10 (zehn !!1!) Jahre ist der NR-Talk von @spielkamp.bsky.social , @nettwerkerin.bsky.social, @kerstingaiml.bsky.social und mir her, um über Algorithmen & Kontrolle zu reden. Es war einer der Impulse für @algorithmwatch.org, bei mir Anfang eines kleinen ;-) Umwegs zur KI. But: Finally there!

EurIPS is coming! 📣 Mark your calendar for Dec. 2-7, 2025 in Copenhagen 📅 EurIPS is a community-organized conference where you can present accepted NeurIPS 2025 papers, endorsed by @neuripsconf.bsky.social and @nordicair.bsky.social and is co-developed by @ellis.eu eurips.cc

📢 Present your NeurIPS paper in Europe! Join EurIPS 2025 + ELLIS UnConference in Copenhagen for in-person talks, posters, workshops and more. Registration opens soon; save the date: 📅 Dec 2–7, 2025 📍 Copenhagen 🇩🇰 🔗eurips.cc #EurIPS @euripsconf.bsky.social‬

Can concept-based models handle complex, object-rich images? We think so! Meet Object-Centric Concept Bottlenecks (OCB) — adding object-awareness to interpretable AI. Led by David Steinmann w/ @toniwuest.bsky.social & @kerstingaiml.bsky.social . 📄 arxiv.org/abs/2505.244... #AI #XAI #NeSy #CBM #ML

@kerstingaiml.bsky.social (TU Darmstadt), Frauke Kreuter (LMU Munich, MCML, @zuseschoolrelai.bsky.social ), Mira Mezini (TU Darmstadt), @igurevych.bsky.social (TU Darmstadt, @athenecenter.bsky.social), [...] (5/🧵)

@kerstingaiml.bsky.social ist Professor für KI und Maschinelles Lernen am @cs-tudarmstadt.bsky.social der @tuda.bsky.social und leitet dort das Artificial Intelligence and Machine Learning Lab (AIML).

🙏 for the invitation to give a talk at the Causal Neuro-Symbolic AI (Causal-NeSy) Workshop at ESWC. Integrating causality, neuro-symbolic reasoning, and machine learning to build more robust, interpretable, and generalizable AI systems is so important.

»Diagnostic Reasoning in Natural Language: Computational Model and Application« by @nilsdy.bsky.social, Matej Zečević, Ilia Kuznetsov, @kerstingaiml.bsky.social, @igurevych.bsky.social 📄 arxiv.org/abs/2409.05367 (8/🧵)

🚀 TU Darmstadt leads the new Cluster of Excellence "Reasonable AI" – advancing trustworthy, efficient & adaptive AI grounded in common sense. A big thank you to the entire team, the university, and the state of Hesse for their tremendous work & support!

👉 Multilingual Text-to-Image Generation Magnifies Gender Stereotypes and Prompt Engineering May Not Help You Joint work with folks from TU Darmstadt, TU Munich, @kathaem.bsky.social, @jlibovicky.bsky.social, @mbrackaiml.bsky.social, @kerstingaiml.bsky.social arxiv.org/abs/2401.16092

Wer tiefer einsteigen will: Im Podcast „Eine letzte Chance“ spricht @thinkBTO über das Digitalministerium – mit Bezug zu unserem Whitepaper zur digitalen Staatlichkeit. Hören lohnt sich! 🎧 👉 podcasts.apple.com/de/podcast/b...

Ich freue mich, in der aktuellen Beyond the Obvious-Kolumne von @thinkBTO im @handelsblatt zum neuen Digitalministerium aufzutauchen. Digitalisierung braucht endlich Rückenwind – jetzt kommt es auf kluge Umsetzung und echten Gestaltungswillen an. 👉

Excited to share that our paper got accepted at #ICML2025!! 🎉 We challenge Vision-Language Models like OpenAI’s o1 with Bongard problems, classic visual reasoning challenges and uncover surprising shortcomings. Check out the paper: arxiv.org/abs/2410.19546 & read more below 👇

Consider following the authors Indraneil Paul (@ineil77.bsky.social), Haoyi Yang, Goran Glavaš, Kristian Kersting (@kerstingaiml.bsky.social) & Iryna Gurevych (@igurevych.bsky.social) if you are interested in more information or an exchange of ideas. (7/7) See you this week in Singapore! #ICLR2025

I'm excited to present our spotlight on meta-causal models at #ICLR2025 next week. We model evolving causal graphs in dynamic systems. Applications to inference and attribution of agent actions. Paper: openreview.net/forum?id=J9V... Visit our poster #441 during the Sat 3pm session.

