Brier Loss
A strictly proper scoring rule for evaluating the accuracy of probabilistic predictions that computes the squared error between predicted probabilities and true outcomes (0 or 1), penalizing both overconfidence and underconfidence.
Timeline (last 90 days)
The correction method tested by the researchers—averaging each answer with the complement of its paired answer—restores complementary consistency by construction and lowers Brier loss averaged over the two phrasings.
Brier loss is a strictly proper scoring rule, and the true probability is the unique optimal strategy; an uninformative predictor that always outputs 0.5 scores 0.25 on balanced data.
All Facts (2)
The correction method tested by the researchers—averaging each answer with the complement of its paired answer—restores complementary consistency by construction and lowers Brier loss averaged over the two phrasings.
50%UnverifiedBrier loss is a strictly proper scoring rule, and the true probability is the unique optimal strategy; an uninformative predictor that always outputs 0.5 scores 0.25 on balanced data.
50%