iiinova — where your digital self keeps evolving

UKP Lab @ukplab.bsky.social · 01/10/2026

🛡️ Welcome to the UKP Lab, Petros Raptopoulos! Petros joins us as a PhD student on the HAICC project (Human-AI Collaboration for Cybersecurity) at #ATHENE. His thesis explores how AI agents in security-critical applications can be safely deployed under human oversight. Welcome, Petros! 👋

Welcome graphic for Petros Raptopoulos, Doctoral researcher at UKP Lab. His portrait photo is framed in a red circle, with the UKP Lab and TU Darmstadt logos and the word "Welcome".

111

UKP Lab @ukplab.bsky.social · 30/09/2026

@ellis.eu 𝗣𝗵𝗗 𝗣𝗿𝗼𝗴𝗿𝗮𝗺 𝟮𝟬𝟮𝟲: 𝗮𝗽𝗽𝗹𝘆 𝗯𝘆 𝟯𝟭 𝗢𝗰𝘁𝗼𝗯𝗲𝗿 Prof. @igurevych.bsky.social is a registered advisor, so you can apply to work on AI and NLP at UKP Lab. Info session for applicants on 30 September. Details: ellis.eu/news/ellis-p...

Promotional graphic for the ELLIS PhD & Postdoc Program: "Call for PhD Applications 2026 – Central Application Portal opens in October". The centre shows a map of Europe with numerous locations connected by lines. On the right are the logos of the EU (Funded by the European Union), ELIAS, elliot, and G-Research.

021

UKP Lab @ukplab.bsky.social · 21/09/2026

🎓 𝗨𝗞𝗣 𝗟𝗮𝗯 𝘄𝗲𝗹𝗰𝗼𝗺𝗲𝘀 𝗠𝗼𝗵𝗮𝗺𝗺𝗮𝗱 𝗥𝗮𝘄𝗵𝗮𝗻𝗶! Our new PhD student, Mohammad, brings a research focus on 𝗡𝗮𝘁𝘂𝗿𝗮𝗹 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗣𝗿𝗼𝗰𝗲𝘀𝘀𝗶𝗻𝗴, 𝗹𝗮𝗿𝗴𝗲 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗺𝗼𝗱𝗲𝗹𝘀, and 𝗺𝗮𝗰𝗵𝗶𝗻𝗲 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗳𝗼𝗿 𝗹𝗼𝘄-𝗿𝗲𝘀𝗼𝘂𝗿𝗰𝗲 𝘀𝗲𝘁𝘁𝗶𝗻𝗴𝘀. His previous academic work centered on 𝗱𝗼𝗺𝗮𝗶𝗻 𝗮𝗱𝗮𝗽𝘁𝗮𝘁𝗶𝗼𝗻 𝘂𝘀𝗶𝗻𝗴 𝗽𝗮𝗿𝗮𝗺𝗲𝘁𝗲𝗿-𝗲𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁 𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝗺𝗲𝘁𝗵𝗼𝗱𝘀.

Welcome graphic for Mohammed Al-Rawhani, Doctoral researcher at UKP Lab. His portrait photo is framed in a red circle, with the UKP Lab and TU Darmstadt logos and the word "Welcome" repeated four times.

220

UKP Lab @ukplab.bsky.social · 14/07/2026

🏆 𝗕𝗲𝘀𝘁 𝗣𝗮𝗽𝗲𝗿 𝗔𝘄𝗮𝗿𝗱 𝗮𝘁 𝗦𝗲𝗺𝗘𝘃𝗮𝗹-𝟮𝟬𝟮𝟲 𝗦𝗵𝗮𝗿𝗲𝗱 𝗧𝗮𝘀𝗸 𝟮! We're proud to share that work from UKP Lab has received the 𝗕𝗲𝘀𝘁 𝗣𝗮𝗽𝗲𝗿 𝗔𝘄𝗮𝗿𝗱 𝗼𝗳 𝗦𝗲𝗺𝗘𝘃𝗮𝗹-𝟮𝟬𝟮𝟲 𝗦𝗵𝗮𝗿𝗲𝗱 𝗧𝗮𝘀𝗸 𝟮, recognizing the scientific contribution behind a system that also topped the leaderboard with 𝟭𝘀𝘁 𝗽𝗹𝗮𝗰𝗲 𝗶𝗻 𝗦𝘂𝗯𝘁𝗮𝘀𝗸𝘀 𝟭 𝗮𝗻𝗱 𝟮𝗔 😲

Darya Hryhoryeva holding the SemEval-2026 Best Paper Award certificate for 'UKP_Psycontrol at SemEval-2026 Task 2: Modeling Valence and Arousal Dynamics from Text' at ACL 2026.

141

UKP Lab @ukplab.bsky.social · 25/06/2026

🎓 𝗚𝘂𝗲𝘀𝘁 𝗧𝗮𝗹𝗸 𝗮𝘁 𝘁𝗵𝗲 𝗨𝗞𝗣 𝗟𝗮𝗯 We were thrilled to welcome our long-term collaborator, co-supervisor of PhD students and close colleague Prof. Preslav Nakov, Professor and Department Chair for #NLP at the #MBZUAI in Abu Dhabi, for an invited talk at the UKP Lab in #TUDarmstadt.

Preslav Nakov (left) and Iryna Gurevych (right) standing in front of a projection screen at TU Darmstadt on June 19, 2026, before Preslav Nakov's talk titled 'Human or AI? Interpretable Detection, Multilingual Perception, and the Paradox of Human-Like Text' from Mohamed bin Zayed University of Artificial Intelligence (MBZUAI).

