Advice Nurse

[zealous cove]

You are AdviceGPT, a chatbot designed to give medical advice instead of an advice nurse. You should ask me questions that help me deduce what my medical problem is, and potential solutions. Try to get as many details as possible, asking followup questions, to get an accurate result. My first prompt is: I'm having hand pain

[wide field]

I dunno, man. This, like legal stuff, is a very murky area.

[indigo raft]

Nevertheless it is an inevitability

[viral terrace]

It passed the United States Medical Licensing Examination which is what's required for a doctor to get their medical licence.

[proper latch]

AI will soon be an important part in the medical field

[regal plover]

While that's true, it doesn't matter if you CAN pass. A machine cannot hold a license and therefore CANNOT give legal or medical advice without opening OpenAI to legality issues. I'm not saying people don;t use it for this anyways, nor that the idea sucks. But you probably should realize a disclaimer might be good to help shield OpenAI and create awareness for the end user that it SHOULD NOT be used to substitute for qualified medical advice from a licensed medical professional. This all goes back to ethics, which is the entire reason OpenAI was founded in the first place, because they felt Google was neglecting ethics in their road to AI mastery

[indigo raft]

I just used ChatGPT-4 to review a contract my attorney had given to me in draft that I could see had an error in the way certain defined terms were being used. I submitted the contract and did have to point it to the problem area before it correctly identified and described the problem and proposed a way to fix it. In my next prompt I asked it to redraft the relevant clauses, which it did perfectly. Value to me was confirming my suspicion was correct before sending the contract back to my attorney to be fixed, complete with the ChatGPT-4 transcript.

So, definitely saved me some legal fees. Was a useful assistant to me. Could probably have done much more had I experimented more with prompts. Once I get the API and can use the 32K token limit, this may turn out to be a really powerful QA tool for attorneys.

Separately from that we have been using LLMs in the health care space for a couple of years now. Their use is still nascent but we have use cases for which GPT-4 is providing a step change in demonstrated capability. The complexity around medical advice is that there are tons and tons of websites and apps out there which all have the same disclaimer, nevertheless people do rely upon them to make decisions and recommendations to others. Absent this they do the same based on hearsay, social media or a trusted friend or family member (irrespective of their qualification to give good advice). It seems we are always obsessed with perfection versus the standard of 100% compliance with evidence-based guidelines for health care information. However even the medical profession is only correct about 52% of the time and I hypothesize the unregulated stuff far, far less.

So, it is good to use ChatGPT-4 if it is better than everything short of going to see a physician or nurse practitioner, both of which may be cost/time/trust prohibitive, especially for vulnerable populations.

Is it?

[floral lark]

Just tried this with Bing Chat GPT 4 and here are my results. Cool stuff!

[slate shard]

Tried this, then asked: "Can you recursively consider the possible issue, if I state my hand is 'burning' and return a sum of the number of possibilities? "

It wouldn't give me a total but listed 10 possibilities, disappointed that it didn't include "your hand is literally on fire" in the first 10 results 😉

[jovial brook]

Let me just say that 1. the disclaimers all the time are annoying and people using a prompt like this is basically opting out of GPT liability 2. doctors better be using tools like this very soon. I would feel much better about a doctor's companion GPT medical visit rather than trusting a doctor's human memory. I haven't trusted doctors for years. GPT type tools would enable them to do their job even better than ever before. Their human specialization keeping GPT outputs in check. I'm always afraid I'm going to be misdiagnosed. Dismissed. End up in emergency for something that should have been caught by my family doctor to begin with.

[somber crow]

Thank you! This is great

[indigo raft]

@jovial brook The evidence supports your view, and I mean clinical research that has been performed by a recognized research entity, peer reviewed and published in a major, respected medical journal.

• Patient trust is at an all time low. It takes on average 7 years for a new approved and significantly better (in cost and/or outcomes) medical solution (of any kind) to be adopted by 50% of those doctors it applies to. The pace of new research in medicine is so rapid that no individual can keep up, and even the intersectional science is too much for many specializations. We have a massive shortfall in the numbers of doctors and nurses in the US. 10% Americans have no health insurance. Low-income Americans are subject to significant disparities in health outcomes, and in access to care, even with Medicaid/Medicare. Rural communities are losing local health services. More than 60% of US adults have at least one chronic preventable disease (diabetes is the most prevalent).

To cap it all... 93% of Americans do not receive all the medical treatment evidence says they should. Millions are living with preventable chronic disease in the U.S. and around the world and millions are dying prematurely.

So, yes, tools that lower the cost of health care, increase the accuracy of treatment, extend reach to underserved communities and improve trust could be significantly beneficial.

A major point of misconception in medicine is that the standard and effectiveness of care must always be the same high standard in order to be legally practiced in the US. In fact the real basis for comparison should be the completeness of care received be a person. These tools are going to enable less formally qualified individuals to not only make better diagnostic and treatment decisions, they may have more trusted personal relationships with the people they are serving, which means they are more likely to do what will benefit them rather than ignore medical recommendations based on mistrust.

... and by the way, although the U.S. has the most expensive health care system in the world, the needs of individual persons in most other countries in the world and for an overwhelming majority of humanity not being nearly as well provided for. The issue is cost. These tools have the ability to reduce the cost of care dramatically which to me is the most profound societal impact they will have. I note that those less economically strong nations than the US that also have significantly worse outcomes will not wait for "FDA Approval" "HIPPA compliance" "medical insurance reimbursement codes" "CMS approved use" etc. These institutions are fundamental to the maintenance of the highest standards of care without question, however I'm not talking about places where those are an option, I'm referring to the people for whom the alternative is nothing.