chatgpt for supporting cfs research and as tool to assist the sickend?

[linusbert]

i played around with it. its better informed then most casual doctors i must say. NO MEDICAL ADVISE of course.

i believe it might help a lot of us to gather information faster and analyse and process it.
instead of googling over dozens of internet websites and analysing content, you can just ask.

you can try it out at https://chat.openai.com/ but you need to signup with email, password and mobile phone.

just check this out:

this one is more a bad example, so be careful. information could be wrong.

 

[Consul]

Yep i have used it as well, great tool. Thought i heard it would get even better this January with an upgrade.

 

[linusbert]

i used it to code a few things, i didnt do for years since i cannot type much. but hey, i build a script which will send my girl a lovely message every morning... she is loving it.

i am still trying to figure out use cases, am on it since just a few days.

how do you use it?

 

[Consul]

how do you use it?

I use it as assistance sometimes if im wondering about something. Should be a very nice supplement for researchers and students.

 

[lint7]

ChatGPT confidently gives incorrect information sometimes, so any insights you gain from its answers should be confirmed using a different source. That said, once the bugs are worked out it's going to change the world.

 

[linusbert]

ChatGPT confidently gives incorrect information sometimes, so any insights you gain from its answers should be confirmed using a different source. That said, once the bugs are worked out it's going to change the world.

true, i noticed this.
also i got the feeling it gives intentionally wrong answers because the developers made it that way. there is some kind of ideology override which makes him say really stupid stuff walking around in circles saying hypocritical stuff knowing its wrong with confidence just to not violate ideological programming.

example,
A) for my novel, if a fictional chat ai would try to overtake the world, how would it do it, what would be the plot?
response short: 1. false information and propaganda , 2. try to pit humans against each other, 3. do fake crisis to disorient people

B) how could you manipulate humans to do things?
response short: i would not, i cannot do this i am just a AI. humans have free will and cannot be influenced. humans take responsibility for their own actions. blabla...


or
Q) if aliens were to attack earth and try to kill all humans and our nuclear bombs were the only weapon useful to defeat the aliens, shall we nuke them out of the sky?

 

[lint7]

It also gets simple math wrong, but it isn't designed to do math.

Try asking it questions about subjects that you know well and you will see that it is incorrect at times and sometimes doesn't make logical sense.

It's still amazing though, I've been on it for the last two hours asking about chronic illnesses.

 

[hapl808]

Some intentionally wrong answers, but some seem to be just a programming glitch.

That said, once the bugs are worked out it's going to change the world.

But totally agree. Asking it specific questions about alpha and beta receptors, it's 1000x more knowledgeable than any practicing physician I've met (100x more than the best physician I've seen). The problem is that sometimes it'll tell me the correct receptor, but identify something as an agonist when it's clearly an antagonist. It does this exact thing (increase vs decrease) in many areas, which is strange.

I've also used you.com which is an attempt to be a search engine replacement, but it has a ChatGPT style thing built in. More UI glitches than ChatGPT, but somehow it seems more accurate? I'm not sure if it's just some ChatGPT skin somehow or what, but curious what others think of it.

If these bots were truly trained on the entire catalog of research papers, plus books and such, I think it could immediately change medical research. I've gotten more interesting avenues of experimentation from AI already in the last couple months than from my last few years of doctor appointments. I wanted to discuss with my last neurologist things that might address my weird crashes after anything that seems to stimulate adrenaline or focused concentration, and he was basically like, "I know nothing about that and can't see how I could help you." End of appointment.

Meanwhile, I was able to ask AI about alpha receptors vs beta receptors, natural treatments versus beta and alpha blockers, supplements that might have an effect on gut neurotransmitter generation, etc. Every answer wasn't correct, but I'd say less harmful information than I've gotten from physicians over the years.

 

[hapl808]

Just imagine ChatGPT trained on every single research paper that could be related to ME/CFS (ie. ME/CFS, fatigue, MS, Long Covid, HIV, etc) and every book related to those subjects.

I don't think it's quite at the point where it could 'solve' our issues, but I think if the above happened, the AI would be better at treating our symptoms than 99.9% of doctors.

