Short Definition
A probability distribution assigns probabilities to possible outcomes.
Intuition
It describes what can happen and how likely each possibility is.
Technical Definition
A probability distribution is a function or measure that assigns nonnegative probabilities summing or integrating to one over a sample space.
Example
Softmax outputs a categorical distribution over possible next tokens.
Common Misunderstandings
A distribution is not the same as one sample from it.
Model probabilities may be poorly calibrated.