Definition
Plain language
Making a system answer with just one word or digit instead of letting it explain itself.
As stated in the literature
An evaluation format that constrains the model's output to a single token from a small answer set, letting the experimenter read the answer distribution directly from the output logits rather than sampling free text.
Also called: forced choice, forced binary, forced-choice format
Why it matters: It gives a clean, directly measurable answer distribution, making small shifts in a model's leanings visible that free-form text would hide.
For example, instead of letting a model explain its reasoning, you require it to answer only "A" or "B" so you can read off exactly how confident it was in each.
Heard on the show
“And on the framing, I agree: what's demonstrated is strong control of forced binary answers on questions where the model had no anchor.”Episode 243 — How a Hundred Meaningless Word Choices Add Up to Flip a Model's Answer