Articles

The article opens with a shocking real case: a 60-year-old man trusted ChatGPT's advice to replace table salt with Sodium Bromide. This led to severe Bromide poisoning and psychiatric symptoms, nearly resulting in hospitalization. This case, published in the authoritative medical journal Annals of Internal Medicine, highlights the serious misguidance AI can provide in non-professional fields, especially health. The article points out that despite improvements in the latest version of ChatGPT on similar issues, the AI hallucination problem and Echo Chamber Effect of Large Language Models (LLM) remain inherent limitations, as they essentially predict the next token rather than truly thinking. The article emphasizes that users' blind trust in AI output is a key factor leading to danger, and advocates for relying on professional advice and evidence-based decisions in health and safety matters, rather than blindly trusting AI.