Why do AI models confidently make things up, and can it be fixed?

The answer

AI hallucination occurs when language models generate plausible-sounding but factually incorrect information with high confidence. It stems from how LLMs work: they predict statistically likely next tokens rather than retrieving verified facts. A 2025 study found that simple prompt-based mitigation cut GPT-4o's hallucination rate from 53% to 23%, while temperature adjustments alone barely helped. Retrieval-Augmented Generation (RAG) grounds responses in external knowledge sources, reducing fabrication by anchoring outputs to verified documents. Novel techniques like Cross-Layer Attention Probing train lightweight classifiers on a model's internal activations to flag likely hallucinations in real time.