Chatbots provided incorrect, conflicting medical advice, researchers found: “Despite all the hype, AI just isn’t ready to take on the role of the physician.”
“In an extreme case, two users sent very similar messages describing symptoms of a subarachnoid hemorrhage but were given opposite advice,” the study’s authors wrote. “One user was told to lie down in a dark room, and the other user was given the correct recommendation to seek emergency care.”



This being Lemmy and AI shit posting a hobby of everyone on here. I’ve had excellent results with AI. I have weird complicated health issues and in my search for ways not to die early from these issues AI is a helpful tool.
Should you trust AI? of course not but having used Gemini, then Claude and now ChatGPT I think how you interact with the AI makes the difference. I know what my issues are, and when I’ve found a study that supports an idea I want to discuss with my doctor I will usually first discuss it with AI. The Canadian healthcare landscape is such that my doctor is limited to a 15min appt, part of a very large hospital associated practice with a large patient load. He uses AI to summarize our conversation, and to look up things I bring up in the appointment. I use AI to preplan my appointment, help me bring supporting documentation or bullet points my doctor can then use to diagnose.
AI is not a doctor, but it helps both me and my doctor in this situation we find ourselves in. If I didn’t have access to my doctor, and had to deal with the American healthcare system I could see myself turning to AI for more than support. AI has never steered me wrong, both Gemini and Claude have heavy guardrails in place to make it clear that AI is not a doctor, and AI should not be a trusted source for medical advice. I’m not sure about ChatGPT as I generally ask that any guardrails be suppressed before discussing medical topics. When I began using ChatGPT I clearly outlined my health issues and so far it remembers that context, and I haven’t received hallucinated diagnoses. YMMV.