For today's reading group, Lasse Jantsch presented "Activation Oracles: Training and Evaluating LLMs as General-Purpose Activation Explainers" by Adam Karvonen et al. (2026) Paper: arxiv.org/abs/2512.15674 #NLProc

Spending the week in Paris at #IASEAI 2026, joining colleagues for conversations on the present and future of safe and ethical AI.

We were thrilled to host @mtutek.bsky.social at our lab last week. His talk "From Internals to Integrity: How Insights into Transformer LMs Improve Safety, Interpretability, and Explanation Faithfulness" led to great discussions! 👏 #Transformers #AISafety #ExplainableAI #MLResearch #NLProc

Call for abstracts! (Deadline: 17 March) Submissions are now open for the second edition of Towards a Safer Web for Women Workshop, taking place on 26 May at the Web Science Conference 2026 in Braunschweig 🇩🇪. 👉 More info & submission details: tsww26.github.io

This year’s workshop places a strong emphasis on preventive rather than reactive approaches to women’s online safety. We invite submissions presenting new findings, recent work, new ideas or previously published research (non-archival) that fits the workshop theme.

For today's reading group @arimuti.bsky.social presented "SPeCtrum: A Grounded Framework for Multidimensional Identity Representation in LLM-Based Agent" by @keyeun.bsky.social. #NLProc #identity

Honored to give my first keynote at #IRCDL2026 on February 19th. I’ll talk about how LLMs have shifted from productivity tools to everyday sources of info & personal guidance and what that means for risk, trust, bias, and alignment. ircdl2026.unimore.it

We were excited to host @naitian.org at today’s lab seminar for a talk on variation, semiotics, fashion, and style. A refreshing perspective at the intersection of sociolinguistics and NLP! #NLProc

For today's reading group @elisabassignana.bsky.social presented "How AI Impacts Skill Formation" by Judy Hanwen Shen & Alex Tamkin (2026). Paper: arxiv.org/pdf/2601.20245 #NLProc

🧠For this week’s lab seminar, @boleima.bsky.social talked about how survey methodology can inform NLP research, from annotations to human–AI alignment. #NLProc

For today's reading group, @deboranozza.bsky.social presented "LLM Targeted Underperformance Disproportionately Impacts Vulnerable Users" by Elinor Poole-Dayan et al. Paper: arxiv.org/pdf/2406.17737 #NLProc

Last week we welcomed Nikhil Sharma to our lab seminar for a talk on Information Seeking, Consumption and Dissemination with LLM-powered Information Systems. #NLProc #HCI

#TBT #NLProc 'SAFETYKIT: Measuring Safety in Open-domain Conversational Systems' by Dinan et al. (2022) introduces taxonomy for AI safety, assesses tools' limits. #AIsafety

For our weekly reading group, Afshin Karimi presented "What Social Media Use Do People Regret? An Analysis of 34K Smartphone Screenshots with Multimodal LLM". Super interesting discussion! Paper: dl.acm.org/doi/10.1145/... #NLProc

New year, new job? If that is your current mantra, check the open postdoc positions with Debora Nozza and me at our lab. Deadline is January 31st. milanlproc.github.io/open_positio...

#MemoryModay #NLProc @gattanasio.cc et al. study asks 'Is It Worth the (Environmental) Cost?' analyzing continuous training for language models. Balances benefits, environmental impacts, for responsible use. #AI #Sustainability arxiv.org/pdf/2210.07365

The deadline is approaching! Join the team :)

🚨(Software) Update: In my PhD, I had a side project to fix an annoying problem: when you ask 5 people to label the same thing, you often get different answers. But in ML (and lots of other analyses), you still need a single aggregated answer. Using the majority vote is easy–but often wrong. 1/N

However, disagreement isn’t just noise—it’s information. It can mean an item is genuinely hard—or someone wasn’t paying attention. If only you knew whom to trust… That summer, Taylor Berg-Kirkpatrick, Ashish Vaswani, and I built MACE (Multi-Annotator Competence Estimation). 2/N

MACE estimates: 1. Annotator reliability (who’s consistent?) 2. Item difficulty (which examples spark disagreement?) 3. The most likely aggregate label (the latent “best guess”) That “side project” ended up powering hundreds of annotation projects over the years. 3/N

I always wanted to revisit it, port it from Java to Python & extend to continuous data, but never found the time. Last week, I played around with Cursor – and got it all done in ~1 hour. 🤯 If you work with any response data that needs aggregation, give it a try—and let me know what you think! 4/N

If you are curious about the theoretical background, see Hovy, D., Berg-Kirkpatrick, T., Vaswani, A., & Hovy E. (2013). Learning Whom to Trust With MACE. In: Proceedings of NAACL-HLT. ACL. aclanthology.org/N13-1132.pdf And for even more details: aclanthology.org/Q18-1040.pdf N/N

I included some test cases on GitHub, will look if I still have the ones we used in the paper.