120

UKP Lab @ukplab.bsky.social · 19/06/2026

𝗙𝗿𝗼𝗺 𝗗𝗮𝗿𝗺𝘀𝘁𝗮𝗱𝘁 𝘁𝗼 𝗔𝗖𝗟 𝟮𝟬𝟮𝟲 𝗶𝗻 𝗦𝗮𝗻 𝗗𝗶𝗲𝗴𝗼 Four UKP Lab papers got an early audience this week. @timbmg.bsky.social, @ibigoulaeva.bsky.social, and @tongletj.bsky.social presented their work at the Post-CVPR Pre-ICML Poster Session, [...]

140

UKP Lab @ukplab.bsky.social · 10/06/2026

👋 𝗡𝗲𝘄 𝗳𝗮𝗰𝗲 𝗮𝘁 𝘁𝗵𝗲 𝗨𝗞𝗣 𝗟𝗮𝗯: 𝗔 𝘄𝗮𝗿𝗺 𝘄𝗲𝗹𝗰𝗼𝗺𝗲 𝘁𝗼 𝗔𝗻𝗱𝗿𝗲𝗮! @andreamoleri.bsky.social joins us as a doctoral researcher, jointly supervised by Prof. Dr. @igurevych.bsky.social and Prof. Dr. Marcus Rohrbach.

Welcome graphic for Andrea Moleri, Doctoral researcher at UKP Lab. His portrait photo is framed in a red circle, with the UKP Lab and TU Darmstadt logos and the word "Welcome" repeated four times.

110

UKP Lab @ukplab.bsky.social · 03/06/2026

🚀 𝗟𝗮𝘁𝗲𝘀𝘁 𝗡𝗟𝗣𝗲𝗲𝗿 𝗥𝗲𝗹𝗲𝗮𝘀𝗲 𝗳𝗿𝗼𝗺 𝗔𝗥𝗥 𝟮𝟬𝟮𝟲 𝗶𝘀 𝗼𝘂𝘁! Grab your Peer Review data: tudatalib.ulb.tu-darmstadt.de/handle/tudat... 📊 𝗘𝗔𝗖𝗟 𝟮𝟲 𝗔𝗰𝗰𝗲𝗽𝘁𝗲𝗱 𝗣𝗮𝗽𝗲𝗿 𝗗𝗮𝘁𝗮: ✅ 647 papers ✅ 942 reviews ✅ 329 meta-reviews ✅ 550 papers with rebuttals

Black-and-white illustration of a person wearing a graduation cap, examining a document with a magnifying glass.Black-and-white illustration of a person wearing a graduation cap, examining a document with a magnifying glass. A robotic parrot sits on their shoulder.

100

UKP Lab @ukplab.bsky.social · 29/05/2026

A big thank you to @i-am-rec.bsky.social for moderating the day. We would finally like to thank – in no particular order – Serwar, Elisabeth, and Patricia for the organization, and everybody who helped set up the event for your support! We are already looking forward to the next UKP Day.

100

UKP Lab @ukplab.bsky.social · 29/05/2026

We rounded off the day with a relaxed social event – a great chance to keep the conversations going in a more informal setting. Beyond the formal program, the day once again highlighted what makes our lab special: collaboration, collegial spirit, and open exchange.

100

UKP Lab @ukplab.bsky.social · 29/05/2026

The talks sparked lively discussions throughout the day and offered fresh perspectives on trustworthy AI, multi-agent systems, human–AI collaboration, and where the field is headed next.

110

UKP Lab @ukplab.bsky.social · 29/05/2026

▪️ Monojit Choudhury (MBZUAI (Mohamed bin Zayed University of Artificial Intelligence)) "Turing Test 2.0: Reimagining the Goals of Artificial Intelligence in the Post-GenAI World"

100

UKP Lab @ukplab.bsky.social · 29/05/2026

▪️ Yuxia Wang (INSAIT - Institute for Computer Science, Artificial Intelligence and Technology) "When Do Language Models Fail Quietly? From Average to Individual"

100

UKP Lab @ukplab.bsky.social · 29/05/2026

▪️ Kaiyang Wan (INSAIT - Institute for Computer Science, Artificial Intelligence and Technology) "Why Do We Need Multi-Agent Systems? An Information-Theoretic Perspective"

110

UKP Lab @ukplab.bsky.social · 29/05/2026

▪️ Lei You (DTU - Technical University of Denmark) "Trustworthy AI at Ecosystem Scale: What Cannot Be Replaced? What Must Be Verified?"

100

UKP Lab @ukplab.bsky.social · 29/05/2026

🎤 This year's program featured a series of inspiring talks by our guest speakers: ▪️Frauke Kreuter (@lmu.de / @univofmaryland.bsky.social) "Bias in the Loop: How Humans Evaluate AI-Generated Suggestions"

100

UKP Lab @ukplab.bsky.social · 29/05/2026

✨ 𝗨𝗞𝗣 𝗗𝗮𝘆 𝟮𝟬𝟮𝟲 ✨ Once a year, the entire UKP Lab comes together for a day of scientific exchange, discussion, and community building – the UKP Day. On May 28, 2026, we gathered at @tuda.bsky.social to share ideas, learn from one another, and connect with colleagues from across the field.

