Iryna Gurevych

Aligned Probing: Relating Toxic Behavior and Model Internals Aligned Probing: Relating Toxic Behavior and Model Internals

Questions? Discussion? Reach out to us.
@dippedrusk.com @a-lauscher.bsky.social Dietrich Klakow @igurevych.bsky.social
Full paper & code: alignedprobing.github.io
(7/7)

In a new briefing by the @sciencemediacenter.de, Prof. Dr. @igurevych.bsky.social (@tuda.bsky.social) notes that the study’s methodology is well aligned with its claims: It extends earlier work by the same lab showing that fine-tuning can lead to broader misalignment.

(2/🧵)

Promotional graphic for a keynote on “Roadmap to Internet Security.” The background shows a glowing digital lock and circuit-board imagery. Large text announces the talk title “How to Make AI-Native Internet Content Secure? Coping with Synthetic and Misleading Data.” On the right is a circular portrait of a woman with shoulder-length curly hair and glasses (Iryna Gurevych), looking toward the camera.

🔐 𝗞𝗲𝘆𝗻𝗼𝘁𝗲 𝗯𝘆 𝗜𝗿𝘆𝗻𝗮 𝗚𝘂𝗿𝗲𝘃𝘆𝗰𝗵 𝗮𝘁 𝗥𝗼𝗮𝗱𝗺𝗮𝗽 𝘁𝗼 𝗜𝗻𝘁𝗲𝗿𝗻𝗲𝘁 𝗦𝗲𝗰𝘂𝗿𝗶𝘁𝘆 𝟮𝟬𝟮𝟱
We are pleased to share that Prof. @igurevych.bsky.social, Distinguished Professor at the @athenecenter.bsky.social and Director of the UKP Lab at @tuda.bsky.social, will deliver a keynote at the ATHENE event ...

Do you like AI for Cybersecurity, Agentic AI research or both? This job is might be for you!

👏🏼Congratulations🍾🥂on this fantastic award from us to 𝗜𝗿𝘆𝗻𝗮 𝗚𝘂𝗿𝗲𝘃𝘆𝗰𝗵, who is not only the🥇LOEWE top professor, but also involved in other LOEWE projects such as LOEWE @emergencity.de and @loewe-dynamic.bsky.social! @igurevych.bsky.social @tuda.bsky.social @unimarburg.bsky.social @royalsociety.org

As part of the ceremony at the Royal Society in London, @igurevych.bsky.social delivered the Milner Prize Lecture titled 𝘏𝘰𝘸 𝘵𝘰 𝘴𝘱𝘰𝘵 𝘢𝘯𝘥 𝘥𝘦𝘣𝘶𝘯𝘬 𝘮𝘪𝘴𝘭𝘦𝘢𝘥𝘪𝘯𝘨 𝘤𝘰𝘯𝘵𝘦𝘯𝘵, addressing current challenges in detecting and countering misleading information.

The @royalsociety.org Milner Award Lecture 2025: Prof. @igurevych.bsky.social's on how AI fact checkers can be misled (similar to how humans are, in many ways) by implicit source citation fallacies, image feature training bias, and misleading chart design, and what we can do about it.

Happens tomorrow, online attendance possible!

Sprachmodelle: Wie weit die Künstliche Intelligenz wirklich ist ChatGPT löste den bislang größten KI-Boom aus. Doch wie weit sind die Sprachmodelle inzwischen? Und wie geht es weiter? Ein Überblick.

It touches on fast-moving model releases, the growing relevance of AI agents and tool-use, the shift towards reasoning-oriented training, and the limits of what benchmark results really tell us.

🔗 Read the article (F+ / paywalled):
www.faz.net/aktuell/wirt...

(2/2)

#INLG2025 main conference starts with not one, not two, but three keynotes! We'll hear from Minlie Huang, Iryna Gurevych (@igurevych.bsky.social), and Verena Rieser (@verenarieser.bsky.social), as well as oral and poster sessions for accepted papers.

Towards privacy-aware mental health AI models - Nature Computational Science In this Perspective, the authors examine privacy risks in mental health AI, and explore solutions and evaluation frameworks to balance privacy–utility trade-offs. They suggest a pipeline for developing privacy-aware mental health AI systems.

In another Perspective, @igurevych.bsky.social and colleagues examine privacy risks in mental health AI, as well as explore solutions and evaluation frameworks to balance privacy-utility trade-offs. #mentalhealthresearch #ArtificialIntelligence

➡️ nature.com/articles/s43...
🔓 rdcu.be/eKp4v

🧵(6/8)

KI mit Schweigepflicht: Forschende der TU Darmstadt & IIT Delhi zeigen, wie KI in der Psychodiagnostik helfen kann – ohne sensible Daten zu gefährden @igurevych.bsky.social
👉 tu-darmstadt.de/universitaet/aktuelles_meldungen/einzelansicht_527552.de.jsp

Logos of the UKP Lab (Ubiquitous Knowledge Processing) and Hugging Face side by side with a plus sign between them. Below, the text reads: “Sentence Transformers is joining Hugging Face.”

