Source: Marginal REVOLUTION
As AI models plateau on benchmark improvements, the constraint has shifted from algorithm design to data quality—and governments sit on the messiest, most consequential datasets. Getting AI to work on healthcare, benefits, permitting, and infrastructure requires not sophisticated models but unglamorous work: standardizing formats, fixing decades of inconsistent record-keeping, and making siloed bureaucratic databases actually talk to each other. This reframes the AI investment narrative from Silicon Valley's model-scaling obsession to the harder, less venture-backable problem of institutional data infrastructure.