»ObscuraCoder: Powering Efficient Code LM Pre-Training Via Obfuscation Grounding« by Indraneil Paul (@ineil77.bsky.social), Haoyi Yang, Goran Glavaš, Kristian Kersting (@kerstingaiml.bsky.social) & Iryna Gurevych (@igurevych.bsky.social) 📄https://openreview.net/forum?id=VYvxrD7aS0 (3/🧵)

xLSTM for time series with Granger causality: arxiv.org/abs/2502.09981 xLSTM again shows superb performance at time series analysis. "Our experimental evaluations on three datasets demonstrate the overall efficacy of our proposed GC-xLSTM model."

🚨 Are you a student passionate about innovation, technology, and building connections? This is your chance to attend leading events connecting female researchers in computer science and engineering 🧳

Doch, wir müssen uns Sorgen machen – aber können handeln 💪 DeepSeek-äquivalente Modelle sind nur mit Investitionen in Teams, Infrastruktur & Daten möglich. Ich freue mich daher über den Austausch mit dem tollen Fraunhofer-Teams der bereits startet 🚀

🚀 Exakt, wir sind bereit, die nächste Stufe im KI-Ausbau zu zünden! TU Darmstadt, hessian.AI & viele weitere Standorte in 🇩🇪 treiben Forschung & Anwendung voran. Zusammen, auch mit der Industrie, kann viel passieren.

ChatGPT-Entwickler: Open AI eröffnet erstes Büro in Deutschland via @faznet

🍰Super proud of our newest work: “The Cake that is Intelligence and Who Gets to Bake it: An AI Analogy and its Implications for Participation” - arxiv.org/abs/2502.03038 🍰We extend Yann LeCun’s AI cake analogy to relate socio-technical outcomes to the AI life-cycle! More details below ⬇️ 1/4

Kahneman und die KI: Schnelles Denken, langsames Denken via @faznet

🚀 Meet NeST, the first neuro-symbolic transpiler! It converts SPLL, a novel probabilistic language, into code across #AI ecosystems like PyTorch & Julia. Unlock advanced reasoning & learning! #NeuroSymbolic #AI #ML 🚨 Int. J. of Approx. Reasoning 👉 https://doi.org/10.1016/j.ijar.2025.109369

Der Informatiker @kerstingaiml.bsky.social (TU Darmstadt / @dfki.bsky.social) ordnet im FAZ-Podcast Digitalwirtschaft #DeepSeek ein und welche Chancen sich für Europa öffnen. overcast.fm/+AANfZj8LS4w

Die KI #Deepseek rüttelt die Tech-Welt auf – wie sie funktioniert und wieso das eine Chance gerade für Deutschland ist, diskutieren @AlaArmbruster und @CarstenKnop mit KI-Professor @kerstingAIML (@TUDarmstadt @Hessian_AI): www.faz.net/podcasts/f-a...

»ObscuraCoder: Powering Efficient Code LM Pre-Training Via Obfuscation Grounding« by Indraneil Paul (@ineil77.bsky.social), Haoyi Yang, Goran Glavaš, Kristian Kersting (@kerstingaiml.bsky.social) & Iryna Gurevych (@igurevych.bsky.social) (3/🧵)

We must never forget

Thrilled to introduce BlendRL— our neuro-symbolic RL that bridges intuitive & symbolic reasoning! 🧠🤖 A step closer to reasoning agents that think & act seamlessly. 🚀 Check out the details in our #ICLR2025 paper! #AI #ReinforcementLearning