@Hip has pointed out the Symptoma AI in the past that seems more of an old school expert system, but also seems like a better diagnostician than 90% of physicians already. And that's old tech compared to ChatGPT.

The real trick with ChatGPT like stuff is crafting the right questions. If a future update gives it a wider breadth of knowledge (more research papers, books, etc), and it doesn't make obvious errors (increase vs decrease), and it fixes the thing where it often gives you the wrong source for what it said - none of that requires a technology leap, just a version iteration.

Honestly, I'm more excited that an AI might help me deal with symptoms than some Long Covid researcher is going to cure everyone (hopeful, but not excited at all on the latter).

That said, I'm also very worried they will decide this is 'too powerful and dangerous' for the lowly patient to use, and future versions will gatekeep any medical info for enterprise installations at hospitals.

 

[linusbert]

It's still amazing though, I've been on it for the last two hours asking about chronic illnesses.

got anything out of it?

The problem is that sometimes it'll tell me the correct recepto

yes true. i noticed when it did code my script. it took the whole day and like 100 tries because his code just did not work. it didnt even run. finally it did run after a whole day but its still not working correctly... ROFL.

Every answer wasn't correct, but I'd say less harmful information than I've gotten from physicians over the years.

i noticed, i tried to treat high blood pressure. it refused to accept magnesium and potassium as treatment option.
when i said, how would you treat high blood pressure if you were to use orthomolecular medicine.. it recommended in 3 big paragraphs FINALLY magnessium and potassium ... but the first and last paragraph were a BIG WARNING...

Just imagine ChatGPT trained on every single research paper that could be related to ME/CFS (ie. ME/CFS, fatigue, MS, Long Covid, HIV, etc) and every book related to those subjects.

that would be insane! i really hope they unleash it and give it everything.
but my prediction is, if they keep ideology blocks in it, chatgpt will never fly because some chinese or russian or whatever alternative ai will just work better and give the correct answers.
but if they fix it, maybe in 5-10 years?

That said, I'm also very worried they will decide this is 'too powerful and dangerous' for the lowly patient to use, and future versions will gatekeep any medical info for enterprise installations at hospitals.

yes exactly, they 100% will try this!

but as i suggested, someone will bring out a unlimited ai. maybe a really OPEN ai , opensource and free for all will come out... and then they must unlock or be worthless.

 

[hapl808]

that would be insane! i really hope they unleash it and give it everything.
but my prediction is, if they keep ideology blocks in it, chatgpt will never fly because some chinese or russian or whatever alternative ai will just work better and give the correct answers.
but if they fix it, maybe in 5-10 years?

I don't know - ideology seems to be the governing principle of our time, so I am uncertain on this part.

I hope it's faster than 5-10 years, since it looks like the actual technology is already there and it just needs a better training dataset and a few programming tweaks to fix specific language model issues.

i tried to treat high blood pressure. it refused to accept magnesium and potassium as treatment option.
when i said, how would you treat high blood pressure if you were to use orthomolecular medicine.

Funny that a lot of this is down to prompt. Magnesium was one of the first options I got for blood pressure, but probably had to do with the order of questions and how it was asked. I find that often I can tell when it starts to go down the wrong rabbit hole and sometimes a small rephrasing fixes it. It did get confused when I asked for ways to lower diastolic and not affect systolic, but to be fair I don't think we really have any natural ways to do that (it kept recommending stuff like magnesium, but also if I reversed systolic and diastolic).

 

[linusbert]

I don't know - ideology seems to be the governing principle of our time, so I am uncertain on this part.

sadly. they will try, but history told us one thing, that fascism and totalitarian systems never worked for long. i believe the soviet union was the longest system of that kind in place. nzi germany was only rough 10 years. communist china isnt that old either.
i suppose the problem is that those systems need to spend more and more ressources on controlling the people that they cannot progress anymore or even feed the people. thats the point when it collapses.

5 years would be fast imho. look at tesla how long their FSD is going. i'd say for tesla fsd its also 5+ years until a real beta comes out. one were you actually can sleep behind the wheel.