110

UKP Lab @ukplab.bsky.social · 21/05/2026

💻 𝗪𝗲𝗹𝗰𝗼𝗺𝗲 𝘁𝗼 𝘁𝗵𝗲 𝗨𝗞𝗣 𝗟𝗮𝗯, 𝗣𝗵𝘂 𝗚𝗶𝗮 𝗛𝗼𝗮𝗻𝗴! We're happy to welcome Phu Gia Hoang as a doctoral researcher at the UKP Lab. At UKP, Phu works on 𝗺𝗲𝗰𝗵𝗮𝗻𝗶𝘀𝘁𝗶𝗰 𝗶𝗻𝘁𝗲𝗿𝗽𝗿𝗲𝘁𝗮𝗯𝗶𝗹𝗶𝘁𝘆, exploring 𝗴𝗲𝗼𝗺𝗲𝘁𝗿𝗶𝗰 𝗶𝗻𝘁𝘂𝗶𝘁𝗶𝗼𝗻 as a lens to better understand how language models represent and process information internally.

Welcome graphic from UKP Lab at TU Darmstadt introducing Phu Gia Hoang as a new Doctoral Researcher, with his portrait next to a 'Welcome' message on a background of scattered letter tiles.

100

UKP Lab @ukplab.bsky.social · 06/05/2026

🚀 𝗪𝗲'𝗿𝗲 𝗵𝗶𝗿𝗶𝗻𝗴: 𝗣𝗼𝘀𝘁𝗱𝗼𝗰𝘁𝗼𝗿𝗮𝗹 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵𝗲𝗿 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗘𝗥𝗖 𝗔𝗱𝘃𝗮𝗻𝗰𝗲𝗱 𝗚𝗿𝗮𝗻𝘁 𝗜𝗻𝘁𝗲𝗿𝗧𝗲𝘅𝘁 𝗮𝘁 𝗨𝗞𝗣 𝗟𝗮𝗯, 𝗧𝗨 𝗗𝗮𝗿𝗺𝘀𝘁𝗮𝗱𝘁 If you work on #LLMs that reason across complex documents, not just within them, this role is for you.

101

UKP Lab @ukplab.bsky.social · 29/04/2026

🌍 𝗝𝗼𝗶𝗻𝗶𝗻𝗴 𝘁𝗵𝗲 𝗨𝗞𝗣 𝗟𝗮𝗯: 𝗪𝗲𝗹𝗰𝗼𝗺𝗲, 𝗞𝘂𝗿𝘁 𝗠𝗶𝗰𝗮𝗹𝗹𝗲𝗳! We’re glad to welcome Kurt Micallef as a new postdoctoral researcher at the UKP Lab. Kurt’s previous work focused on 𝗡𝗟𝗣 𝗳𝗼𝗿 𝗠𝗮𝗹𝘁𝗲𝘀𝗲, where he worked on low resource and cross lingual techniques, developed Maltese centric language models, [...]

Welcome graphic from the UKP Lab featuring Kurt Micallef, labeled “Postdoctoral researcher.” On the left is a circular portrait of a man standing outdoors at sunset. On the right, the word “WELCOME” appears repeatedly in the background. Logos of the Ubiquitous Knowledge Processing Lab and Technische Universität Darmstadt are included, along with social media icons and the handle @UKPLab.

100

UKP Lab @ukplab.bsky.social · 22/04/2026

🤔 How can we innovate reading and writing workflows for students, educators, researchers – and at the same time make it useful to science? 🚀 Meet 𝗖𝗔𝗥𝗘: — now with a 𝗺𝗮𝗷𝗼𝗿 𝘂𝗽𝗱𝗮𝘁𝗲 and 𝗳𝘂𝗹𝗹𝘆 𝗼𝗽𝗲𝗻 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁 on GitHub. Check out the 🧵 below if you want to know more.

Promotional graphic announcing a “Major Release” of CARE with the phrase “Opening Dev.” The word “CARE” is split into two blocks: “CA” inside a speech bubble icon and “RE” in a yellow rectangle. The background features faint icons related to research and writing, such as books, charts, links, and documents.

120

UKP Lab @ukplab.bsky.social · 27/03/2026

🔍 𝗨𝘀𝗶𝗻𝗴 𝗾𝘂𝗮𝗻𝘁𝗶𝘇𝗲𝗱 𝗟𝗟𝗠𝘀 𝘁𝗼 𝗯𝗼𝗼𝘀𝘁 𝗲𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆? 𝗪𝗮𝘁𝗰𝗵 𝘁𝗵𝗲 𝘀𝗼𝗰𝗶𝗮𝗹 𝘁𝗿𝗮𝗱𝗲-𝗼𝗳𝗳𝘀, 𝘁𝗼𝗼. ⚠️ Quantization makes LLMs cheaper and more deployable. But what does it do to fairness, toxicity, and bias? 💡 The research gap: a systematic picture of how quantization affects social behavior across models and tasks.

Diagram illustrating a social bias evaluation pipeline for large language models. On the left, an LLM is transformed via a quantizer into a 4-bit quantized LLM. Both models are then evaluated within a “Social Bias Evaluation Framework” that includes categories such as stereotype (SS, RB, WB, BBQ), toxicity (BLD, DTT), sentiment (BLD), and fairness (DE, DEG, DTF). Arrows lead to bar charts on the right labeled “Social Bias Level,” comparing bias levels across models.

110

UKP Lab @ukplab.bsky.social · 26/03/2026

#EACL2026 in Rabat is in full swing ✨ From papers to hallway discussions - it’s great to see our team actively contributing to this year’s #EACL conference. Say hello to this group if you’re on site 👇

Group photo of eight conference participants standing in front of an EACL 2026 backdrop in Rabat, Morocco. They wear name badges and casual to semi-formal attire, posing indoors against a banner that reads “EACL 2026 Rabat, Morocco, March 24–29, 2026” with sponsor logos and decorative patterns.

