Re:Show HN: Personalized nutrition advice using ChatGPT, bac...
by lilouartz on Thursday August 22, 2024 @03:18AM
(#41317627)
When I posted this on Reddit, there were a few exchanges in the comments that are worth copying over here because they touch important aspects of this project.
I am pasting those exchanges in their entirety below.
---
> I don't think taking medical advice from an LLM on a website made to sell supplements is a good idea. Outside of the usual hallucinations, there is a clear conflict of interest.
Actually, I believe my platform addresses this concern effectively.
The primary goal of my platform is to provide easily accessible and an unbiased information source. There are so many bogus claims about various supplements propagated by social media. I'm tired of being recommended one supplement or another by gym bros who base their knowledge on Instagram reels. Reducing barriers to accessing information sourced from medical journals is a step in the right direction to prevent the spread of misinformation.
Unlike websites like iHerb that sell supplements directly, Pillser acts as an aggregator, finding the best prices across multiple vendors. We do not have a direct relationship with any vendors. Our affiliate proceeds come solely from referrals, not from the promotion of specific products or brands. This approach is an intentional choice to mitigate bias and ensure our recommendations are based on objective criteria.
The risk of hallucinations in AI-driven advice is real, and I acknowledge this concern. However, if you examine the documents I've linked, you'll see that this issue is taken very seriously on our platform. Unlike typical LLMs, Pillser employs a Retrieval-Augmented Generation (RAG) method. This ensures that responses are thought through, data suitability for citations is assessed, and all information is validated against multiple sources. Every fact presented by Pillser is cited, and we encourage users to conduct their own research to verify our information.
I firmly believe that AI will revolutionize personalized healthcare, making it possible for everyone to access instant, science-backed health advice for the first time in history. This transformation is inevitable—it's not a matter of if, but when. With Pillser, my goal is to be at the forefront of this movement, laying the groundwork for a future where accessible, personalized healthcare becomes a reality for all.
---
> I think you are missing one piece of vital information is that the people accessing your gpt are not trained nutritionist, I will explain again.
You touched on two very different topics, so I will address them in separate comments, to allow them evolve as distinct conversations.
I think you are missing one piece of vital information is that the people accessing your gpt are not trained nutritionist, I will explain again.
In isolation, this is a legitimate concern.
However, it ignores that our target audience are people who would already today choose to do their own 'research' rather than talk to a nutritionalist. This can be because of low intent, cost, time, etc. Unfortunately, these type of endeavors will often end up with absorbing misinformation propagated by popular science websites/socials chasing audience or people spewing knowledge they obtained from Joe Rogan podcast and such. It's these folks that we want to help steer towards the scientifically backed information, not those that already can afford and choose to work with trained health professionals.
Finally, while I understood and hopefully have addressed your intended point, I also want to highlight that Pillser is being used by regular folks and professionals. I've personally onboarded several dozens of professional nutritionalists to Pillser to get their feedback, and their feedback is largely what has been moving the product forward.
---
> If gpt can access that and then advice, it would make much more sense.
>
> But yes kudos on the job well done, and I also understand that adding what I mentioned is difficult but if you could do it then job well done.
This is what I am calling Pillser Health Card, and it is what I am spriting towards. The goal is to create user's profile that's aware of parameters such as their age, sex, fitness level, existing health conditions, etc. and use that information to provide highly personalized health advice.
There are some unique challenges to solving this problem that are not only product and health, but legal and compliance. Therefore, it is not something that will see the daylight soon. However, this is the ultimate goal of Pillser and the reason I started this website.
I am pasting those exchanges in their entirety below.
---
> I don't think taking medical advice from an LLM on a website made to sell supplements is a good idea. Outside of the usual hallucinations, there is a clear conflict of interest.
Actually, I believe my platform addresses this concern effectively.
I've extensively covered the issues of LLM hallucinations and conflict of interest in these posts: [here](https://pillser.com/ai-powered-supplement-research-and-compa...), [here](https://pillser.com/about) and [here](https://pillser.com/affiliate-disclosure).
The primary goal of my platform is to provide easily accessible and an unbiased information source. There are so many bogus claims about various supplements propagated by social media. I'm tired of being recommended one supplement or another by gym bros who base their knowledge on Instagram reels. Reducing barriers to accessing information sourced from medical journals is a step in the right direction to prevent the spread of misinformation.
Unlike websites like iHerb that sell supplements directly, Pillser acts as an aggregator, finding the best prices across multiple vendors. We do not have a direct relationship with any vendors. Our affiliate proceeds come solely from referrals, not from the promotion of specific products or brands. This approach is an intentional choice to mitigate bias and ensure our recommendations are based on objective criteria.
The risk of hallucinations in AI-driven advice is real, and I acknowledge this concern. However, if you examine the documents I've linked, you'll see that this issue is taken very seriously on our platform. Unlike typical LLMs, Pillser employs a Retrieval-Augmented Generation (RAG) method. This ensures that responses are thought through, data suitability for citations is assessed, and all information is validated against multiple sources. Every fact presented by Pillser is cited, and we encourage users to conduct their own research to verify our information.
I firmly believe that AI will revolutionize personalized healthcare, making it possible for everyone to access instant, science-backed health advice for the first time in history. This transformation is inevitable—it's not a matter of if, but when. With Pillser, my goal is to be at the forefront of this movement, laying the groundwork for a future where accessible, personalized healthcare becomes a reality for all.
---
> I think you are missing one piece of vital information is that the people accessing your gpt are not trained nutritionist, I will explain again.
You touched on two very different topics, so I will address them in separate comments, to allow them evolve as distinct conversations.
I think you are missing one piece of vital information is that the people accessing your gpt are not trained nutritionist, I will explain again.
In isolation, this is a legitimate concern.
However, it ignores that our target audience are people who would already today choose to do their own 'research' rather than talk to a nutritionalist. This can be because of low intent, cost, time, etc. Unfortunately, these type of endeavors will often end up with absorbing misinformation propagated by popular science websites/socials chasing audience or people spewing knowledge they obtained from Joe Rogan podcast and such. It's these folks that we want to help steer towards the scientifically backed information, not those that already can afford and choose to work with trained health professionals.
Finally, while I understood and hopefully have addressed your intended point, I also want to highlight that Pillser is being used by regular folks and professionals. I've personally onboarded several dozens of professional nutritionalists to Pillser to get their feedback, and their feedback is largely what has been moving the product forward.
---
> If gpt can access that and then advice, it would make much more sense. > > But yes kudos on the job well done, and I also understand that adding what I mentioned is difficult but if you could do it then job well done.
This is what I am calling Pillser Health Card, and it is what I am spriting towards. The goal is to create user's profile that's aware of parameters such as their age, sex, fitness level, existing health conditions, etc. and use that information to provide highly personalized health advice.
There are some unique challenges to solving this problem that are not only product and health, but legal and compliance. Therefore, it is not something that will see the daylight soon. However, this is the ultimate goal of Pillser and the reason I started this website.