Dirt up AI examples -- sometimes I wonder about the disconnect between the epistemic cultures of digital cloud techie types and folks that deal with stuff irl. A buddy from HS who does digital marketing for fancy global folks shared a draft email for my (successful!) local water board campaign. There were a lot of factual errors. It was a clever use of AI in some ways though I wonder about 1) how AI gets local data wrong and 2) a sort of cavalier cloud attitude about what's useful. Regarding 1) that's not something that fancier models will solve per say. There are data availability and organization issues. See WhatWaterCosts.org for a great protocolized approach to address those. Regarding 2) that's why I think that as AI deploys there will be a lot more dirt up innovation from folks that have that local knowledge. That's more scarce. The cloud central stuff is increasingly getting commodified further and further upstream.