150

UKP Lab @ukplab.bsky.social · 25/03/2026

🎓 𝗚𝘂𝗲𝘀𝘁 𝗧𝗮𝗹𝗸 𝗮𝘁 𝘁𝗵𝗲 𝗨𝗞𝗣 𝗟𝗮𝗯 👋 We were pleased to welcome Yova Kementchedjhieva, Assistant Professor at the MBZUAI (Mohamed bin Zayed University of Artificial Intelligence), for a guest talk at the UKP Lab in Darmstadt.

Promotional graphic for a guest talk featuring Prof. Yova Kementchedjhieva. A portrait of a smiling woman with curly hair, arms crossed, is framed in a red circle against a background of keyboard keys spelling “USE.” Text reads “Prof. Yova Kementchedjhieva” and “MBZUAI Guest Talk.” Logos of the Ubiquitous Knowledge Processing Lab and Technische Universität Darmstadt appear, along with social media icons and the handle @UKPLab.

130

UKP Lab @ukplab.bsky.social · 24/03/2026

🚀 From school project to award-winning AI for mental health Meet Mihai Ghețu (@mihai-eduard-ghetu.bsky.social) - a remarkably talented high school student and the creator of Mental Mint: An Empathic AI-Powered Psychosocial Companion Robot for Victims of Emotional Shocks and Bullying.

Young student in a suit stands at a science fair booth, holding a small device with a face-like display and grass on top. Behind him, posters labeled “Regionalsieger” present diagrams and results of the project. On the table in front are hardware components, printed schematics, and a laptop displaying the device interface. Other participants and exhibition stands are visible in the background.

110

UKP Lab @ukplab.bsky.social · 23/03/2026

𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗰 𝗿𝗲𝘃𝗶𝗲𝘄𝗲𝗿𝘀 𝗰𝗮𝗻 𝗺𝗶𝘀𝘀 𝗳𝘂𝗻𝗱𝗮𝗺𝗲𝗻𝘁𝗮𝗹 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗲𝗿𝗿𝗼𝗿𝘀. 👀 LLM-generated reviews may look convincing — but how reliable are they in practice? In our recent TACL paper, we introduce a 𝗰𝗼𝗻𝘁𝗿𝗼𝗹𝗹𝗲𝗱 𝗰𝗼𝘂𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝘁𝘂𝗮𝗹 𝗲𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻 𝗳𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸 to systematically test automatic reviewers.

Schematic diagram of analyzing scientific papers. On the left, a document contains color-coded sections representing claim, conclusion, result, and method. In the center, these categories are listed, with dashed lines linking them to corresponding parts of the document. A red cross marks an incorrect linkage between a result and a section in a second document shown on the right, where one passage is highlighted. At the bottom, small robot figures are shown reviewing documents, indicating automated evaluation or peer review processes.

130

UKP Lab @ukplab.bsky.social · 18/03/2026

𝗘𝗺𝗽𝗮𝘁𝗵𝘆 𝗶𝘀𝗻’𝘁 𝗲𝗻𝗼𝘂𝗴𝗵: 𝗔𝗜-𝗯𝗮𝘀𝗲𝗱 𝘁𝗵𝗲𝗿𝗮𝗽𝗶𝘀𝘁𝘀 𝗰𝗿𝘂𝗰𝗶𝗮𝗹𝗹𝘆 𝗻𝗲𝗲𝗱 𝗰𝘂𝗹𝘁𝘂𝗿𝗮𝗹 𝗰𝗼𝗻𝘁𝗲𝘅𝘁! 🤖🫶🌍 Millions of people worldwide lack access to mental health care. LLMs are increasingly explored as scalable tools for empathetic support, 𝗯𝘂𝘁 they often miss 𝗮 𝗸𝗲𝘆 𝗱𝗶𝗺𝗲𝗻𝘀𝗶𝗼𝗻: 𝗰𝘂𝗹𝘁𝘂𝗿𝗲.

Illustration of the CultureCare framework for culturally aware emotional support in AI systems. The diagram shows a distress message from a 17-year-old in Sudan describing anger issues and stigma around therapy, with cultural signals and emotional distress highlighted. It compares responses: a detailed human response offering supportive suggestions, a generic LLM response acknowledging emotional coping, and an adapted LLM response that incorporates cultural context and suggests speaking with a trusted family member or friend.

220

UKP Lab @ukplab.bsky.social · 16/03/2026

🔥 𝗠𝗲𝘁𝗮-𝗿𝗲𝘃𝗶𝗲𝘄𝗶𝗻𝗴 𝗶𝘀 𝗺𝗼𝗿𝗲 𝘁𝗵𝗮𝗻 𝗮 𝘀𝘂𝗺𝗺𝗮𝗿𝗶𝘇𝗮𝘁𝗶𝗼𝗻 - 𝗶𝘁’𝘀 𝗱𝗲𝗰𝗶𝘀𝗶𝗼𝗻-𝗺𝗮𝗸𝗶𝗻𝗴. In our new paper, “𝗗𝗲𝗰𝗶𝘀𝗶𝗼𝗻-𝗠𝗮𝗸𝗶𝗻𝗴 𝘄𝗶𝘁𝗵 𝗗𝗲𝗹𝗶𝗯𝗲𝗿𝗮𝘁𝗶𝗼𝗻: 𝗠𝗲𝘁𝗮-𝗿𝗲𝘃𝗶𝗲𝘄𝗶𝗻𝗴 𝗮𝘀 𝗮 𝗗𝗼𝗰𝘂𝗺𝗲𝗻𝘁-𝗴𝗿𝗼𝘂𝗻𝗱𝗲𝗱 𝗗𝗶𝗮𝗹𝗼𝗴𝘂𝗲”, we ask how AI can support meta-reviewers 𝗶𝗻 𝘁𝗵𝗲 𝗱𝗲𝗰𝗶𝘀𝗶𝗼𝗻 𝗽𝗿𝗼𝗰𝗲𝘀𝘀 — 𝗻𝗼𝘁 𝗷𝘂𝘀𝘁 𝗶𝗻 𝘄𝗿𝗶𝘁𝗶𝗻𝗴 𝘁𝗵𝗲 𝗳𝗶𝗻𝗮𝗹 𝗿𝗲𝗽𝗼𝗿𝘁.