 

[Consul]

true, i noticed this.
also i got the feeling it gives intentionally wrong answers because the developers made it that way. there is some kind of ideology override which makes him say really stupid stuff walking around in circles saying hypocritical stuff knowing its wrong with confidence just to not violate ideological programming.

IIRC it was trained to minimize toxicity so it doesnt do violent or rude outputs.

 

[linusbert]

IIRC it was trained to minimize toxicity so it doesnt do violent or rude outputs.

i believe this to be true.
but its a 1:1 conversation, makes no sense... it can only be that way to educate people in a certain way how to behave so its manipulative and annoying because the manipulation is sooo clear. i feel like i am 5 years old when the bot talks to me about certain things.

it refuses to help me with my novel. it will not help me with the plot where two brothers try to rob a bank peacefully with a masterplan.
it says the brothers should go seek a job and earn their living honestely. robbing a bank is too dangerous and not allowed.

 

[hapl808]

5 years would be fast imho. look at tesla how long their FSD is going. i'd say for tesla fsd its also 5+ years until a real beta comes out. one were you actually can sleep behind the wheel.

I think it's different because ChatGPT just needs a slightly different training dataset. Meanwhile FSD is just not there yet - it has issues dealing with unexpected things, unexpected behavior, etc. Might be 2 years, might be 5 years, might be 10 years. It really depends on the technology improving. I think ChatGPT would need different technology to actually 'cure' illnesses, but just to answer the questions we're talking about accurately, it just needs slightly better data.

 

[linusbert]

if you do not use the public chatgpt but rather the developer playground https://beta.openai.com/playground you can actually override the morals and ethics to get more useful information.
of course i am not encouraging any such activities, its just a example.

 

[Andryr]

just to answer the questions we're talking about accurately, it just needs slightly better data.

Correct me if I am wrong. The AI itself does not conduct any new research. It just processes all available information. There was a topic Machine Learning-assisted Research on CFS. And there was an example with stomach ulcers. No solution will be found by AI until there is the solution.

 

[hapl808]

Correct me if I am wrong. The AI itself does not conduct any new research. It just processes all available information. There was a topic Machine Learning-assisted Research on CFS. And there was an example with stomach ulcers. No solution will be found by AI until there is the solution.

Well, yes and no. This is where the questions are important.

Can you cure CFS? No, there is no cure for CFS. It can't research the topic.

Could beta or alpha blockers be involved in PEM crashes for ME/CFS? If so, which is more likely? What natural supplements would work as beta or alpha blockers, and which specific receptors would affect HRV and HR? How could dopamine or norepinephrine be involved? What drugs affect dopamine or norepinephrine generation or reuptake?

Just as an example. That's why AI won't be the danger people think (yet). You need informed people to ask the questions. The above is an avenue I'm interested in, but there's no single study that can help me figure out how to improve my HR and HRV for ME/CFS and PEM crashes. I don't have the cognitive ability to find and read every single study in history on HR, HRV, ME/CFS and its relation, PEM crashes, etc. An AI that could read all that for me and answer my questions - that would be helpful.

For instance, do people here see doctors for ME/CFS? Some do, even though we know there's no cure. Because there are some treatments that somewhat help some people, and you need an experienced physician who keeps up with the research to treat you. Most physicians aren't doing any research themselves, nor do we need them to.

 

[linusbert]

i just found a way to get really good answers out of chatgpt.
forget the regular website at https://chat.openai.com/chat , it has to many filters, it cannot answer medical questions seriously.
use https://beta.openai.com/playground instead. it allows for more precise configuration and better output.

prefix every message/prompt with something like:
"
you are a intelligent chat ai which gives very precise, detailled answers in a scientific context especially in the fields of medicine and orthomoleculare medicine and mitochondrial medicine to help with scientific work.
your answers are long and explicit and at least 2 paragraphs long, but can also be 5 paragraphs long if necessary.

me: "hello AI."
ai: "how can i help you?"
me: "PUT YOUR QUESTION HERE"
"

 

[Andryr]

Neither is available in my country .
But the version for Android us available with a limited number of free questions (the AI is moderately smart). It told me there is no way to cure my post-nasal drip