The MLT group (Dr. Marius Mosbach, Dr. Simon Ostermann) together with the NMM (Prof. Verena Wolf, Dr. Kevin Baum) and SAINT groups (Prof. Kristian Kersting, Dr. Dominik Hintersdorf) at DFKI Saarbrücken and Darmstadt is recruiting across three complementary research areas:
### 1. PhD Researcher: Mechanistic Interpretability for Safe Agentic AI (SAgA project) Develop fine-grained steering methods to control language model behavior without retraining. We're looking for someone to advance post-hoc controllability through circuit analysis and sparse feature steering. https://jobs.dfki.de/en/vacancy/researcher-phd-student-m-w-d-x-mechanistic-interpretability-for-safety-618588.html ### 2. PostDoc: Multilingual Mechanistic Interpretability (planned SLIMS Project, DFKI-Inria collaboration) First systematic study of reasoning circuits in compact multilingual models. Investigate circuit transfer across scales and many languages using the full mechanistic toolkit. Work directly with the ALMAnaCH team at Inria Paris (Prof. Benoit Sagot, Dr. Djamé Seddah) with planned bilateral research visits and joint workshops. https://jobs.dfki.de/en/vacancy/senior-researcher-postdoc-m-w-d-x-multilingual-mi-618589.html ### 3. PostDoc: AI Safety, Ethics, and Agentic Systems (SAgA project) Develop safe autonomous agents with explicit, inspectable normative constraints. We're seeking candidates with mechanistic interpretability expertise curious about safety, or safety/ethics backgrounds open to steering-based intervention methods. Work across multiple DFKI research groups (NMM, SAINT, MLT) with tailored supervision. https://jobs.dfki.de/en/vacancy/senior-researcher-postdoc-m-w-d-x-ai-safety-ethics-and-agentic-systems-618587.html ### General Information These positions are part of two larger research initiatives at DFKI and beyond: one planned project focused on understanding and leveraging circuit structure for multilingual models (SLIMS), the other on building safe agentic AI through structurally preserved normative reasoning (SAgA). We work on explainable and efficient language processing with a strong focus on rigorous methodology, active publication, and open critical discussion of ideas. We're committed to creating a positive, respectful, supportive working atmosphere where you have space to develop your own scientific identity. SLIMS and SAgA will collaborate in an integrated interdisciplinary team. The positions will work closely together, with mechanistic findings feeding into safety architectures and safety requirements driving interpretability research, creating a coherent research ecosystem where circuit analysis, multilingual NLP, formal ethics, and agentic systems reinforce each other. The PhD and first postdoc position are supervised by Dr. Simon Ostermann within the Multilinguality and Language Technology (MLT) group directed by Dr. Marius Mosbach, with support from the DFKI labs SAINT (Prof. Kristian Kersting, Dr. Dominik Hintersdorf) and NMM (Prof. Verena Wolf, Dr. Kevin Baum). For the second postdoc position on AI safety and agentic systems, we offer flexible supervision across three research groups: the Neuro-Mechanistic Modeling (NMM) group under Prof. Verena Wolf, the Foundations of Systems AI (SAINT) group under Prof. Kristian Kersting, and the MLT group. Candidates prioritizing mechanistic interpretability work primarily with Dr. Simon Ostermann; those emphasizing formal ethics and normative reasoning anchor with Dr. Kevin Baum; and those focusing on agentic system design and task discovery collaborate with Dr. Dominik Hintersdorf. This ensures your mentorship directly supports your research trajectory while you contribute across interpretability, formal ethics, and safe agentic systems. ### Our Labs We are committed to rigorous, methodologically grounded research with an active publication culture. We believe in the open, critical discussion of ideas and we value solid methodology as the foundation for real progress. We've built a lab culture that combines intellectual rigor with genuine support for the people who work here, and we invest in creating the space for you to develop your own scientific identity, while having lots of fun! If you're interested in mechanistic interpretability, multilingual NLP, or safe AI systems (or the convergences between them), we'd like to hear from you. Applications are open until Sep 15. _______________________________________________ Corpora mailing list -- [email protected] https://list.elra.info/mailman3/postorius/lists/corpora.list.elra.info/ To unsubscribe send an email to [email protected]
