Why do hallucinations happen?

AI models don't know what a hand or words are, like we know. They just learn patterns from their training data.

Issues with spelling or jumbled text can happen because models learn based on pixels and not letters.

Logical errors, like stoves that are turned on with nothing on them, can be caused by stoves being turned on in training images.

Randomness: the model is "guessing" what the image should look like and sometimes gets it wrong.