Industrializing a small field: Lessons from Vannevar

Crossposted from canaryinstitute.ai/blog/lessons-from-vannevar . Related LW pieces on AI Safety field-building and the push-vs-pull question. Related posts A system overview for near-term, low-trust AI compute verification — Naci Cankaya. A concrete example of the verification work built on Aumann and Lindell's covert-adversary framework that this essay cites. AI Safety has a scaling problem — Boyd Kane. Diagnoses the pull-model bottleneck directly: fellowships reject 95%+ of qualified applicants because mentor capacity doesn't scale, and proposes push-side research bounties as an alternative. The case for AI safety capacity-building work — abergal. Marshals survey evidence that funded capacity-building programs are among the highest-leverage interventions in the field. Industrializing a small field: Lessons from Vannevar AI Safety needs to quickly transform from a "community" to an "industry"; the Manhattan Project is a good example! Many folks know about the "Einstein letter" [1] which was dated 87 years ago today. As legend has it, this led to the formation of the Manhattan Project. But actually this is only half the story: the original forays were plagued by institutional apathy, with multiple occasions when key scientific reports just didn't move. In one famed instance, the Brits realized that the bomb would only require a few kilograms of uranium (instead of tons from earlier estimates), and sent a report to the head of the American uranium committee... who put it in h…