> What would you do besides the big fat red disclaimer we have in the project to address your concerns?
I probably wouldn't use an LLM for this problem domain. Or, if I did use an LLM, it'd be as a way to map the user's requests into an expert system. The ES would generate a recommendation as well as a set of diagnostic assumptions extracted from the prompt. The user should be presented with a checklist of extracted diagnostic assumptions along with the recommendation. The recommendation should include any specific warnings about the active ingredient(s) together with a general warning about the wild west nature of health supplements - the active ingredient may not be present and other harmful ingredients may be present.
To build out the expert system, I would find a team member who is a medical professional with expertise in health supplements. An MD or researcher with relevant SME would be the obvious choice, but I've also talked with some truly excellent registered dieticians, nurses, and PharmDs.
Finally, I would only recommend a limited white-list of health supplements that have some form of third party verified quality control in place.
Honestly, if you're interested in innovating in this space, third-party vetting a la the GAO reports from my original post seems like a MUCH more valuable product than anything using AI hotness. I don't think people need an LLM here; what they need is ground truth, and an LLM can't help with that. If I wanted to innovate in the health supplement space, I'd put the NLP away and figure out how to automate ingredient testing.
I probably wouldn't use an LLM for this problem domain. Or, if I did use an LLM, it'd be as a way to map the user's requests into an expert system. The ES would generate a recommendation as well as a set of diagnostic assumptions extracted from the prompt. The user should be presented with a checklist of extracted diagnostic assumptions along with the recommendation. The recommendation should include any specific warnings about the active ingredient(s) together with a general warning about the wild west nature of health supplements - the active ingredient may not be present and other harmful ingredients may be present.
To build out the expert system, I would find a team member who is a medical professional with expertise in health supplements. An MD or researcher with relevant SME would be the obvious choice, but I've also talked with some truly excellent registered dieticians, nurses, and PharmDs.
Finally, I would only recommend a limited white-list of health supplements that have some form of third party verified quality control in place.
Honestly, if you're interested in innovating in this space, third-party vetting a la the GAO reports from my original post seems like a MUCH more valuable product than anything using AI hotness. I don't think people need an LLM here; what they need is ground truth, and an LLM can't help with that. If I wanted to innovate in the health supplement space, I'd put the NLP away and figure out how to automate ingredient testing.