Aaron Tay @aarontay.bsky.social · 11hLooking at the new pilot - Linked Discoveries in Pubmed linkeddiscoveries.ncbi.nlm.nih.gov/35472127/ - I used a paper I am fairly familar with as seed pubmed.ncbi.nlm.nih.gov/35472127/ /1 132
Reposted by Aaron TayAltmetric @altmetric.com · 28/09/2026Papers discussed in @cassyni.bsky.social talks will now receive a seminar mention on their Altmetric badges. A seminar mention will carry the same weight as a policy document or clinical guideline.itbrief.com.auAltmetric adds Cassyni seminars to research scoresResearch talks will now feed into Altmetric Attention Scores, giving universities and publishers a broader view of how scholarly work is discussed. 163
Aaron Tay @aarontay.bsky.social · 28/09/2026Im only on the 20usd tier of openai and anthropic which is limited. But combo of Opus5.5/medium + Gpt6-luna high effort really goes very far, for first time I can run multiple subagents with little problems. Unimaginable how things will be a year from now 130
Aaron Tay @aarontay.bsky.social · 26/09/2026A bit weird to see my workshops on ai powered search and agentic flows classed under "ai literacy" 130
Aaron Tay @aarontay.bsky.social · 24/09/2026Finally had time to try the new gen Opus 5.5, Sol/luna 6. Crazy how cheap they are 000
Aaron Tay @aarontay.bsky.social · 24/09/2026Compared against CLEF-TAR 8 DTA used in Akinseloyin 2024/2026. It is slightly worse at WSS95 & better at WSS100 using essemble of Claude 3 + Gemini reranking. 2026 LLM esemble = GPT-4o Mini + Gemini 1.5 Flash + Claude 3 Haiku, with embedding reranking - not bad at all /1 110
Aaron Tay @aarontay.bsky.social · 24/09/2026Just did first workshop internally on agents (claude code/codex). Scary but interesting experience when doing brand new content. Really hard to decide what level to pitch at and examples to show because audience was mix of admin staff, students and even faculty. On hindsight should have restricted 000
Aaron Tay @aarontay.bsky.social · 24/09/2026Playing with jev as title-abstract screener. Achieves WSS of 42% with recall of 100% in my first test /1. Not bad.. But not SOTA I think... cost is 0.00020 120
Aaron Tay @aarontay.bsky.social · 18/09/2026Singapore is well known for having very liberal laws for text mining/ai training. But if these recommendations go through its going to get even more so /1 sal.org.sg/wp-content/u...sal.org.sg 101
Aaron Tay @aarontay.bsky.social · 18/09/2026Hmm a possible breakthrough in retrieval if Jev this is truly sota reranker & much cheaper/faster. I wonder if this works well for title abstract screening for evidence synthesis 110
Aaron Tay @aarontay.bsky.social · 18/09/2026Example of what most people actually want/need. Some systems that spits out probabilities from a fixed set of alternatives + predefined flows. typesafe.ai/blog/introdu...typesafe.aiIntroducing System One Models & Jev - TypeSafe AI BlogTypeSafe AI is an AI lab building machine-native intelligence infrastructure for automation, designed to make decisions within software. Try our first System One Model, Jev, in early access. 130
Aaron Tay @aarontay.bsky.social · 17/09/2026I realise a lot of what librarians want is not agents but what anthropic calls "agentic workflows" or "sys where LLMs & tools are orchestrated through predefined code paths" vs "sys where LLMs dynamically direct their own processes , tool usage maintaining control over how they accomplish tasks" /1 330
Aaron Tay @aarontay.bsky.social · 12/09/2026Not sure how good the categories stand up but the q is still important. Libraries with a-z databases now have "ai" tags. What does that mean? "Ai" used in =retrieval eg generate boolean, query expansion? Dense embedding? Ai used to summarise individual items? Synthesise answer across multi docs? 020
Aaron Tay @aarontay.bsky.social · 11/09/2026I must say the RAISE statement from evidence synthesis organisations (pleasantly) surprises me. It reads cautiously optimistic to me on AI use. Given how careful and rigorous such standards are, this is not a small thing... 011
Aaron Tay @aarontay.bsky.social · 11/09/2026What Changes When an LLM agent Searches Your Library Catalogue? aarontay.substack.com/p/what-chang...aarontay.substack.comWhat Changes When an LLM agent Searches Your Library Catalogue?I connected an LLM to search Primo and found it was excellent at academic database discovery to help less experienced users 052
Aaron Tay @aarontay.bsky.social · 07/09/2026I suspect the best leading indicator for when AI really starts transforming work isn't benchmarks. It's when people who think AI is only about prompting, dont understand harnesses , suddenly start routinely delegating work to Gemini/Copilot because Google/Microsoft put it everywhere /1 230
Aaron Tay @aarontay.bsky.social · 07/09/2026[Blogged] AI Search capabilities in Scopus are expanding aarontay.substack.com/p/ai-search-... - Scopus added two ai search feature a) AI query builder (converts query to nested boolean) & b) Semantic search(uses vector embedding), but which is more reproducible? /1 151