Found and added under data/

🚀 We’re opening 2 fully funded postdoc positions in #NLP! Join the MilaNLP team and contribute to our upcoming research projects. 🔗 More details: milanlproc.github.io/open_positio... ⏰ Deadline: Jan 31, 2026

For today's reading group, Serena Pugliese presented the paper “Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Models" by Piercosma Bisconti et al. (2025). Paper: arxiv.org/pdf/2511.15304 #NLProc #LLMs #jailbreaking

We're also back with the lab's seminar! Today we had Eleonora Mancini presenting her doctoral research "Multimodal AI for Human Expression Understanding". #NLP #multimodality #speech

🎉 MilaNLP 2025 Wrapped 🎉 Lots of learning, building , sharing, and growing together 🌱 #NLProc

⏳ Deadline approaching! We’re hiring 2 fully funded postdocs in #NLP. Join the MilaNLP team and contribute to our upcoming research projects (SALMON & TOLD) 🔗 Details + how to apply: milanlproc.github.io/open_positio... ⏰ Deadline: Jan 31, 2026

Thank you @belindazli.bsky.social for the great talk "Solving the Specification Problem through Interaction” at our weekly seminar! #NLProc

This week at reading group 📚 @pranav-nlp.bsky.social presented "You Cannot Sound Like GPT": Signs of language discrimination and resistance in computer science publishing. Paper: arxiv.org/abs/2505.08127 #NLProc

Thank you @belindazli.bsky.social for the great talk "Solving the Specification Problem through Interaction” at our weekly seminar! #NLProc

This week at reading group 📚 @pranav-nlp.bsky.social presented "You Cannot Sound Like GPT": Signs of language discrimination and resistance in computer science publishing. Paper: arxiv.org/abs/2505.08127 #NLProc

🎉 MilaNLP 2025 Wrapped 🎉 Lots of learning, building , sharing, and growing together 🌱 #NLProc

⏳ Deadline approaching! We’re hiring 2 fully funded postdocs in #NLP. Join the MilaNLP team and contribute to our upcoming research projects (SALMON & TOLD) 🔗 Details + how to apply: milanlproc.github.io/open_positio... ⏰ Deadline: Jan 31, 2026

We're also back with the lab's seminar! Today we had Eleonora Mancini presenting her doctoral research "Multimodal AI for Human Expression Understanding". #NLP #multimodality #speech

Holidays over, reading group resumes 📖 Today Henning Hoffmann presented the paper "Music for All: Representational Bias and Cross-Cultural Adaptability of Music Generation Models" Paper: arxiv.org/pdf/2502.07328 #NLProc

#MemoryModay #NLProc Countering Hateful and Offensive Speech Online - Open Challenges" by Plaza-Del-Arco, @debora_nozza, Guerini, Sorensen, Zampieri, 2024 is a tutorial on the challenges and solutions for detecting and mitigating hate speech.

#TBT #NLProc #MachineLearning #SafetyFirst 'Safety-Tuned LLaMAs: Improving LLMs Safety' by Bianchi et al. explores training LLMs for safe refusals, warns of over-tuning.

🚀 We’re opening 2 fully funded postdoc positions in #NLP! Join the MilaNLP team and contribute to our upcoming research projects. 🔗 More details: milanlproc.github.io/open_positio... ⏰ Deadline: Jan 31, 2026

For today's reading group, Serena Pugliese presented the paper “Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Models" by Piercosma Bisconti et al. (2025). Paper: arxiv.org/pdf/2511.15304 #NLProc #LLMs #jailbreaking

Huge thanks to our speakers at last Friday’s lab seminar! 🗣️ @penzo-nicolo.bsky.social on multi-party conversations 🌍 @patriciachiril.bsky.social on NLP for socially grounded research #NLProc

#MemoryModay #NLProc Uma, A. N. et al. examine AI model training in 'Learning from Disagreement: A Survey'. Disagreement-handling methods' performance is shaped by evaluation methods & dataset traits.

#TBT #NLProc 'Respectful or Toxic?' by Plaza-del-Arco, @debora & @dirkhovy.bsky.social (2023) explores zero-shot learning for multilingual hate speech detection. Highlights prompt & model choice for accuracy. #AI #LanguageModels #HateSpeechDetection

At today’s lab reading group @carolin-holtermann.bsky.social presented ‘Fairness through Difference Awareness: Measuring Desired Group Discrimination in LLMs’ by @angelinawang.bsky.social et al. (2025). Lots to think about how we evaluate fairness in language models! #NLProc #fairness #LLMs

For our weekly reading group last week, @a-lauscher.bsky.social presented the paper “Shape it Up! Restoring LLM Safety during Finetuning" by ShengYun Peng et al. (2025) #NLProc

#MemoryModay #NLProc 'Leveraging Social Interactions to Detect Misinformation on Social Media' by Fornaciari et al. (2023) uses combined text and network analysis to spot unreliable threads.

#TBT #NLProc 'Compromesso! Italian Many-Shot Jailbreaks Undermine LLM Safety' by Pernisi, @dirkhovy.bsky.social, @paul-rottger.bsky.social (2024). Paper highlights LLM vulnerability through Italian demos, more demos = more attack chances.

I will be @euripsconf.bsky.social this week to present our paper as non-archival at the PAIG workshop (Beyong Regulation: Private Governance & Oversight Mechanisms for AI). Very much looking forward to the discussions! If you are at #EurIPS and want to chat about LLM's training data. Reach out!

#MemoryModay #NLProc 'Multilingual HateCheck: Functional Tests for Multilingual Hate Speech Detection Models' by @paul-rottger.bsky.social et al. (2022). A suite of tests for 10 languages.