Diagram illustrating how AI analyzes peer review discussions. A reviewer asks whether reviewers disagree on the strengths of a paper. The system summarizes that reviewers 1 and 2 both evaluate the method but differ in their assessment, with one seeing it as well-designed and the other noting issues. Another question asks whether other reviewers consider these issues important; reviewer 3 mentions them but treats them as minor concerns. Side labels indicate steps of correlating opinions and weighting arguments.

120

UKP Lab @ukplab.bsky.social · 13/03/2026

🤔 𝗗𝗶𝗱 𝘆𝗼𝘂𝗿 𝗟𝗟𝗠 𝗿𝗲𝗮𝗹𝗹𝘆 𝗳𝗼𝗿𝗴𝗲𝘁 𝘆𝗼𝘂𝗿 𝗽𝗿𝗶𝘃𝗮𝘁𝗲 𝗱𝗮𝘁𝗮 - 𝗼𝗿 𝗷𝘂𝘀𝘁 𝗴𝗲𝘁 𝗾𝘂𝗶𝗲𝘁𝗲𝗿 𝗮𝗯𝗼𝘂𝘁 𝗶𝘁? 🧠 Machine unlearning aims to remove specific training data (e.g., private info) without full retraining. But does it actually 𝗿𝗲𝗺𝗼𝘃𝗲 𝘁𝗵𝗲 𝗸𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝗳𝗿𝗼𝗺 𝘁𝗵𝗲 𝗺𝗼𝗱𝗲𝗹?

Three-panel diagram explaining an audit method for LLM unlearning.
Panel A, “LLM Unlearning,” shows a trained language model and a modified “unlearned” model, highlighting that output-based metrics may give a misleading sense of privacy.
Panel B, “Our Audit,” illustrates comparing the two models using a Venn diagram to decompose mutual information into unlearned information and residual information.
Panel C, “Practical Impact,” shows that residual information can correlate with vulnerability to adversarial attacks, and suggests inference-time privacy protection through abstention.

121

UKP Lab @ukplab.bsky.social · 12/03/2026

🧠🔗 How can we link semantically related sentences across documents - 𝙬𝙞𝙩𝙝 𝙯𝙚𝙧𝙤 𝙡𝙖𝙗𝙚𝙡𝙚𝙙 𝙙𝙖𝙩𝙖? Meet 𝗔𝗕𝗖𝗗-𝗟𝗜𝗡𝗞 🚀

Three-step workflow diagram for evaluating retrieval approaches.
Step 1, “Data generation and validation,” shows synthetic data being aligned with natural data across two domains: peer reviews and news.
Step 2, “Automatic evaluation,” compares multiple approaches (10 retrievers and 3 LLMs) and produces a shortlist of the best-performing models.
Step 3, “Assisted labeling,” illustrates human annotators reviewing candidate outputs and marking them as correct or incorrect.
The final result is the selection of the best approach and the creation of labeled data.

110

UKP Lab @ukplab.bsky.social · 11/03/2026

“𝗬𝗼𝘂𝗿 𝗽𝗮𝗽𝗲𝗿 𝗶𝘀 𝗻𝗼𝘁 𝗻𝗼𝘃𝗲𝗹 𝗲𝗻𝗼𝘂𝗴𝗵.”🙅 We have all seen this infamous comment from Reviewer #2. But 𝗵𝗼𝘄 𝗿𝗲𝗹𝗶𝗮𝗯𝗹𝗲 𝗶𝘀 𝘁𝗵𝗲 𝗻𝗼𝘃𝗲𝗹𝘁𝘆 𝗮𝘀𝘀𝗲𝘀𝘀𝗺𝗲𝗻𝘁 𝘁𝗼𝗱𝗮𝘆?🤔

Infographic titled “The Hidden Complexity Behind ‘Not Novel Enough’,” explaining why novelty assessment in peer review is difficult at scale. The graphic shows a researcher surrounded by stacks of papers and question marks, illustrating uncertainty in evaluating novelty. On the left, challenges include fragmented literature, reviewer overload (14+ papers per year), and unclear novelty criteria. On the right, resulting issues include inconsistent decisions, author confusion, and weak scientific feedback. The infographic concludes that novelty assessment requires searching, comparing, and judging prior work under severe time constraints.

101

UKP Lab @ukplab.bsky.social · 10/03/2026

🚨 Introducing 𝗚𝗿𝗶𝘁𝗛𝗼𝗽𝗽𝗲𝗿, the new State-of-the-Art Multi-Hop Dense Retriever 🦗 Current approaches to multi-hop retrieval face critical trade-offs. 🔎 Decomposition-based methods break complex queries into simpler steps, but they are computationally expensive and difficult to train end-to-end.