Aaron Tay @aarontay.bsky.social · 06/09/2026"SMU rewrote its promotion and tenure framework. The new one, dubbed RISE — research, impact, service, education — treats impact as broader purpose, underpinning the other three, accepts policy papers and other grey lit as evidence alongside journal publl." sponsored.chronicle.com/smu-is-build...sponsored.chronicle.comBeyond the language of impact: How SMU is building the systems behind its promise (COPY)What Travels (and What Doesn’t) 110
Aaron Tay @aarontay.bsky.social · 04/09/2026How should LibGuides Describe "Those people" . - This might be controversial... hxlibraries.substack.com/p/how-should...hxlibraries.substack.comHow Should LibGuides Describe “Those People”?On designing information literacy tools regarding controversial topics while adhering to first principles. 130
Aaron Tay @aarontay.bsky.social · 03/09/2026Playing with the new JSTOR AI dynamic search about.jstor.org/blog/ai-is-c... /1 050
Aaron Tay @aarontay.bsky.social · 03/09/2026Hmm Webmcp. Something for acacemic library sites? developer.chrome.com/docs/ai/webmcpdeveloper.chrome.comWebMCP | AI on Chrome | Chrome for DevelopersWebMCP has two APIs that allow browser agents to take action on behalf of the user. 011
Aaron Tay @aarontay.bsky.social · 01/09/2026Slowly reading 2024 to 2025 stuff by anthropic on agents eg context engineering, harness engineering etc. I mostly understood these concepts but it feels more real now I have some experience messing around with codex and Claude code /1 100
Aaron Tay @aarontay.bsky.social · 01/09/2026[Blogged] AI Search capabilities in Scopus are expanding - From LLM-to-Boolean conversion to vector retrieval, filters and query interpretation aarontay.substack.com/p/ai-search-...aarontay.substack.comAI Search capabilities in Scopus are expandingFrom LLM-to-Boolean conversion to vector retrieval, filters and query interpretation 131
Reposted by Aaron TayEric Schares @eschares.bsky.social · 31/08/2026This analysis combined OpenAlex journal-year article counts with an accompanying APC dataset that's free and open for your reuse. Tracking 2,540 unique journals across 14 publishers and seven years, for 69,856 journal-year combinations. doi.org/10.7910/DVN/...doi.org 2135
Aaron Tay @aarontay.bsky.social · 29/08/2026[Read] Open Access Phantom citations: An empirical study of non-existent and unverifiable references in scholarly literature asistdl.onlinelibrary.wiley.com/doi/10.1002/... - pretty standard "I ran my home brew verifier over a lot of papers since 2023 and found x% of papers don't seem to exist" /1asistdl.onlinelibrary.wiley.com<em>Journal of the Association for Information Science and Technology</em> | Wiley Online LibraryThe integrity of scholarly communication depends critically on the accuracy and verifiability of cited references. Citations enable readers to trace prior work, assess evidence, and situate new contr... 173
Aaron Tay @aarontay.bsky.social · 29/08/2026Finally read the proposed Open Knowledge Format (OKF) and it's not that hard to grasp if you have any experience playing with LLM agents /1 100
Aaron Tay @aarontay.bsky.social · 29/08/2026[Read] Information Austerity Threatens the Future of Library Work. Let’s Teach Future Librarians to Fight Back. katinamagazine.org/content/arti... some thoughts /1katinamagazine.orgInformation Austerity Threatens the Future of Library Work. Let’s Teach Future Librarians to Fight Back.The information infrastructure on which democratic life depends is being dismantled. We must teach library and information science students not only how to function within degraded conditions, but how... 110
Aaron Tay @aarontay.bsky.social · 29/08/2026Everyone "supports open access". Question is what price point :) 141
Aaron Tay @aarontay.bsky.social · 29/08/2026[Read] Auditing GenAI Literature Search Workflows: A Replicable Protocol for Traceable, Accountable Retrieval in Student-Facing Inquiry www.mdpi.com/3042-8130/2/... - lot of work done but I see a lot of flaws /1mdpi.com 163
Aaron Tay @aarontay.bsky.social · 27/08/2026I have the impression the way We support apc payment in my institution is pretty unusual maybe even unique but im no expert on library TA deals, open access etc 100
Aaron Tay @aarontay.bsky.social · 26/08/2026Looking at 2 new Scopus "AI" search features. #1 AI query builder - converts your natural language query to Boolean. It might be using LLMs to do this but help doesnt say specifically. You get 5 queries a month, 50 if you have Scopus AI /1 110
Aaron Tay @aarontay.bsky.social · 26/08/2026Coming up for air after studying implementation of guard rails using hooks in Claude code/codex. Really fascinating... 140
Aaron Tay @aarontay.bsky.social · 26/08/2026Amused. Adding the word "agentic" to library workshops = Huge increase in demand /1 121
Aaron Tay @aarontay.bsky.social · 26/08/2026[Blogged] Prompt Engineering Is the Wrong Mental Model for AI Agents aarontay.substack.com/p/prompt-eng...aarontay.substack.comPrompt Engineering Is the Wrong Mental Model for AI AgentsWhat building an absurdly complicated PubMed search agent taught me about tools, deterministic checks and the importance of deep domain expertise 240
