Hallucination

AI makes things up. Confidently. Without flagging. Not malice, not a bug — the mechanism of the probability machine, working as designed, sampling tokens that 'look right' next without checking whether the resulting claim is true.

Hallucination is the confident assertion of unsupported, fabricated, or false content by a generative model, arising from the stochastic nature of token prediction. The model optimizes for linguistic coherence and conditional probability distributions rather than factual veracity. As the mechanism functions as designed—sampling tokens that satisfy local sequence patterns—it inherently lacks a grounding mechanism to cross-reference claims against verifiable external reality, leading to high-confidence outputs that are epistemically decoupled from the training data or provided context.