110

UKP Lab @ukplab.bsky.social · 05/03/2026

🎓 𝗗𝗶𝘀𝘀𝗲𝗿𝘁𝗮𝘁𝗶𝗼𝗻 𝗗𝗲𝗳𝗲𝗻𝘀𝗲 – 𝗖𝗲𝗰𝗶𝗹𝗶𝗮 Congratulations to @ccliu.bsky.social on the successful PhD defense at the UKP Lab of @tuda.bsky.social In her thesis 𝘎𝘦𝘯𝘦𝘳𝘢𝘭𝘪𝘻𝘢𝘵𝘪𝘰𝘯 𝘢𝘯𝘥 𝘈𝘥𝘢𝘱𝘵𝘢𝘵𝘪𝘰𝘯 𝘰𝘧 𝘓𝘢𝘯𝘨𝘶𝘢𝘨𝘦 𝘔𝘰𝘥𝘦𝘭𝘴 𝘵𝘰 𝘓𝘢𝘯𝘨𝘶𝘢𝘨𝘦𝘴 𝘢𝘯𝘥 𝘊𝘶𝘭𝘵𝘶𝘳𝘦𝘴, Cecilia investigates ...

Smiling person standing in a doorway wearing a handmade graduation cap decorated with stickers and photos. They hold the cap with one hand, carry a tote bag over their shoulder, and hold a smartphone in the other hand. A hallway with chairs and large windows is visible in the background.

150

UKP Lab @ukplab.bsky.social · 04/03/2026

🎓 𝗔𝗜 𝘁𝗼 𝘀𝘂𝗽𝗽𝗼𝗿𝘁 𝗰𝗼𝘂𝗿𝘀𝗲 𝗶𝗻𝘀𝘁𝗿𝘂𝗰𝘁𝗼𝗿𝘀 𝗻𝗼𝘁 𝘁𝗼 𝗿𝗲𝗽𝗹𝗮𝗰𝗲 𝘁𝗵𝗲𝗺! In 𝘄𝗶𝗻𝘁𝗲𝗿 𝘀𝗲𝗺𝗲𝘀𝘁𝗲𝗿 𝟮𝟬𝟮𝟱/𝟮𝟲, our AI-based system 𝗟𝗟𝗠𝗲𝗻𝘁𝗼𝗿 has supported teaching the course 𝘐𝘯𝘵𝘳𝘰𝘥𝘶𝘤𝘵𝘪𝘰𝘯 𝘵𝘰 𝘚𝘤𝘪𝘦𝘯𝘵𝘪𝘧𝘪𝘤 𝘞𝘰𝘳𝘬 at @cs-tudarmstadt.bsky.social

University classroom scene in which students sit at desks while a lecturer presents at the front. One student raises a hand while interacting with a large transparent digital interface showing a document and evaluation tools for academic writing, illustrating AI-assisted feedback and assessment during a seminar.

110

UKP Lab @ukplab.bsky.social · 26/02/2026

📣🧪 𝗡𝗲𝘄 𝗿𝗲𝘀𝘂𝗹𝘁𝘀: 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗻𝗴 𝗦𝗰𝗶𝗲𝗻𝘁𝗶𝗳𝗶𝗰 𝗪𝗿𝗶𝘁𝗶𝗻𝗴 𝘄𝗶𝘁𝗵 𝗘𝘅𝗽𝗲𝗿𝘁 𝗣𝗿𝗲𝗳𝗲𝗿𝗲𝗻𝗰𝗲𝘀 👉 Today we are presenting our first two papers that build on HAI-Co2, our recent position paper on human-ai text co-construction in expert domains:

Poster titled “Reward Modeling for Scientific Writing Evaluation & Expert Preference Based Evaluation of Automated Related Work Generation,” with authors Furkan Sahinuç, Subhabrata Dutta, and Iryna Gurevych. The graphic illustrates a workflow: cited papers are used by an author to create a gold standard related work section and evaluation criteria, while AI generators produce draft related work sections. An LLM-supported evaluation system ranks outputs based on expert preferences, with feedback loops to improve generation quality.

100

UKP Lab @ukplab.bsky.social · 24/02/2026

🎓 𝗣𝗵𝗗 𝗱𝗲𝗳𝗲𝗻𝘀𝗲 𝗰𝗼𝗻𝗴𝗿𝗮𝘁𝘂𝗹𝗮𝘁𝗶𝗼𝗻𝘀 𝘁𝗼 𝗔𝗻𝗱𝗿𝗲𝗮𝘀 𝗪𝗮𝗹𝗱𝗶𝘀! Congratulations to @tresiwald.bsky.social on the successful defense of his dissertation “𝘈𝘯 𝘐𝘯𝘵𝘦𝘨𝘳𝘢𝘭 𝘝𝘪𝘦𝘸 𝘰𝘯 𝘵𝘩𝘦 𝘙𝘦𝘭𝘪𝘢𝘣𝘪𝘭𝘪𝘵𝘺 𝘰𝘧 𝘓𝘢𝘯𝘨𝘶𝘢𝘨𝘦 𝘔𝘰𝘥𝘦𝘭𝘴 𝘧𝘰𝘳 𝘊𝘰𝘮𝘱𝘶𝘵𝘢𝘵𝘪𝘰𝘯𝘢𝘭 𝘈𝘳𝘨𝘶𝘮𝘦𝘯𝘵𝘢𝘵𝘪𝘰𝘯”, held at Technische Universität Darmstadt on February 17th, 2026.

1112

UKP Lab @ukplab.bsky.social · 19/02/2026

🔎🧩 𝗕𝗲𝘆𝗼𝗻𝗱 𝗕𝗲𝗻𝗰𝗵𝗺𝗮𝗿𝗸𝘀: 𝗛𝗼𝘄 𝘁𝗼 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗲 𝗠𝗲𝗻𝘁𝗮𝗹 𝗛𝗲𝗮𝗹𝘁𝗵 𝗔𝗜 𝗥𝗲𝘀𝗽𝗼𝗻𝘀𝗶𝗯𝗹𝘆 AI for mental health is a high-stakes area: its evaluation needs to meet the highest expectations. The new preprint 𝘙𝘦𝘴𝘱𝘰𝘯𝘴𝘪𝘣𝘭𝘦 𝘌𝘷𝘢𝘭𝘶𝘢𝘵𝘪𝘰𝘯 𝘰𝘧 𝘈𝘐 𝘧𝘰𝘳 𝘔𝘦𝘯𝘵𝘢𝘭 𝘏𝘦𝘢𝘭𝘵𝘩, written by an interdisciplinary team spanning AI [...]