Reposted by Aaron TayCharleston Hub @chashub.bsky.social · 24/08/2026@aarontay.bsky.social guides us through the ongoing shakeup of the academic search market by AI with this comprehensive analysis of Web of Science Smart Search, Google Scholar’s Scholar Labs, and Consensus. In Katina Magazine: katinamagazine.org/content/arti...katinamagazine.orgHow Is AI Transforming Academic Search?Aaron Tay guides us through the ongoing shakeup of the academic search market by AI with this comprehensive analysis of Web of Science Smart Search, Google Scholar’s Scholar Labs, and Consensus. 154
Aaron Tay @aarontay.bsky.social · 24/08/2026Studying news source in altmetric. The most interesting bit is besides the usual tracking with direct links/identifers, it does text mining. This is important because many news stories dont allow links. /1 110
Aaron Tay @aarontay.bsky.social · 24/08/2026Forcing myself to teach a workshop on altmetric.com really forces me too go deep and try to anticipate what people will ask. I'm probably overdoing it but as a librarian i like to go deep into things I'm studyingaltmetric.comHomepageAltmetric tracks and analyzes online attention markers to published research to see where research is making a difference. 030
Aaron Tay @aarontay.bsky.social · 24/08/2026More exploration of altmetric.com. Institutional edition allows us to upload a list of research outputs belonging to our institution. This allows you to filter to it in advanced search or browse. Main use of this I think is so you can filter/browse by depts/schools lower than institution level /1 141
Aaron Tay @aarontay.bsky.social · 23/08/2026[Read] AI Skepticism Makes Moral and Practical Sense. We Should Still Move Full Speed Ahead. - this one almost seems designed to be misunderstood. But the core message is to engage actively AND be extremely skeptical at interrogating it /1 katinamagazine.org/content/arti...katinamagazine.orgAI Skepticism Makes Moral and Practical Sense. We Should Still Move Full Speed Ahead.When it comes to AI in academic libraries, we’ve organized ourselves into a polarity: skepticism vs. enthusiasm. Our professional future depends on our figuring out not how to resolve that tension, bu... 162
Aaron Tay @aarontay.bsky.social · 21/08/2026Going back and forth with LLM and trying to verify, this is the best I can do listing the various @altmetric.com sources and how they are picked up. the LLMs GPT-sol and Opus-5 also found quite a few contradictions in the support documentations found.... 321
Aaron Tay @aarontay.bsky.social · 21/08/2026Using chatgpt to try to grasp this. It found a contradiction "Current documentation is inconsistent. The dedicated Reddit page, updated 31 Mar 2026, says Altmetric tracks posts & comments from all subreddits. Other documentation still says original posts only." help.altmetric.com/en/articles/... /1 100
Aaron Tay @aarontay.bsky.social · 21/08/2026[Read] NISO RP-8-202x, Journal Article Versions (draft for public comment groups.niso.org/higherlogic/... - fascinating suggestions. Even if approved will the industry adopt it?groups.niso.org Projects - NISO 011
Aaron Tay @aarontay.bsky.social · 19/08/2026I think articles that claim librarians or information professionals have an advantage in the ai world cause searching = prompt engineering are wrong. We have an advantage but not the way you think /1 journals.sagepub.com/doi/10.1177/...journals.sagepub.com 140
Aaron Tay @aarontay.bsky.social · 19/08/2026Now that we have institutional access to altmetric.com, I'm trying to really get a deep understanding of it and anticipate what faculty will ask /1 140
Aaron Tay @aarontay.bsky.social · 19/08/2026"The danger is taxonomy by intuition: take familiar words, arrange them into a plausible hierarchy, and then reason from the categories as though they were discovered rather than constructed." - hmm 030
Aaron Tay @aarontay.bsky.social · 15/08/2026Of course I can see someone obsessively trying all the new ai search tools maybe getting a good feel of each tools pros and cons but even then it is for their use case and domain only which may not match yours.. 111
Aaron Tay @aarontay.bsky.social · 15/08/2026Whenever I see a workshop whether by librarians or researcher claim they can teach you the “right” way to use AI for literature reviews, I get wary. Literature review practice is wildly heterogeneous across disciplines & purposes- there is no one correct way even if your workflow has branches /1 141
Aaron Tay @aarontay.bsky.social · 15/08/2026[Blogged] Hot take: Don't brainstorm keywords with LLMs. Let them Search.https://aarontay.substack.com/p/hot-take-dont-brainstorm-keywords 030
Aaron Tay @aarontay.bsky.social · 13/08/2026i help researchers navigate information abundance by supporting scholarly sense-making: finding relevant evidence, identifying influential ideas, mapping research landscapes, understanding expert consensus interpreting impact signals & integrating AI-powered discovery tools into the research process 100
Aaron Tay @aarontay.bsky.social · 13/08/2026Hot take: library information literacy is a bit of a Frankenstein's monster. Many information literacy programmes now include database searching, source evaluation, verification /handling disinfo, prompt engineering/AI ethics/AI literacy & critical thinking. We are good at absorbing things /1 160