Thanks, though I don’t find this engages with my feelings much. I try to point at a central example
inadvertently training one capabilities researcher in exchange for training one safety researcher is likely still worth it (I think their ratio is much better than this).
Let us imagine that the current speed of AI progress requires 9 capabilities researchers for every safety researcher. Many people want to be capabilities researchers, for the large salaries. If one trains safety and capabilities researchers at a rate of 1-1, does that speed up or slow down AI progress?
In terms of the other fellowships those seem more straightforwardly good, though I don’t know what % of the $50m will go on that.
Thanks, though I don’t find this engages with my feelings much. I try to point at a central example
Let us imagine that the current speed of AI progress requires 9 capabilities researchers for every safety researcher. Many people want to be capabilities researchers, for the large salaries. If one trains safety and capabilities researchers at a rate of 1-1, does that speed up or slow down AI progress?
In terms of the other fellowships those seem more straightforwardly good, though I don’t know what % of the $50m will go on that.