uncertainty quantification

The estimation and communication of how uncertain a model's predictions are. Two sources are commonly distinguished: aleatoric uncertainty, irreducible randomness or ambiguity in the data (a blurry image, a question with several valid answers), and epistemic uncertainty, the model's lack of knowledge, which more data could reduce and which grows on out-of-distributionout-of-distribution, adj.: Describes inputs that differ from anything an AI system saw in training — like a model trained on daytime photos facing a night scene — where its answers become less reliable. inputs. Methods include BayesianBayesian inference, n.: Reasoning that treats unknowns as uncertain beliefs and updates them with evidence using Bayes' theorem. The result is not one answer but a range of answers, each with a probability. approaches (posteriors over weights, Laplace approximations, variational inferencevariational inference, n.: A way of approximating a hard-to-compute probability distribution by picking the closest match from a simpler family of distributions, turning a hard problem of adding up every possibility into an optimization problem.), deep ensembles of independently trained networks, Monte Carlo dropout, distance-aware and evidential models, distribution-free conformal predictionconformal prediction, n.: A way to wrap any prediction model so that, instead of one guess, it gives a range or set of answers that contains the truth a chosen share of the time, such as 90%., and post-hoc calibrationcalibration, n.: How well a model's stated confidence matches reality: when a well-calibrated system says it is 80% sure, it turns out to be right about 80% of the time.. For language modelslarge language model, n.: A computer program trained on enormous amounts of text to predict the words that come next. At vast scale, that simple skill lets it chat, answer questions, translate, summarize and write code., uncertainty is estimated from token log-probabilitieslog probability, n.: The logarithm of the probability a model gives to a token or a text. Computers use it because multiplying many tiny probabilities runs into rounding problems, whereas adding their logarithms does not., from the entropyentropy, n.: A measure of unpredictability. A fair coin toss has more entropy than a rigged one; an AI that is unsure of its next word has high entropy. or disagreement of several sampled answers ('semantic entropy'), from verbalized confidence, or from probesprobing, n.: Testing what information is hidden inside an AI model by training a small, separate predictor on the model's internal activity: if the predictor can recover something, the model has stored it somewhere. on internal activationsactivation, n.: The numbers a neural network computes inside itself as it processes an input — a reading of which of its artificial neurons are firing, and how strongly, at that moment., and is used to trigger abstentionabstention, n.: When an AI declines to answer because it is not sure enough — saying 'I don't know' or asking for more information — instead of guessing and risking a confident mistake., retrieval or human review and to flag likely hallucinationshallucination, n.: When an AI states something false or made up — a fake fact, quotation or source — fluently and confidently, as if it were true..

Disagreement among five sampled answers was our best uncertainty signal for routing questions to a human.