Imagine having a casual conversation with your computer, mentioning a few health details, and it accurately predicts your risk of a heart attack. While this sounds like science fiction, researchers are actively exploring the potential of large language models (LLMs), like the technology behind ChatGPT, to do just that. A recent study, using data from the extensive UK Biobank, put LLMs like ChatGPT and GPT-4 to the test, analyzing their ability to predict myocardial infarction (MI), commonly known as a heart attack. The results? While promising, there's still a long way to go. The study revealed that these powerful AI models struggled to achieve the accuracy needed for reliable clinical use. Although GPT-4 showed the best performance, its accuracy was still below the desired threshold. One surprising finding was that feeding the LLM information in a step-by-step, logical way (a technique called "Chain of Thought" prompting) actually *reduced* the accuracy of the predictions. This suggests that these models don’t reason like human doctors, who build a diagnosis piece by piece. Instead, they seem to perform better when given all the information upfront. So, where does this leave the dream of AI-powered heart attack prediction? While current LLMs aren’t ready to replace doctors, the research points to some critical next steps. Future medical AI will likely need a deep understanding of both language and quantitative medical data, combined with the ability to reason logically. This research reminds us that even the most sophisticated AI needs careful development and rigorous testing before it can be trusted with our health.