Digital humanities researchers often care about fine-grained similarity based on narrative elements like plot or tone, which don’t necessarily correlate with surface-level textual features. Can embedding models capture this? We study this in the context of fanfiction!

I love Zohran y’all, but posting “Proven: politics not about compromise and we never have to compromise again” because the Democratic nominee won the election for … mayor of NYC … might slightly overstate available evidence? Still, 💕 and hugs.

Aghion ce matin sur France culture : "Moi y a une chose qui me révolte, c'est la pauvreté et la misère." 🤡

I'm honestly not sure whether that's an insane or brilliant question

A l’occasion de la présidentielle samedi en Côte d’Ivoire, une semaine 🇨🇮 dans @culturesmonde.bsky.social 1. A. Ouattara, une réélection jouée d’avance 2. Désinformation, le Burkina Faso 🇧🇫 pointé du doigt 3. Cacao, une rente menacée 4. La musique 🎶 au pouvoir ➡️ www.radiofrance.fr/francecultur...

Monty Python understood p-hacking

Evolving thread with the best papers that I read. It will be sociology, economics, psychology and computational stuff (god I have to choose a discipline someday) Let's start with this absolute banger (thx @arnaud-wolff.bsky.social) : www.journals.uchicago.edu/doi/abs/10.1...

Beaucoup trop bien : www.radiofrance.fr/francecultur...

Question pour les vrais économistes par ici : pourquoi les AER et Econometrica sont systématiquement imbitables, alors que les articles sont souvent délicieux à lire dans Review of Economic Studies, Economic Inquiry ou Journal of Economic Perspectives?

"Aveuglé par tant de fourberie, je n'entends que les cris odieux du fond de mes poches" www.youtube.com/watch?v=b3RF...

It's been brewing for months: @inriaparisnlp.bsky.social releases CoMMA (Corpus of Multilingual Medieval Archives) ! 📚 2.5bn tokens of mostly Latin and French texts 🕰️ 800→1600 CE 📜 23k manuscripts 🖥️ 18k on the reading interface: comma.inria.fr 🔍 Paper: inria.hal.science/hal-05299220v1 (1/🧵)

Le macronisme c'est long, surtout vers la fin.

The last time I posted the income relationship to presidential vote among White respondents to the @electionstudies.bsky.social ANES, people asked for additional estimates among all voters. Updated estimates here:

A. Conchon, L. Feller, C. Virlouvet et moi-même y présenterons notre récente Histoire économique et sociale de la France. Venez nombreux et nombreuses

Un projet beaucoup trop cool

BTW the reason there aren’t actually any consistently right wing AI models really is more or less because the right wing is wrong and they aren’t going to be able to fix this overall.

Even in Denmark, cultural tastes are socially stratified. With some quite cool library data we find strong gradients by education and wealth - important as wealth is rarely studied in cultural stratification research doi.org/10.1093/esr/... OA: osf.io/ar2xc_v2/dow... @europeansocreview.bsky.social

Présentement en train de travailler sur mon macbook dans un café branché d'un quartier londonien gentrifié, à me demander à quel moment exactement je suis devenu un connard et dans quelle mesure c'est réversible

Ever wanted to read about an old problem almost nobody cares about anymore? Well, I wrote about it. 🧵

Un peu vertigineux de retrouver ma grand mère, sa fratrie et leurs 9 (!) domestiques. Et après on s'étonne que mon daron sache à peine faire des pâtes

Tellement bien ! Bravo et merci @sandrabree.bsky.social et ses collègues

Grâce au projet #POPP dirigé par @sandrabree.bsky.social (@larhra.ish-lyon.cnrs.fr ; @cnrs.fr ) et le laboratoire LITIS pour la partie informatique

Come work with us @crestsociology.bsky.social We are looking for a post-doc with a background in social sciences, interested in NLP and in media & journalism studies. Details here: www.css.cnrs.fr/post-doctora...

I think this miiight be the best feed for machine-learning news on Bluesky. It strikes a good balance between new/popular and whitelist/regex factors in the algorithm.

Super article www.mediapart.fr/journal/poli...

