The other day I was rather stuck on a search.

So I thought I’d try Copilot to see what it would throw up (note if you use Copilot, choose either the Creative or (more likely) the Precise option, as those two give you access to the more powerful ChatGPT4. The default middle option (Balanced) only provides access to version 3.5)

I put in the query as a natural language query – as all good medical librarians are taught to do – and it threw up 5 citations. None were hallucinatory (as visited each one), one was very relevant (which was then used as a seed article to find related articles), three others weren’t too bad, and one wasn’t useful. So, in this instance it was surprisingly helpful. Having said that, I tried it on another search today and the results weren’t useful, so it can bit a bit hit and miss, and probably depends how the question is phrased. In any case, another option to keep in mind

Copilot can be accessed at the top right of the Edge browser or directly at https://copilot.microsoft.com/

Browser Extensions For The Stretched Health Librarian, 2023
https://www.johila.org/index.php/Johila/article/view/143/212

What amuses me most is the fact that it significantly increases the already critical importance of asking the right questions in the “reference interview” of ChatGPT in order to make sure we elicit the most useful information in order to be able to provide the actual human patron with the answer they actually need! Thanks bookoodles for all your work, Michelle!

Yes. IMHO I think it depends on how much time you save. Believe it or not both the first post and the second post I wrote took me less than an hour. My posts used to take me a lot longer than that to do. It definitely helped save me time when I was writing a professional and technical email about library stuff to an important non-librarian. When writing those types of emails I spend a lot of time de-jargoning the library terminology so as to get my point accros in a professional but understandable way. I used ChatGPT to write the email. The email was entirely too long but I cut and pasted parts of it when I crafted my own and it made the process pretty quick and easy.
Right now in its current iteration, I wouldn’t spend the time on ChatGPT for medical searching. BUT be wary when we talk about amount of time to get results because there are ton of MEDLINE searches that I have spent hours on within PubMed or Ovid MEDLINE trying to get good results. I think time spent vs output received is important but not the only marker.
Some have emailed me to look at Elicit, Lori, Perplexity as AI’s to think about for search PubMed. One person said they used http://www.perplexity.ai on medical topics and it reference some articles from PubMed. It still has paywall problems too but those are ones to try and see if input time vs output results is worth it.

After an hour of using every tool at my disposal, I finally asked the patron if they had used ChatGPT while searching and when they said yes, I said, “Well, THERE’S your problem!”

I would not use it right now for that kind of research because it might given an answer but ChatGPT3.5 will not always provide correct references and it is difficult to say where found the information it displayed. Right now I would do that kind of search in PubMed or other appropriate bibliographic databases. Doctors, nurses, medical professionals should not be using ChatGPT3.5 to find literature to answer medical queries. The risk of AI hallucination, lack of transparency regarding where it found the answers, its inability to go behind fire walls for published journal articles, books, etc., and its date limit of 2021 really hamper it for finding that type of information.