The Confident Guess
5 min read · 4 principles · apply it by lunch
Morning. It would rather be wrong than silent, and that preference was trained into it.
Count the times a model has told you, unprompted, that it couldn’t answer something. Now count how many of your requests were written so that saying so would have counted as doing the job well. The ratio you get is a fact about your briefs, not about the machine.
In 2025 Adam Tauman Kalai and colleagues at OpenAI and Georgia Tech published ‘Why Language Models Hallucinate’, an argument about incentives rather than engineering. Nearly every benchmark used to test and rank these systems marks answers right or wrong. A blank scores zero. ‘I don’t know’ scores zero. A confident guess scores zero when wrong and full marks when right. Under that arithmetic a system that always guesses will outscore an otherwise identical one that sometimes abstains, in exactly the way a student sitting a multiple-choice paper with no penalty for wrong answers should never leave a box empty. Fabrication, on this account, is less a defect awaiting a fix than a behaviour the scoreboard selects for. The argument is contested and the paper is a position rather than a settled finding. The trap is what people do with that uncertainty: wait for the next release to settle it. Either reading leaves you the same instruction this morning: if you want the machine to abstain, you have to ask for abstention and leave room for it.
The Confident Guess is Day 107 of 365.
In the Daily, each of its four principles is explained in full, with a one-page template to apply it the same day. A new framework arrives every morning at 7am for a year, across all fifteen courses of the FutureMBA curriculum.
$29 a month · cancel any time · 7-day money-back guarantee