{"id":6378,"date":"2025-12-19T13:33:40","date_gmt":"2025-12-19T13:33:40","guid":{"rendered":"https:\/\/www.stanventures.com\/news\/?p=6378"},"modified":"2025-12-19T13:33:40","modified_gmt":"2025-12-19T13:33:40","slug":"even-the-best-ai-models-still-hallucinate-study-finds","status":"publish","type":"post","link":"https:\/\/www.stanventures.com\/news\/even-the-best-ai-models-still-hallucinate-study-finds-6378\/","title":{"rendered":"Even the Best AI Models Still Hallucinate, Study Finds"},"content":{"rendered":"

A new benchmark shows that advanced language models, including newer GPT versions, continue to produce incorrect answers at a noticeable rate, raising ongoing concerns for businesses that rely on AI for factual analysis and decision-making.<\/b><\/p>\n

The <\/span>study<\/span><\/a> examined how well large language models handle a basic but critical task. Can they accurately pull facts from a given document without adding information that is not there?<\/span><\/p>\n

Researchers tested 37 models using 60 questions based on CNN News articles. Each question required a precise answer drawn directly from the text. If the article did not include the information, the correct response was \u201cnot given.\u201d Any attempt by a model to guess or infer missing details is counted as a hallucination.<\/span><\/p>\n

This setup closely mirrors how companies use AI in real workflows. From summarizing reports to answering questions from internal documents, businesses expect systems to stick to the source material. In these cases, guessing can be more damaging than saying nothing at all.<\/span><\/p>\n

\n
\n

<\/p>\n<\/span><\/div>\n