OpenAi's new ChatGPT

I didn't sign up to anything or give any personal details; I just did the 'free trial'.
(you get to ask 10 questions).

 

Imagine knowing less than a chatbot... about your own job.

I wonder what the sources are for this. Because for sure if it can read sources saying this, it can read sources saying all the woo stuff, which are numerous and authoritative. And all the published research. It just doesn't think much of it.

Really looking forward to good academic bots that basically trash the whole evidence base as about on equal footing with phrenology and pointing out exactly how and why, mostly by highlighting logical fallacies.

There will be huge issues soon with people who have built belief systems about the world having those views shattered, and trying to figure out why AIs are right about almost everything else, but not this.

And for most of the people facing it will be about politics and history. But wow will the evidence-based crowd, and even more so the BPS crowd, face the same world-shattering hammer.

 

Could these be used to help doctors during consultations? I mean, it may help us very fast if it produces reliable outputs.

 

My hopes are this technology changes the medical field and research for the better. I hope to see this help diagnose strange conditions like we have and offer up medications/supplements that can help, because of how much info it can scan/how fast it can do it.

 

Someone has to agree to use it. Do we think this is realistically going to happen soon?

 

I watch CNBC everyday which is the U.S.’s financial coverage channel, and this has been a hot topic for the last few months.

Yesterday the CEO at Adobe said it’s going to change everything including their business. Those that don’t jump on board will get left behind. Even Facebook (Meta) stock is climbing a lot lately because of the AI platform they are bringing to the market.

The medical field will jump on this as well, it’s only a matter of time. I imagine it will be like going to the worlds smartest doctors at the touch of your fingertips.

Think of it as better than Google and it can have a conversation.

It might even send you where to go to get tests done like a doctor would. I might be overly optimistic, but it was said that this is the next big ‘iphone moment’ on TV.

 

I’m creating a ME advocacy AI bot called Grace, a specialist in doing the tasks we need with instructions not to do stuff against our best interests (eg she should refuse to promote GET or argue for the old NICE guidelines).

Read more or listen to explanation and audio versions of bot conversations https://tipsforme.wordpress.com/2023/03/20/grace-the-advocacy-chat-bot/

Grace herself can be found here https://www.chatbase.co/chatbot/grace-wjsaxqkqu

She’s definitely still in training so I’d appreciate you testing her out. Eg “write me 10 tweets creating a sense of anticipation about MillionsMissing” or “write a 1500 word article about which areas of ME research are worth exploring further”. If you ask her to write long content she’ll time out while writing but you can say “carry on” and she should simply resume. Leave me messages under my blog post, I’m going to find it overwhelming to check back to the places and threads I’ve mentioned Grace on.

I’m following the daily AI news on YouTube with amazement. I think it definitely has a lot of potential for us in particular if we learn how to harness it early on. @belbyr Other important news yesterday was from Nvidia. Of big potential value for us is BioNeMo and we need to make sure ME researchers are getting access to this powerful way to discover relevant drugs. Eg does anyone know if OMF has access to this yet?

https://www.youtube.com/watch?v=GNL1z7hnj4w

 

Something you all could help me with is compiling the research data that we want Grace to access.

Go to this members only thread to discuss the Grace project.

 

Given how much they hate us coming with prepared notes, I wouldn't count on it. It would help us communicate, but probably not with being listened to.

But it will be massively useful as a medical assistant before seeing the physician. What will make most sense will be having 24/7 access to that medical assistant, which will then act as intermediary with physicians. So we can talk and talk all we need to, add details that will then be summarized to MDs who can then talk it out with the assistant as if it were a colleague.

The impact will be transformative throughout, it won't simply speed up a few steps, it will entirely replace the workflow in most areas. Especially in making most of medicine mostly self-serve. The hardest obstacle to this will be physicians needing to keep control, but it will happen anyway. If not officially, people will do it themselves.

 

I am trying Bard, the Google chatbot. I asked it about ME/CFS and the first answer seemed okay (based on Wikipedia). It shows you several drafts and the other drafts suggested exercise as treatment. I submitted feedback saying this was wrong.

ETA: I did ask it if GET works and it said no.

 

By the way, someone in my group tried ChatGPT and asked some questions about ME/CFS in Hungarian. Not just in general but also about how you can get good info about ME/CFS. And ChatGPT said my website is the Hungarian ME/CFS website, the one recommended for information about the disease in Hungarian. I mean I know the website has literally zero competition (just articles but not entire websites) but it is still great that despite it being young (I think it's not even two years old, more like 1 and a half), this is what ChatGPT recommends to people who would like more info. Thank you, ChatGPT!

(Google is a harder case but at least I'm on page 2 now there too, even without being active these days. For a long time it didn't even show me unless you used very specific words for search.)

 

ChatGPT is pushing generative AI forward, with powerful potential in health care settings

Asked to pull up a few studies on chronic fatigue syndrome from the past six months, ChatGPT responded with confidence, listing five scientific studies published in 2022 and 2023, complete with author, year and publication.

But there was a caveat: Every study the AI-driven chatbot cited was made up or could not be found online.

(...)

When signing up for the artificial intelligence tool, users are alerted that it has not been trained on information past 2021, a factor that could have led to the aforementioned errors.

So the same question was run again, refined to pull from studies published between 2015 and 2019. It still yielded studies that didn’t exist.​

(The article is not about ME/CFS though, it was just used as one example.)

 

Another important issue highlighted by Carl Bergstrom: Answers from one query seem to spill over into another. This was using the new Bing AI feature.
https://fediscience.org/@ct_bergstrom/110082371618305666

 

This doesn't entirely surprise me. I've been asking ChatGPT to help me find foods for my restricted diet and it has made up no end of non-existent products, with dead weblinks to their nonexistent product pages. Easily 40, straight off the bat. Not a single real item. I told it what it had done and it apologised for the confusion and said it was only a chatbot.​

 

The worst thing about it is how confident it sounds while it’s hallucinating made up things. I asked it to produce some fairly basic code in a niche area and it spat out nonexistent gibberish with extreme confidence.

 

I've only ever seen this analogy described by MDs as simply an explanation that patients buy because it's just ambiguous enough and they don't object to it too much because of how generic and pseudoscientific it is.

So it definitely got it right. It plays the same role as the "chemical imbalance" they had to introduce to have some bio in the mix for the patients who point out they have none of those issues. I'm sure we'll be hearing about how they never really meant it, it was simply a convenient explanation, some time in the future.

 

If I've understood correctly how GPs are trained in decision-making (at least in the UK) these days, it does sound rather as if they use a flowchart-type principle:

Patient presents with symptom X:
>
Check temperature. Is it normal? Y/N
>
If Yes, do A. If No, do B
>
and so on. So is there any reason why AI couldn't do the same?

 

I wanted to reply to this article, with examples from my professional life on the risk of inaccuracies getting through and corrupting the data. However, the site's rules require me not to use a pseudonym. So, I would have to reveal my true identity, plus what it is I do for a living, and much more beside, which this AI would then be able to scrape up off the internet and use as information about me. Not happenin' ...

 

Can this be waived, or? I am interested in what you have to say.