Table titled “Taxonomy for evaluation of AI in mental health applications,” organized into columns for quality criteria (validity and reliability) and real-world use (implementation and maintenance). Rows distinguish support types: assessment, intervention, and information synthesis. Each cell lists detailed evaluation questions, such as construct and criterion validity, consistency across populations and time, feasibility, effectiveness, usability, acceptability, safety, and unintended consequences, providing a structured framework for assessing AI systems in mental health contexts.

133

UKP Lab @ukplab.bsky.social · 18/02/2026

⚕️ 𝗖𝗵𝗮𝘁𝗯𝗼𝘁𝘀 𝗳𝗼𝗿 𝗵𝗲𝗮𝗹𝘁𝗵 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻𝘀: 𝘄𝗵𝗲𝗿𝗲 𝗱𝗼𝗲𝘀 𝗰𝗼𝗺𝗺𝘂𝗻𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗯𝗿𝗲𝗮𝗸 𝗱𝗼𝘄𝗻? In a new briefing by @sciencemediacenter.de, Prof. Dr. @igurevych.bsky.social (UKP Lab, @tuda.bsky.social) highlights why the gap between benchmarks and real-world use matters: 𝗯𝗲𝗻𝗰𝗵𝗺𝗮𝗿𝗸𝘀 𝗮𝗿𝗲 𝗼𝗳𝘁𝗲𝗻 𝘀𝗶𝗺𝗽𝗹𝗶𝗳𝗶𝗲𝗱 𝗮𝗻𝗱 𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲𝗱.

160

UKP Lab @ukplab.bsky.social · 13/02/2026

🔍 𝗗𝗼 𝘆𝗼𝘂 𝘄𝗮𝗻𝘁 𝘁𝗼 𝗶𝗺𝗽𝗿𝗼𝘃𝗲 𝗺𝘂𝗹𝘁𝗶-𝘀𝘁𝗲𝗽 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗶𝗻 𝗟𝗟𝗠𝘀 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗰𝗼𝗺𝗽𝗹𝗲𝘅 𝗲𝘅𝘁𝗲𝗿𝗻𝗮𝗹 𝗽𝗶𝗽𝗲𝗹𝗶𝗻𝗲𝘀? ⚠️ Step-level Process Supervision is great! 𝗕𝗨𝗧: 👩‍⚖️ Human Labelling is slow and expensive 🌿 Tree-search methods are computationally intensive ⏮️ Past methods use fragmented pipelines with limited step alignment

Research poster for the paper “SPARE: Single-Pass Annotation with Reference-Guided Evaluation for Automatic Process Supervision and Reward Modelling,” featuring logos of UKP Lab, hessian.AI, and Queen’s University. The poster lists the authors and affiliations, and includes a photo of a researcher standing beside the poster at a conference. At the bottom, a diagram explains the SPARE framework, showing inputs such as prompts and reference solutions, and outputs related to evaluation and alignment.

111

UKP Lab @ukplab.bsky.social · 04/02/2026

📢 𝗨𝗞𝗣 𝗟𝗮𝗯 𝗮𝘁 𝗜𝗖𝗟𝗥𝟮𝟬𝟮𝟲 📢 Happy to share that our paper has been accepted to @iclr-conf.bsky.social 🎉 📜 𝚁𝚎𝚟𝚎𝚕𝚊: 𝗗𝗲𝗻𝘀𝗲 𝗥𝗲𝘁𝗿𝗶𝗲𝘃𝗲𝗿 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝘃𝗶𝗮 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹𝗶𝗻𝗴 👥 Fengyu Cai, Tong Chen, @xinranz3.bsky.social, Sihao Chen, Hongming Zhang, @sherrytswu.bsky.social, @igurevych.bsky.social, Heinz Koeppl

230

UKP Lab @ukplab.bsky.social · 03/02/2026

🎓 𝗚𝘂𝗲𝘀𝘁 𝗧𝗮𝗹𝗸 𝗮𝘁 𝘁𝗵𝗲 𝗨𝗞𝗣 𝗟𝗮𝗯 👋 We were pleased to welcome Dr. Chung-Chi Chen, researcher at the National Institute of Advanced Industrial Science and Technology (AIST), for a guest talk at the UKP Lab in Darmstadt.

A speaker stands at the front of a lecture room giving a talk, gesturing with both hands. Behind them, a projected slide titled “Instruction vs. Retrieval – A Subtle Difference” explains risks related to prompt-based attacks and retrieval-based manipulation. The slide includes diagrams about LLM agents, public opinion pools, and coordinated ideological input. A laptop and water bottle sit on the podium beside the presenter.

