I saw this today, new LLMs pretend to have inner thoughts.... not sure what to think about it. 🤔 I find them already too talkative in general. 😅

Yes I had immediately the same question 🤣 I'm feeling stupid 😅🫠

Aujourd'hui de 14h à 16h, je fais un tuto mate-shs : les LLMs pour les nuls. Soyez les bienvenus ! mate-shs.cnrs.fr/actions/tuto...

I just asked Claude about me and it said I am a VERY obscure person. 🥲

Then, it asked whether I knew her, I said 'Yes, I am her.' It replied 'Oh, I apologize Dr. Seminck!'. 😂😅

The Green AI project at University of A Coruña (Spain) has a PhD position open! Your work will be on efficient NLP. Contact me if interested!

Sorry, "relatively obscure". Already better than "fairly obscure " 😅

Fairly obscure 🤣😂

I love the fact you can plant them outside and they come back every year.

I was wondering whether performance could be enhanced working on very explicit promting. You ask again and again to do a better job and explain why the output is wrong, going on until you get something acceptable. Then, repeat 100x and save these chats, use them as training data. Could that work?

Relaying a message on #esu2025 which will take place in #Besançon!

The 15th European Summer University in Digital Humanities (ESU-2025) will take place from July 22nd to August 1st, 2025, in Besançon, France.

The local host will be Université de Franche-Comté, which will soon be renamed […]

Oh great, happy to hear that. Merry Christmas by the way! 🎅
And fore anyone who fancies some more 🥐🥐 m.youtube.com/shorts/oBJ3y...

Dear Ted, this reminds me of an article I wrote some years ago. We developed a so-called 'reference corpus' which is available in the paper's materials, but I can also send it to you by email (all in French though 🥐) . culturalanalytics.org/article/3758...

Facinating... I put this publication on my list to read.

Revised conclusion after input on thread from @oseminck.bsky.social, @mariaa.bsky.social and @dbamman.bsky.social: there is probably no consensus yet, because degrees of memorization vary widely and their effects on downstream tasks are even more variable, depending on the task!

It is already a week since Aarhus University welcomed #CHR2024. We look back on inspiring days of talks and research!

Next year, the C²DH at the University of Luxembourg will host us. We hope to see many of you there! 🇱🇺

2024.computational-humanities-research.org/announcement...

And the runner ups of the Best Short Paper Award:

🎖"Abbreviation Application: A Stylochronometric Study of Abbreviations in the Oeuvre of Herne’s Speculum Scribe" by Caroline Vandyck & @mikekestemont.bsky.social

🎖"SCIENCE IS EXPLORATION" by @rmmhicke.bsky.social & Ross Deans Kristensen-McLachlan

Our congratulations to the winners of the #CHR2024 Best Short Paper Award 🏆

Automated Image Color Mapping for a Historic Photographic Collection

by Taylor Arnold & @nolauren.bsky.social

2024.computational-humanities-research.org/announcement...

Yes I agree, it depends on the research question. I think that researchers should be aware that a little number of books shows signs of a high degree of verbatim memorization. It might be useful to exclude them from experiments to avoid bias, depending on what you're doing exactly.

And now the big announcement... #CHR2025 will be in Luxembourg! 🇱🇺

That's all for now, signing out from #CHR2024 🫡

Thank you to the @comphumresearch.bsky.social for the #chr2024 Best Short Paper recognition. Taylor and I are deeply appreciative, and grateful to be a part of this wonderful community!

🚀📢 BookNLP-fr is here!

A major step for french computational literary studies by the Lattice Lab team—Frédérique, Olga, Martial, Clément, @tpoibeau.bsky.social and I!

🌍 The French expansion of @dbamman.bsky.social’s BookNLP project

📖 doi.org/10.48694/jcl...

#ComputationalHumanities #NLP

@jbarre.bsky.social hello!