This Diabetes Technology Network UK has issued a statement proposing nine policy points to help ensure safety and effective use of Generic Large Language Models (LLMs) – such as ChatGPT or Grok – for people living with diabetes.
While it is considered that LLMs offer potential as supplementary tools for diabetes management by assisting with tasks like initial carb estimation from food images and identifying patterns in glucose data, DTN-UK believe that it is paramount that healthcare professionals and people living with diabetes understand these systems are not a substitute for medical expertise and cannot safely provide definitive medical advice or insulin dosing recommendations.
Generic LLMs have several critical limitations and demonstrate significant variability and unreliability in visual carb counting, often misinterpreting food items or providing vastly disparate estimates. When analysing personal glucose data, these models can misinterpret images or misrepresent data duration and content, leading to potentially misleading analyses. Furthermore, LLMs have been shown to occasionally provide erroneous or dangerous advice (eg freezing insulin), necessitating extreme caution.
The nine policy points can be found HERE.
