Small but time-aware!

At the " #LLM for for the History, Philosophy, and Sociology of Science workshop" @tuberlin Jochen Büttner presented a compact small language model (39M parameters) that integrates a temporal dimension for historical text analysis.

A proof of concept using historical UK weather reports shows the model successfully reflects both seasonal patterns and linguistic shifts

Wanna try it?
github.com/j-buettner/time_transformer

I asked #ChatGPT to write a Word Sense Disambiguation system using the #ViterbiAlgorthm and code reviewed it's results.

https://petebleackley.github.io/code_reviewing_chatgpt.html

#ArtificialIntelligence #NLP #LLM
@data_science

We've got a new article out

*Angst M, Müller NN, Walker V. Automated extraction of discourse networks from large volumes of media data* (open access)

https://doi.org/10.1017/nws.2025.4

As part of our sustainability.discourses project, we developed an automated method to gather and track discourse networks around urban sustainable mobility, based on media data.

The article describes the approach in detail and reports an evaluation of the internal validity (do we measure what we ourselves think we measure) of the pipeline.

tl;dnr: Automating discourse network data gathering can work well to capture broader tendencies. In many cases, it is likely not worth investing in automating data gathering to understand discourse however.

Want to see the daily updated pipeline in action? Check out the web apps and resources here: https://sustainability.discourses.ch/en/

Classifying Genre in Historical Medical Periodicals

Next in line: Vera Danilova presents her work on genre classification in digitized periodicals from European patient organizations (1951–1990) using #LLMs as part of the #ActDisease project.

XLM-RoBERTa (UDM) led Q&A tasks with 32% more correct answers than mBERT/hmBERT.
hmBERT (UDM) topped Administrative classification (+16%)
CORE-based models excelled in legal genre prediction.

#DigitalHumanities @tuberlin #classification #NLP

Call for participation: *ClimateCheck* Shared Task ( https://sdproc.org/2025/climatecheck.html)

@NFDI4DS members Raia Abu Ahmad, Aida Usmanova and Georg Rehm are organizing a shared task “Scientific Fact-Checking of Social Media Claims on Climate Change (ClimateCheck)” on July 31 or August 1st, 2025 in Vienna, Austria, hosted as part of the SDP 2025 Workshop.

Deadline for system submissions: May 16, 2025

#NLP
#LLM
#climatecheck
#socialmediaposts
#climatechange
#misinformation
#SDP2025
#ACL2025
#Vienna
#NFDI4DS

Just caught myself using the phrase "#Inform7 source code if Inform 7 was made for MCAD". What in the world happened to #NLP research? It would be so much more useful than what we have now, and wouldn't burn down the planet to make a database query.

On Tuesday, I gave a lecture on harms of AI and NLP. Turns out, the ChatGPT update the same day started a viral wave of Ghibli style image gen.

Worth seeing Miyazaki's response if you haven't. It is absolutely brutal: https://youtu.be/ngZ0K3lWKRc

We discussed: Can generative models be used to make art? What is the role of intent? Why are we using gen AI models and should we?

https://techcrunch.com/2025/03/26/openais-viral-studio-ghibli-moment-highlights-ai-copyright-concerns/
https://www.404media.co/email/5d418149-ed10-42f3-b535-c74d61ed5b6c/

Dami Lee: https://youtu.be/FRnKY-5TqOM

Call for participation: 2nd LLMs4Subjects Shared Task (https://sites.google.com/view/llms4subjects-germeval/)

The challenge is part of the German Evaluation (GermEval 2025) on 10. - 12. September, 2025 in Hildesheim, Germany, co-located with KONVENS 2025.

Release of training/ testing data: March 8/ May 23, 2025
Deadline for system submissions: June 2, 2025

#NLP
#LLM
#SharedTask
#Challenge
#LMMs4Subjects
#Classification
#energyefficiency
#KONVENS2025
#Hildesheim
#NFDI4DS

Greetings from the FIZ Karlsruhe Research Retreat at Königstuhl, above the city of Heidelberg. I had the pleasure to introduce and conduct a Design Thinking session together with 40 brilliant colleagues to work on how to improve our research

@fiz_karlsruhe @fizise @zbMATH @tabea @GenAsefa @heikef @enorouzi @sourisnumerique #semanticweb #knowledgegraphs #AI #llms #nlp #innovation #designthinking

An extra-fun Acuitas update for March, as I go back to work on game-playing. I'm trying to set him up to play a text version of a favorite game that helped *me* learn how to think. https://writerofminds.blogspot.com/2025/03/acuitas-diary-82-march-2025.html
#AI #ArtificialIntelligence #chatbots #NLP

Mobile email UIs added AI replies "on top". These popups hide draft & email, so users need to remember all info from the email while prompting & checking AI results. We've redesigned this with local responses, resulting in flexible workflows - both with & without AI. #CHI2025 #HCI #NLP #NLProc #AI

** MCP: DAS ENDE DER KI-HALLUZINATIONEN? **

Ist das Model Context Protocol (MCP) wirklich der Game-Changer, auf den wir gewartet haben? Ein neuer Ansatz, der KI-Halluzinationen minimieren soll – klingt fast zu gut, um wahr zu sein!

#ai #ki #artificialintelligence #NLP #MachineLearning #LLM #DeepLearning

https://kinews24.de/mcp-das-model-context-protocol/