Learn AI · Level 2 · How AI works · step 2 of 20 · word 1 of 2
What is calibration in AI?
A model is well calibrated when how sure it says it is matches how often it is right. A confident AI answer is not automatically a correct one.
In plain English
If a weather forecast says “70% chance of rain”, it should rain on about seven of every ten such days. AI models are the same: when a well-calibrated model is 90% sure, it should be right about 90% of the time. Chatbots are often poorly calibrated, especially in the confidence they express in words. They can state a wrong answer in the same firm tone as a right one.
An example
Ask a chatbot for the year a small local company was founded. It may answer with a precise year and no hint of doubt, even though it is guessing.
What to watch out for
Don’t judge an answer by how confident it sounds. OpenAI researchers said in 2025 that common tests reward guessing over saying “I don’t know”. See how to check an AI answer.
Learn AI path · Level 2 · How AI works · step 2 of 20
Next wordData contamination →See all of level 2