Standard evaluations often reward a confident guess more than calibrated abstention. This brief shows how incentives turn uncertainty into a losing strategy and proposes interface patterns that make 'I don't know' useful rather than embarrassing.
Status · CommissionedCommissioned. Research roots are public; the full essay is in the queue.
Editorial status · Commissioned. Research roots published; full essay in editorial queue.
Research roots · hallucination incentives; calibration; selective prediction.
Standard evaluations often reward a confident guess more than calibrated abstention. This brief shows how incentives. Reviewer 2's verdict: Hold for evidence, publish the question.
← Back to the ledger