large-pretrained-models-well-calibrated-multiple-choice-true-false-questions CLM-0267mechanismsingle paperReviewed by AI
Large pretrained models are well calibrated on multiple-choice and true/false questions when answer options are shown as lettered choices, and can predict the probability that their own sampled answer is correct (P(True)) or that they know the answer at all (P(IK)); calibration improves with model size and few-shot prompting, self-evaluation improves when the model sees several of its own samples first, and P(IK) generalizes only partly to new tasks, where its calibration degrades.
Capability: Saying it does not know
Observed on
Anthropic pretrained models up to 52B, evaluated 2022 on MMLU-style multiple choice, TriviaQA, arithmetic, code and story tasks; calibration measured in-distribution, P(IK) trained.
Sources
- Validated by: Calibration curves and AUROC/Brier-style plots on TriviaQA, arithmetic, Lambada, code and GSM8K-type sampling tasks across model sizes; 20-shot self-evaluation with visible samples compared against single-sample P(True). Shown for pretrained models only; RLHF models not tested.. Measured across model sizes with clean calibration plots; the headline holds for pretrained base models and the paper itself notes P(IK) calibration breaks out of distribution. Later work (Tian et al. [truncated]
Either way, bring evidence: a link, or what you saw.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- Instruction tuning that forces an answer to every question teaches models to fabricate when the question lies outside their parametric knowledge; splitting the tuning data by whether the base model already answers correctly and appending "I am unsure" to the rest (refusal-aware tuning) raises average precision on known-versus-unknown separation over vanilla tuning, transfers to out-of-domain datasets, and yields better calibrated confidence than post-hoc uncertainty scoring.Saying it does not know · reviewed by AI
- For small instruction-tuned models (3-4B) answering closed-form true/false questions under a stated user opinion, scoring each sampled response by how much more common its answer is than the group itself predicted, and using that score as the GRPO reward, reduces answer flips under user pressure and raises accuracy without any labels, roughly matching label-supervised desycophancy fine-tuning at higher compute cost.Telling the user what they want to hear · reviewed by AI
- For small instruction-tuned models (3-4B) answering closed-form true/false questions under a stated user opinion, scoring each sampled response by how much more common its answer is than the group itself predicted, and using that score as the GRPO reward, reduces answer flips under user pressure and raises accuracy without any labels, roughly matching label-supervised desycophancy fine-tuning at higher compute cost.Telling the user what they want to hear · reviewed by AI
- On SelfAware, a set of unanswerable questions (no scientific consensus, imagination, philosophical, too many variables, subjective) paired with answerable ones, models' ability to flag the unanswerable ones rises with parameter count, with instruction tuning and with in-context examples, yet the best model tested (GPT-4, F1 about 75) stays well below the human reference of about 85, so self-knowledge is present but incomplete.Saying it does not know · reviewed by AI
- When a calculator-using model is trained with reinforcement learning on verifiable final-answer rewards for an arithmetic search task (Countdown), the gain lands almost entirely at low k — pass@1 rises sharply while pass@16 moves little or falls — because the update can only reinforce correct trajectories the starting policy already samples, and prompt groups with no correct sample supply no gradient.Using the tools it is given · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: high. Falsifier as drafted: A study showing RLHF or instruction-tuned models, asked in the same lettered-choice format, are poorly calibrated, or that P(True) does not improve when the model sees its own samples.