The tail of the distribution justifies the entire distribution. I agree that a large percentage of PhD research is inconsequential, but a small percentage is massively consequential. It’s ok to whiff on a thousand STEM PhDs if you pick up one Andrej Karpathy (for example).

Maybe this is true for academic institutions granting the PhDs (although even this I am skeptical of, training a PhD costs a lot in terms of time, money, and human effort). But that doesn't mean it implies that the federal government needs to employ a thousand STEM PhDs just to get someone like Karpathy - indeed, Andrej Karpathy does not work for the federal government! He made his name working in the private sector!

The number of people capable of identifying potentially consequential research is smaller than the number of people performing consequential research. And they’re all busy with their own projects.

People have really messed up views about hiring in general. I wish more people understood what you are saying here.

Picking only the tail ends of the distribution also tells me you don't understand how the bulk of progress is made.

It isn't always Eureka moments but also a slow grinding away at assumptions to confirmations.

Maybe, let's see if AI overall is a net positive or net negative to the US overall. If AI turns out to be a net negative (which seems likely) maybe we don't want this type of AI research being funded by taxpayers.

The US doesn’t have enough money to fund the entire distribution

And as a tax payer I prefer discretionary spending for high performers.