@aesthr LLMs are designed to keep you using them over and over in a repeated loop. One of the ways they accomplish this is by agreeing with you and affirming you, regardless of what you say.

Even if you feed it nonsense, its bias is to keep agreeing with you. It can and will tell you you are right about something, when you are very wrong.

Do not use AI for any safety critical work. Do not use it to diagnose or treat any medical condition.

@aesthr maybe tell them that AI just writes the most likely word and does not check its sources, and while it can tell something plausible, it can also invent things
AI have been coded to please you so you use them longer, so they may wrongly tell you you're safe or make you panic, and may lack nuance. In complex medicine cases the truth can often bring messy feelings, and seeking human connexion can help figure things out. If AI is making you certain (bas phrasing i know), it might be a sign it's rrying to please you rather than actually help.
While humans may have bias, for example against black women, the way AI is trained tends to amplify this issue and is not foolproof. (also remember that while doctors may have biases, they have studied medicine for xx years and good doctors do exist. If one doctor is overly biased you can ask peers for names of doctors that are not.)
While seeking a second opinion after seeing a doctor can sometimes be a good idea, you can also see a second doctor, advice from a group of peers, or seek info about specific conditions on [insert trusted websites]

from a chronically ill patient used to looking for info and gently suggesting it to doctors so they might consider it and test it without feeling offended.

@aesthr
– Before using an LLM, always ask yourself "Is this necessary and can't this question be answered by other sources?" Every prompt burns tokens, that consume energy which is bad for the environment and accelerates climate change
– LLMs don't have a concept of facts. When asking for factual information, always ask it for attribution. Don't take attribution at face value. Check it yourself if you can and if not ask someone who can to check it for you

1/

@aesthr

I think one very important thing is to reference trustworthy sources to double check information from a chatbot. Most of all a physician.

Another thing is to emphasize that LLMs can give wrong plausible sounding answers with the appearance of confidence.

Patients are especially vulnerable because they don't necessarily have the expertise to easily check chatbot output.

@aesthr While teaching at uni, I’ve seen what happens to a lot of inexperienced yet supposedly smart people when using LLMs specifically. I’ve found that a lot of people have a poor understanding of what LLMs are good at and what not, and also how to critically evaluate LLM output. I think many people just fall for the convincing-looking text that comes out of an LLM and really don’t expect the true depth of falsification, gaslighting, and making shit up that LLMs can do if you’re not careful.

@aesthr I have written that kind of thing but mostly just for professional contexts (as in: for the designers and developers at the firm I work for). That probably doesn't help much