Yes and no. Yes, you're right in that it does analyze info you post pretty well. No, in that it does have some ability to reference papers - althought it's very hit or miss, and not reliable. For instance, I've often asked for what's the research and science between two receptors / two metabolites and asked to include sources. But yeah, it could be way better. Can't tell if that's from its creators handicapping it and changing it with new version or just the nature of it.
Agreed. But I wasn't saying it has no ability to reference papers - just that LLMs do not 'remember' everything they've ever seen. People sometime seem to think that an AI remembers everything it's seen, but that's not how training works.
So if a paper is important and frequently mentioned, they may 'remember' it in much more detail. But that's my point - since references to specific papers are unreliable, that's an area it can't be used.
It will rarely get a receptor function wrong or be unable to explain the krebs cycle or be unfamiliar with a common drug, but it will frequently mix up specific papers if you ask about a 1956 study or something. Again, it's not dissimilar to a person. I will remember famous papers, but other ones I will vaguely remember.
I don't think it's creators handicapping it - but there are no 'sources' during inference - that's not really how it works. It's not looking up anything, it's using the weights in its 'neural net' to answer.
The negative side is that AI is often incorrect in its assumptions - meaning that if you find something statistically significant, you still have to dig through that to manually confirm the assessment.
AI is good to use to identify absolutes - those things that do almost nothing at all, are in contest (think like something that works about half the time), or work almost every time. Outside of that though, manual research is necessary. So it is about as useful as a very basic intern or an uneducated assistant. It can't replace an expert, a scientist, a professional, or even a hobbyist.
I am willing to bet any person in these forums could do a better job given the general patient level of knowledge about our condition.
I would respectfully disagree with most of that. AI is actually not great for certain types of absolutes (mathematics, etc), but can be really good for many areas. If you're using the tool in the right way, my 'success rate' is nowhere near 50% (ETA: Haha, as a faulty human, I misread what you were saying here. But still disagree - more useful than any intern or assistant I've ever had. It's like having a full time assistant with a Master's degree.).
Of course, manual research is ALWAYS a good idea, whether it's your doctor (especially) or a scientist or forum member.
But saying AI is often incorrect in its assumptions is usually cherry picking very specific areas that it's known to have problems. In my day-to-day, it's probably right more than 90% of the time for how I use it.
Even in areas where it struggles (like math), the other day I asked it to calculate the electrolytes and sugar in 8oz of V8 Original and 8oz of Gatorade Lemon Lime. This is an area it usually is not good (looking up numbers), but after a quick mental math check of nutrition facts, it looked very accurate.
For instance, I asked AI a very general question:
Is exercise a good treatment for symptoms of major fatigue and 'crashes' after activity?
I think maybe 100% of the 50+ doctors I've seen would get this wrong unless there were further context - like when I already understood how badly exercise affected me and included that in my health history. This is the beginning of GP4o:
Exercise can be a beneficial treatment for symptoms of major fatigue and "crashes" after activity, but it depends on the underlying cause of these symptoms. Here are some considerations:
- Chronic Fatigue Syndrome (CFS) / Myalgic Encephalomyelitis (ME):
- For people with CFS/ME, exercise can be a double-edged sword. Graded Exercise Therapy (GET) was once recommended, but many patients found it exacerbated their symptoms. Now, a more tailored and gentle approach, known as pacing, is often advised. Pacing involves balancing activity and rest to avoid post-exertional malaise (PEM).
- General Fatigue:
- Regular, moderate exercise can improve energy levels, reduce fatigue, and improve sleep quality. Activities like walking, swimming, and yoga are often recommended.
- Mental Health:
- Exercise has been shown to improve mood and energy levels, which can be beneficial for individuals experiencing fatigue related to depression or anxiety.
I always warn people to be careful about AI like they would about their doctor or any Reddit post or whatever. But it's not been my personal experience that it's incredibly unreliable. The problem is when you do ask the type of questions it can't answer, it may just make up numbers. So you need to learn the types of questions it can't answer, but beyond that I think it's much better than most people realize.