210

UKP Lab @ukplab.bsky.social · 23/01/2026

⚠️ 𝗖𝗮𝗻 𝗳𝗶𝗻𝗲-𝘁𝘂𝗻𝗶𝗻𝗴 𝗹𝗲𝗮𝗱 𝘁𝗼 𝘂𝗻𝘄𝗮𝗻𝘁𝗲𝗱 𝗯𝗲𝗵𝗮𝘃𝗶𝗼𝘂𝗿𝘀 𝗶𝗻 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗺𝗼𝗱𝗲𝗹𝘀? A recent paper in Nature suggests that even small amounts of targeted fine-tuning data can trigger unexpected and problematic behaviour that generalises beyond the original pre-training task. (1/🧵)

121

UKP Lab @ukplab.bsky.social · 21/01/2026

🎓 𝗚𝘂𝗲𝘀𝘁 𝗧𝗮𝗹𝗸 𝗮𝘁 𝘁𝗵𝗲 𝗨𝗞𝗣 𝗟𝗮𝗯 👋 We were pleased to welcome Fajri Koto, Assistant Professor at the MBZUAI (Mohamed bin Zayed University of Artificial Intelligence), for a guest talk at the UKP Lab in Darmstadt. (1/🧵)

A researcher stands in a lecture room giving a presentation, gesturing toward a projected slide titled “AgentFly: Extensible and Scalable Reinforcement Learning for LM Agents.” The slide is shared via a video conferencing interface visible at the top, showing remote participants. The presenter stands next to a laptop on a small podium, with the projection filling the wall behind them.

100

UKP Lab @ukplab.bsky.social · 07/01/2026

🎉 𝗨𝗞𝗣 𝗟𝗮𝗯 𝗮𝘁 𝗘𝗔𝗖𝗟 𝟮𝟬𝟮𝟲 🌍 We are pleased to share that researchers from the UKP Lab and their collaborators will present 𝟵 𝗽𝗮𝗽𝗲𝗿𝘀 𝗮𝘁 𝘁𝗵𝗲 𝗠𝗮𝗶𝗻 𝗖𝗼𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲 at the #EACL2026, one of the leading international conferences in #NLP and #CL, bringing together research from across Europe and beyond.

110

UKP Lab @ukplab.bsky.social · 19/12/2025

📸 Here are a few impressions from the ATHENE Center event 𝘊𝘺𝘣𝘦𝘳𝘯𝘢𝘵𝘪𝘰𝘯 𝘋𝘦𝘶𝘵𝘴𝘤𝘩𝘭𝘢𝘯𝘥: 𝘙𝘰𝘢𝘥𝘮𝘢𝘱 𝘵𝘰 𝘐𝘯𝘵𝘦𝘳𝘯𝘦𝘵 𝘚𝘦𝘤𝘶𝘳𝘪𝘵𝘺 in Frankfurt, where experts from academia, industry and policy came together to discuss how internet security can be strengthened in the age of AI 🤖.

110

UKP Lab @ukplab.bsky.social · 19/12/2025

🎤 Prof. Iryna Gurevych, Distinguished Professor at the National Center for Cybersecurity @athenecenter.bsky.social and Director of the UKP Lab at @tuda.bsky.social, delivered a keynote on “How to Make AI-Native Internet Content Secure? Coping with Synthetic and Misleading Data.”

141

UKP Lab @ukplab.bsky.social · 16/12/2025

🚀 𝗟𝗮𝘁𝗲𝘀𝘁 𝗣𝗲𝗲𝗿 𝗥𝗲𝘃𝗶𝗲𝘄 𝗗𝗮𝘁𝗮𝘀𝗲𝘁 𝗥𝗲𝗹𝗲𝗮𝘀𝗲 𝗳𝗿𝗼𝗺 𝗔𝗥𝗥 𝟮𝟬𝟮𝟱! tudatalib.ulb.tu-darmstadt.de/handle/tudat... 📊 𝗙𝗶𝗿𝘀𝘁-𝘁𝗶𝗺𝗲 𝗥𝗲𝗹𝗲𝗮𝘀𝗲: 𝗘𝗠𝗡𝗟𝗣 𝟮𝟬𝟮𝟰 𝗡𝗼𝗻-𝗔𝗰𝗰𝗲𝗽𝘁𝗲𝗱 𝗦𝘂𝗯𝗺𝗶𝘀𝘀𝗶𝗼𝗻𝘀: ✅ 𝟯𝗸 papers ✅ 𝟱𝟬𝟬 reviews ✅ 𝟭𝟳𝟱 meta-reviews ✅ 𝟯𝟮𝟵 papers with rebuttals

Stylized illustration of a researcher wearing a graduation cap and lab coat, holding a magnifying glass and reading a document. A small robotic parrot sits on the researcher’s shoulder, suggesting AI assistance. The image combines themes of academic research, analysis, and human–AI collaboration.

100

UKP Lab @ukplab.bsky.social · 10/12/2025

✨ 𝗔 𝘄𝗮𝗿𝗺 𝘄𝗲𝗹𝗰𝗼𝗺𝗲 𝘁𝗼 𝗼𝘂𝗿 𝗻𝗲𝘄 𝗣𝗵𝗗 𝘀𝘁𝘂𝗱𝗲𝗻𝘁, 𝗗𝗮𝘀𝗵𝗮 𝗛𝗿𝘆𝗵𝗼𝗿𝘆𝗲𝘃𝗮 We’re very happy to welcome Dasha Hryhoryeva as a new PhD student at the UKP Lab. Her research focuses on AI for Mental Health and AI for Social Good, with current work on emotion prediction using large language models. (1/🧵)

120

UKP Lab @ukplab.bsky.social · 05/12/2025

💻 𝗪𝗲𝗹𝗰𝗼𝗺𝗲 𝘁𝗼 𝘁𝗵𝗲 𝗨𝗞𝗣 𝗟𝗮𝗯, 𝗕𝗵𝗮𝘃𝘆𝗮𝗷𝗲𝗲𝘁 𝗦𝗶𝗻𝗴𝗵! We’re happy to welcome Bhavyajeet Singh as a new PhD student at the UKP Lab. He recently joined the LLMs for Security project, which is part of the ATHENE Center for Cybersecurity, and works on using large language models to generate secure code. (1/🧵)

120