AI Hallucination Risks Loom Large in Mathematics and Healthcare

Large language models work primarily by predicting the next token. Their output may appear fluent, but that does not mean they understand facts or can perform rigorous reasoning. Errors in mathematics, data analysis, medical advice and academic citations could directly affect research, diagnoses and decision-making, making professional verification essential.

Recent reports said AI hallucination rates reached as high as 60% in some tests, with mathematical calculations, medical information and source citations particularly prone to errors. The reports did not identify the research institutions, sample sizes or publication dates, so the 60% figure still needs to be checked against the original research. Experts recommend a “use as a reference, but verify” approach, with humans retaining final oversight.