AIExplainer

What is hallucination?

A false or misleading perception generated by a machine learning model

In AI, a hallucination occurs when a model produces an output that is not based on any actual input or data, but rather on the model's own biases or errors. This can happen in tasks such as image or text generation, where the model may produce something that is not grounded in reality.

A hallucination in AI is like a dream or a false memory in humans - it's a creation of the mind that isn't based on actual events or facts

A language model that generates a sentence that is grammatically correct but semantically nonsensical, such as 'the color blue is a type of food', is an example of a hallucination

Hallucinations can be used to evaluate the performance of AI models, particularly in tasks that require generating new data. They can also be used to identify biases or errors in the model.

One common misconception is that hallucinations in AI are always a bad thing. However, they can also be a sign of creativity or innovation in the model.

The term 'hallucination' in AI was first used in the 1990s to describe the phenomenon of models producing false or misleading outputs. Since then, it has become a widely recognized concept in the field.

false positive misclassification error

Three products for different needs — explore what’s relevant to you.