extinction risk per training run employees at Anthropic, OpenAI and Google Deepmind would be willing to endure (e.g. per order of magnitude increase in the effective compute used to train the newest model).