Thrilled to share our #ICLR2025 work on Meta-Causal States! 🌟 Causal graphs evolve with dynamic systems & agent actions. We show how to cluster causal models by qualitative behavior, revealing hidden dynamics & emergent relationships 🚀 #Causality #ML https://arxiv.org/abs/2410.13054

⬛ Emerging #AI technologies have immense potential to transform industries. ➡️ Learn More. Download & study white paper: tinyurl.com/2s3pp4et | #WEF2025 @akwyz.com @constanzeholzwarth.bsky.social @kerstingaiml.bsky.social @mitshapingwork.bsky.social @sgenner.bsky.social @rafbuff.bsky.social

Code-LMs need to go beyond surface syntax for better coding. Meet ObscuraCoder at #ICLR2025, pre-trained on 272B tokens with obfuscated code grounding via our ObscuraX dataset (55M source-obfuscated pairs, 7 languages). Stay tuned for the paper & models! #AI #ML #CodeGeneration

New Year. Four weeks. Three multi-university project reviews. One evaluation review. FIVE accepted papers: 3x ICLR, 1x AISTATS, 1 IJAR. 😵‍💫 These are the kind of weeks that remind you why teamwork is so important & coffee exists. Crazy times, but exciting ones, too! 🚀 #AI #ResearchLife

Happy to see our joint work of #researchTUDa, #alpeh_alpha and #DFKI on T-Free to be in the 🚀 AI House Davos 2025. A fantastic line-up of speakers. Here is the link to the recent EMNLP publication on T-Free: https://aclanthology.org/2024.emnlp-main.1217/

Hallo Bluesky! Wir sind die Deutsche Forschungsgemeinschaft. #DFG Als Selbstverwaltungsorganisation der #Wissenschaft in Deutschland fördern wir erkenntnisgeleitete #Forschung. Hier erhaltet Ihr Infos u.a. zu Förderangeboten, Ausschreibungen und Anträgen. Wir freuen uns auf den Austausch!

Aus Gründen teile ich nochmal die Empfehlungen des Wissenschaftsrats zu Geschlechterforschung. Darin betont der WR die Relevanz von #GenderStudies & plädiert u.a. dafür, dass sie verlässliche institutionelle Strukturen erhalten. Wer das Gegenteil will, hat es auf Wissenschaft als Ganzes abgesehen.

Over 60 German universities and research institutions announced their departure from X today.

Klasse! 🤩 Darmstadt gehört zu den Top Fünf der Gründerstädte in Deutschland. 🎉

Wird 2025 das Jahr der KI-Agenten? Was hält 2025 in Sachen #KI für uns bereit? "2025 wird voraussichtlich von spezialisierteren Modellen geprägt sein, die weniger Energie verbrauchen und präziser arbeiten.", so @kerstingaiml.bsky.social! Kompletter Beitrag 👉 www.zdf.de/nachrichten/....

8 up to the minute reasons not to take Altman’s new claim that AGI is near seriously: garymarcus.substack.com/p/sam-altman...

Frohes neues Jahr: Der Vatikan gibt sich KI-Regeln 👇

Zum Abschluss des Jahres 2024. Viel Spaß beim Lesen. Und alles Gute für 2025!

Surprise!

Season's Greetings from Our Learning Machines to Yours! Wishing You a Joyful Holiday Season and an Insightful Start to 2025 🎄🤖✨

Is it fair to say ARC got Winogranded?🤔 Demise of ARC reminded me of this 👇from three years back..

Some of my thoughts on OpenAI's o3 and the ARC-AGI benchmark aiguide.substack.com/p/did-openai...

When building human allied AI systems, it is imperative to think of who is in control. In our blue sky paper at #AAAI, we argue that a deeper understanding of whether human is in the loop or if AI is in the loop is essential! arxiv.org/pdf/2412.14232

Our work on a unified framework for learning PCs will appear in #AAAI our awesome trio of Athresh, Sahil and Saurabh have worked on this really awesome work. We can learn with varied types of rich domain knowledge! arxiv.org/pdf/2405.02413 we have applied this on real adverse pregnancy outcomes data