Definition
Plain language
A cutoff in a probability distribution, like the value below which a certain percentage of outcomes fall.
As stated in the literature
A point in a probability distribution corresponding to a specified cumulative probability; in LLM forecasting, models are often asked to output a set of quantiles characterizing a predictive distribution.
Also called: quantiles
Why it matters: Quantiles let a forecast express uncertainty honestly — a range of outcomes with calibrated probabilities — rather than collapsing everything into a single point guess.
For example, the 90th-percentile quantile of next month's sales forecast is the figure that should be exceeded only 10% of the time.
Heard on the show
“They ask for five quantiles — basically the floor, the lower-middle, the median, the upper-middle, and the ceiling — so they can read off the entire shape of what the model believes.”Episode 069 — When Smarter Models Forecast Worse: The Hidden Failure Mode in LLM Predictions