Wanted to share this. I think this is the first preprint discussing the new custom GPT feature in a medical context. We noticed that ChatGPT was unsurprisingly

Wanted to share this. I think this is the first preprint discussing the new custom GPT feature in a medical context. We noticed that ChatGPT was unsurprisingly very bad at most clinical calculations doctors do on a daily basis (things we mostly just do on MDCalc.com). With code interpreter/web search, it will attempt a lot of them, but fails in unpredictable/chaotic ways. We made a simple API for some of these calculations, hooked it up to custom GPT and tested the custom GPT, showing it performed much better with a calculation tool on the tasks we assessed. To people on HN, this will not be a surprise, but we wanted to show the medical community the future of LLMs: able to interact with an environment of tools just like clinicians do.