Close Netflix and open this: probably the best modeling paper I've ever read

Dans le n°5 de ma newsletter "Parlementer", je reviens sur les enjeux de la réélection à venir du bureau de l'Assemblée. Dans une Assemblée balkanisée, des conceptions du pouvoir et de la façon de gérer le RN, s'affrontent. pjanuel.kessel.media/posts/pst_78... (newsletter gratuite sur inscription)

#Cartographie | Découvrez la France par carreaux de 200 m de côté à l’aide de l’outil interactif Cartographie infracommunale (population, niveau de vie, pauvreté…). Les cartes proposées représentent des données issues du recensement de la population et d'autres sources 👉 www.insee.fr/fr/outil-int...

📣 Job alert! *Assistant Prof in Computational Social Science*. We're a friendly department, with sharp students, at a great institution, in a lovely city. We have real strengths in computational social science & are looking for a colleague to build this further. Share and reach out with quesions!

Le dernier rempart contre le Front national s'appelle Julian Alaphilippe

Fréquence de "stigmate(s)" dans les revues de socio de Persée et Cairn (avec mon petit Gallicagram)

🚨 New working paper alert 🚨 Missing summer — and the Tour de France? Don’t worry, we got you covered. 🚴‍♂️🚴‍♂️🚴‍♂️ In this paper, we show that being on the route of Tour de France reduces far-right voting. osf.io/preprints/so...

Delighted to announce the publication of 'From Questions to Knowledge', my new and updated statistics and data analysis handbook. www.danielnettle.eu/2025/09/14/f...

Ce soir Julian ALAPHILIPPE a gagné et ma timeline en a probablement rien à fiche mais moi je suis HEUREUX In other news, je déménage demain pour Londres pour commencer lundi mon postdoc à la LSE

Great question! Performance and LLM hacking risk are negatively correlated. So easy tasks do have lower risk. But even tasks with 96% F1 score showed up to 16% risk of wrong conclusions. Validation is important because high annotation performance doesn't guarantee correct conclusions.

A grim but great paper! One question comes to mind: what if the task is really easy? is the risk small enough for obvious tasks, e.g. classifying news articles between sport, politics, culture etc?

🚨 New paper alert 🚨 Using LLMs as data annotators, you can produce any scientific result you want. We call this **LLM Hacking**. Paper: arxiv.org/pdf/2509.08825

1/ Thrilled to share that my main PhD article has just been published in Nature Human Behaviour! 🎉 👉 “A randomised controlled trial on the effect of administrative burden and information costs on social inequalities in early childcare access in France” rdcu.be/eFAEA

My very first paper has just been published in Nature Human Behaviour! In it, we demonstrate that both a chatbot and a more traditional pedagogical intervention can help improve adolescents' knowledge of vaccines and their attitudes towards vaccination.

Je viens d'installer gemini-cli, je crois que ma vie vient de prendre un tournant

Do streaming platforms trap us in cultural filter bubbles? We like to think so but the evidence says otherwise. In a new paper @abelaussant.bsky.social and I find the use of streaming platform to be associated with an increase in consumption diversity. sociologicalscience.com/articles-v12...

We've just released an amazing Embedding model: EmbeddingGemma, the new best-in-class open embedding model! 🚀 🏆 Top multilingual model on MTEB (<500M) 💾 Runs on <200MB RAM ⚙️ Customizable output for on-device use 🧩 Integrated with your favorite tools developers.googleblog.com/en/introduci...

"It just predicts the next token based on data in the training set" is a poor abstraction of LLMs because it makes poor predictions For example, we might predict that it's very difficult to have an LLM emit the phrase "the quick brown fox jumps over the lazy cvnpmnzq", but it's trivial

Dans Le Monde, on abandonne progressivement la formule "le problème..." (le problème algérien, le problème social)

i wanted to understand the current landscape better, so i did a deep dive on current embedding sizes and architectures vickiboykis.com/2025/09/01/h...

Now that school is starting for lots of folks, it's time for a new release of Speech and Language Processing! Jim and I added all sorts of material for the August 2025 release! With slides to match! Check it out here: web.stanford.edu/~jurafsky/sl...

Not sure which is more clearly a millennial religion The belief that existing technologies are just the heralds and forerunners of “AGI” Or the belief that existing tech is a “bubble,” a fraudulent Babylon whose millions of deceived followers will melt away like snow when its falsehood is unmasked