According to a TU Berlin study, ChatGPT models tend to make overly cautious recommendations, and are currently inadequate for the targeted management of patients in healthcare systems