Perplexity
Noun · AI & Machine Learning
Definitions
A metric that measures how well a probability model predicts a sample. Lower perplexity means the model is less 'surprised' by the data. Informally, it measures how confused the model is — a perplexity of 1 means perfect prediction.
In plain English: A score measuring how surprised an AI language model is by a piece of text. Lower scores mean the model predicted the words better, so it understands the language more.
Etymology
- 1977
- Information theory formalizes perplexity as a measurement of how well a probability model predicts a sample — lower is better
- 1990s
- NLP researchers adopt perplexity as the standard metric for evaluating language models
- 2022
- Perplexity AI launches as a search engine, choosing the name to signal its foundation in language model evaluation