🤝 𝗦𝗲𝗻𝘁𝗲𝗻𝗰𝗲 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗲𝗿𝘀 𝗷𝗼𝗶𝗻𝘀 𝗛𝘂𝗴𝗴𝗶𝗻𝗴 𝗙𝗮𝗰𝗲
Originally developed at the UKP Lab at @tuda.bsky.social, Sentence Transformers has become one of the world’s most widely used open-source libraries for semantic embeddings in natural language processing.

(1/🧵)

Join Professor Iryna Gurevych for the Milner Prize Lecture on 17 November. She will demonstrate how machine learning and artificial intelligence can be applied to protect people and machines from misinformation: royalsociety.org/science-even... @igurevych.bsky.social @tuda.bsky.social

🙌 𝗛𝘂𝗴𝗲 𝘁𝗵𝗮𝗻𝗸𝘀 𝘁𝗼 𝗼𝘂𝗿 𝗰𝗼𝗹𝗹𝗮𝗯𝗼𝗿𝗮𝘁𝗼𝗿𝘀!
This release wouldn’t be possible without the contributions of: Sheng Lu, Nils Dycke, @atnafu.bsky.social, Thamar Solorio, Xiaodan Zhu, Koen Dercksen, Lizhen Qu, Margot Mieskes, @dirkhovy.bsky.social and @igurevych.bsky.social.

(4/🧵)

Quantum AI and NLP Conference 2025 Quantum AI and NLP Conference 2025 Website. The conference will be held from the 6th to the 8th of August 2025 in Bloomington, Indiana at Indiana University.

This work was made possible through a great collaboration with

👩‍🔬 Anna Schroeder & Mariami Gachechiladze (Quantum Computing Group, @tuda.bsky.social)

👨‍🔬 Yue Zhang (Westlake University)

👩‍💻 @igurevych.bsky.social (@ukplab.bsky.social, @tuda.bsky.social)

(2/🧵)

ELLIS PhD Program: Call for Applications 2025 The ELLIS mission is to create a diverse European network that promotes research excellence and advances breakthroughs in AI, as well as a pan-European PhD program to educate the next generation of AI...

🎓 The application portal for the #ELLISPhD Program is open today! Info & link: https://bit.ly/45DSe75

🎓 What could a pan-European PhD program focusing on #MachineLearning & AI be like?

Check out @wafaamohammed.bsky.social's reason for joining the #ELLISPhD Program.

You can apply to the via our central recruiting portal starting on Oct 1st.

Get all the details now 👉 bit.ly/45DSe75

📢 Call for Posters – ELLIS UnConference 2025

Have a 2025 published paper? Present it in Copenhagen on Dec 2!

🔹 Showcase your research
🔹 Get feedback & connect with peers
🔹 Grow your network

📝 Submit here: https://forms.gle/vUUdM9mah34swX7U7
📅 Deadline: Oct 23, 17:00 CEST

#EurIPS

Promotional graphic for the Discovery Science 2025 International Conference keynote. The text reads: “Please meet AI, our dear new colleague. In other words: can scientists and machines truly cooperate?” Keynote by Prof. Iryna Gurevych. On the right is a photo of Prof. Iryna Gurevych standing in front of bookshelves. The logos of UKP Lab and Discovery Science are included at the bottom.

🚀 𝗞𝗲𝘆𝗻𝗼𝘁𝗲 𝗯𝘆 𝗣𝗿𝗼𝗳. 𝗜𝗿𝘆𝗻𝗮 𝗚𝘂𝗿𝗲𝘃𝘆𝗰𝗵 𝗶𝗻 𝗟𝗷𝘂𝗯𝗹𝗷𝗮𝗻𝗮
Prof. @igurevych.bsky.social, Head of the UKP Lab, is giving a keynote today at the 28th International Conference on Discovery Science 2025 in Ljubljana.

#Dagstuhl Seminar "Open Scholarly Information Systems:
Iryna Gurevych @igurevych.bsky.social introduces us to AI, our dear new colleague: Are we becoming obsolete?

We’re excited to share that @igurevych.bsky.social, Director of the UKP Lab at @tuda.bsky.social, will join the expert panel to discuss the opportunities and challenges of Generative AI from a research perspective.

