#medlibs #canmedlibs
A devastating loss for the librarian community
05.02.2026 01:17
👍 6
🔁 1
💬 1
📌 0
#medlibs #canmedlibs
A devastating loss for the librarian community
first conclusion. It runs a search then evaluates the results likely with Gemini 3? At certain points you can make it go deeper, but it seems to always stop when it has found 50 relevant results OR looked at the top 300 results! (1).
Hey #medlibs, I built a tool for making it easier to create proximity search strings in PubMed. Check it out!
I presented at PNC-MLA this afternoon and will follow up with a link to the recording when it's posted.
[blogged] “We’re Good at Search”… Just Not the Kind That the AI era Demands - a Provocation aarontay.substack.com/p/were-good-...
Your annual #medlibs Halloween reminder that "Chocolate"[Mesh] has only been in use since 2017. (OK, to be fair previous indexing is "Cacao"[Mesh] 1963-2016, but STILL.)
Also, "Spirit Possession"[Mesh] is a thing, with only 28 results in PubMed, "Spirit Possession"[tiab] gets 144 results.
🎃🍫🍬👻
“Generative AI / large language models have added to this flood by making papers – especially papers not introducing new research results – fast and easy to write. While categories across arXiv have all seen a major increase in submissions, it’s particularly pronounced in arXiv’s CS category‘’
ResearchRabbit big updates www.youtube.com/watch?v=St4C... - first impression much better interface , a lot of the new features are similar to LitMaps (there is a partnership). Sadly some features are now under paywall (advanced search) unlike researchrabbit in the past. (1)
At a fancy-pants conference where everyone's a professor and let me tell you the timetable is not holding up well.
I was working on a slide today about "searching is like using a flashlight in a dark cave." Coincidentally, Caulfield also referenced caves, but in the context of AI mode being like having a drone map the cave for you! ...can I be forgiven for thinking of it as "generation-augmented retrieval?"😊
Mike Caulfield's SMU Masterclass on AI mode definitely evolved my mental model of genAI with search.... We used to have to make sense of our search results by reviewing hits individually. But now suddenly we can ask search results to organize *themselves* into meaningful structures, like tables.
This is a good tip, I have never even heard of ECIL but this looks like a conference I wish I had been at. I am going to need to start running similar product trials soon and would be nice to benefit from others' work!
For sure, if it were me, I would rather the vendors get a front-file data load and do all their pre-processing. Otherwise all the tools lose their retrieval special sauce. Consensus has all those fancy filters for example that you can apply before even running the search.
Aha, I forgot to think about MCP. If publishers start offering full text via agents it doesn't give any one product the clear content advantage... and allows the vendors in this class of product to continue duking it out. Wish I was going to be able to see this panel!
Consensus is making a very aggressive push for market share right now. I wonder if their ultimate goal is to be purchased by a publisher-agnostic library vendor like Clarivate.
I've just been revisiting your blog post on abstracts as the "petrol tank for AI discovery" and I wonder how long these tools can continue to exist before publishers find a way to collect licensing fees for their front-file abstracts - and in that scenario, which products will survive.
If Google is supposed to know so much about me why does it always suggest the Modern Language Association conference website when I Google "MLA annual." 😂 At least let me benefit from my loss of privacy, lol!
Rainbows and renewables in the Columbia gorge 😊
Just now seeing OpenAI's research that 80% of casual AI use is information-seeking. It doesn't surprise me but I think it really highlights that information literacy is a major component of AI literacy, and libraries have a role to play in AI literacy instruction. cdn.openai.com/pdf/a253471f...
But if it helps you evaluate and select which system your library should invest in… that’s valuable knowledge!! Eg if I need a reason we need to pay for e.g. Scite vs the corny AI implementations from traditional library vendors you were mentioning earlier… I think this kind of thing helps for sure!
I don’t mean to project my own limitations onto the profession but I wonder why we aren’t talking about this more
Are we focusing on these features as a profession because most of us didn’t get into the technical side of information search and retrieval in library school? When I was in school semantic search barely existed. But I don’t feel like I see a lot of presentations on this topic at library conferences
That surprises me, does that require them to maintain both a lexical index and a vector embedding database separately and update them both whenever they get new data loads?
I was able to meet with a Consensus rep and was interested to hear how they do retrieval by both keyword and embedding search. I wish I could’ve learned more but I’m sure after a point what they do becomes proprietary but they seem to have done some thoughtful development of the info retrieval part
[Blogged] Implications of AI powered academic search open.substack.com/pub/aarontay...
Number 2 seems to be the most important one, and it seems like everyone forgets to look at the sources! They forget especially that the generated text is NOT an "authored" text.
Just saw this announcement that NNLM's Region 5 grant funding has been reinstated?!? #Medlibs news.nnlm.gov/region_5/r5r...
Humans win! 🥇 #MLANET25 #medlibs
Dang I missed this! Whose presentation was this if you don’t mind me asking?
RFK Jr. ordered the NIH’s Integrated Research Facility in Frederick to stop all research on serious infectious diseases yesterday at 5:00 PM.
They study pathogens like Ebola there. You may be familiar with it from The Hot Zone.
www.wired.com/story/hhs-ni...