(2/🧵)

🎤 Panel with Peter Buxmann (@tuda.bsky.social), Holger Hanselka (‪@fraunhofer.bsky.social‬), Florian Rentsch (Verband der Sparda-Banken e.V.), Sara Jourdan (@tuda.bsky.social) , Carsten Knop (@faznet.bsky.social), and @igurevych.bsky.social (@ukplab.bsky.social / @tuda.bsky.social)

(3/🧵)

Iryna Gurevych in die Academia Europaea gewählt 06.06.2025 ATHENE Principal Investigator Prof. Iryna Gurevych ist in die Academia Europaea gewählt worden. Damit wird ihre langjährige akademische Exzellenz und internationale Sichtbarkeit in den Be...

🎓 ATHENE Principal Investigator Prof. 𝗜𝗿𝘆𝗻𝗮 𝗚𝘂𝗿𝗲𝘃𝘆𝗰𝗵 @igurevych.bsky.social wurde in die Academia Europaea @acad-euro.bsky.social gewählt – für ihre langjährige akademische Exzellenz in den Bereichen Künstliche Intelligenz #KI & Natural Language Processing. 🎉

👉 www.athene-center.de/aktuelles/ne...

This is a unique opportunity to work with @igurevych.bsky.social and the team on the intersection of AI Safety, Natural Language Processing and Machine Learning.

(2/🧵)

Super important initiative, thanks for doing this great work!

🛌 𝗔𝗺 𝟮𝟭. | ProLOEWE 🛌 𝗔𝗺 𝟮𝟭. 𝗝𝘂𝗻𝗶 𝗶𝘀𝘁 𝗧𝗮𝗴 𝗱𝗲𝘀 𝗦𝗰𝗵𝗹𝗮𝗳𝗲𝘀. 𝗪𝗮𝘀 𝗦𝗰𝗵𝗹𝗮𝗳 𝗺𝗶𝘁 𝗽𝘀𝘆𝗰𝗵𝗶𝘀𝗰𝗵𝗲𝗿 𝗚𝗲𝘀𝘂𝗻𝗱𝗵𝗲𝗶𝘁 𝘇𝘂 𝘁𝘂𝗻 hat? 𝗘𝗶𝗻𝗲 𝗴𝗮𝗻𝘇𝗲 𝗠𝗲𝗻𝗴𝗲! #Schlafmangel kann nicht nur kurzfristig unsere Stimmung trüben, er ist auch ein zentraler Fakt...

21.06.= #TagdesSchlafes. Schlaf ist zentral für die psychische🧠Gesundheit+spielt bei #LOEWE-DYNAMIC eine wichtige Rolle. #GrundlagenForschung #Hessen
@unimarburg.bsky.social @jlugiessen.bsky.social @tuda.bsky.social @goetheuni.bsky.social @dipfaktuell.bsky.social ➡️ www.linkedin.com/posts/proloe...

Portrait of Iryna Gurevych. ©Markus Scholz

𝗜𝗿𝘆𝗻𝗮 𝗚𝘂𝗿𝗲𝘃𝘆𝗰𝗵 𝗲𝗹𝗲𝗰𝘁𝗲𝗱 𝗮𝘀 𝗺𝗲𝗺𝗯𝗲𝗿 𝗼𝗳 𝘁𝗵𝗲 𝗔𝗰𝗮𝗱𝗲𝗺𝗶𝗮 𝗘𝘂𝗿𝗼𝗽𝗮𝗲𝗮
@tuda.bsky.social computer scientist Prof. Dr. Iryna Gurevych (@igurevych.bsky.social) has been elected as a member of the Academia Europaea (@acad-euro.bsky.social‬), the pan-European academy of sciences, humanities and letters.
(1/🧵)

DeepSeek-Modelle auf dem Prüfstand In einem Gastbeitrag geben Professorin Iryna Gurevych und Irina Bigoulaeva vom Ubiquitous Knowledge Processing (UKP) Lab am Fachbereich Informatik der TU Darmstadt einen Einblick in ihre Forschung zum...

DeepSeek: In einem Gastbeitrag geben Prof. Iryna Gurevych und Irina Bigoulaeva vom Ubiquitous Knowledge Processing Lab am Fachbereich Informatik einen Einblick in ihre Forschung zum Leistungsvermögen von generativer Künstlicher Intelligenz @cs-tudarmstadt.bsky.social @ukplab.bsky.social

Photo of the Vienna Giant Ferris Wheel with the lettering "UKPLab @ACL2025"

We are happy to announce that 10 papers authored or co-authored by UKP members have been accepted by the 63nd Annual Meeting of the @aclmeeting.bsky.social (#ACL2025) taking place this year in beautiful Vienna 🎶☕.

Congratulations to everyone involved! 🚀📚

The 8 main conference papers are:
(1/🧵)