Tobias Eder

Doctoral Researcher NLP · Social Computing Group · TU Munich

prof_pic.jpg

Munich, Germany

I’m a doctoral researcher in the Social Computing Group at TU Munich, supervised by Prof. Georg Groh. My work sits at the intersection of NLP, multimodal learning, and the social dimensions of online communication with a particular interest in model robustness, trustworthy AI, and understanding where and why current systems fail.

I like working across boundaries: between modalities, between languages, and between academic research and applied problems that benefit from computational rigor. This has led me from content moderation and social media analysis to legal NLP and standardization work on digital media integrity. I’m drawn to questions where the technical challenge is entangled with real-world complexity.

news

Sep 14, 2026 Our paper “Beyond the 5G-COVID moment: the scope of EMF mis- and disinformation on German-language social media” got accepted at the IOP Journal of Radiological Protection as part of the Special Issue “Communicating Risk in a Changing Media Environment”.
Jun 25, 2026 We contributed to the newly published Research Handbook of Public Communication by Edward Elgar Publishing. Our contribution can be found in the subchapter on “More than debunking misinformation: a holistic communication model for public authorities. The case of electromagnetic fields and 5G in Germany”.
Apr 28, 2026 Heading to ACL 2026 in San Diego — presenting “Safer Reasoning Traces: Measuring and Mitigating Chain-of-Thought Leakage in LLMs” at the PrivateNLP workshop on July 3rd.
Apr 08, 2026 Our paper “Beyond Hate: Differentiating Uncivil and Intolerant Speech in Multimodal Content Moderation” was accepted at the CySoc workshop at ICWSM 2026 — come find me in Los Angeles on May 26.
Nov 16, 2025 Presenting our completed COLEX project at the Software Campus Summit in Berlin, December 3–4.

selected publications

  1. safe_traces_2026.png
    Safer Reasoning Traces: Measuring and Mitigating Chain-of-Thought Leakage in LLMs
    Patrick Ahrend, Tobias Eder, Xiyang Yang, and 2 more authors
    arXiv preprint arXiv:2603.05618, 2026
  2. beyond_hate_2026.png
    Beyond Hate: Differentiating Uncivil and Intolerant Speech in Multimodal Content Moderation
    Nils A Herrmann, Tobias Eder, Jingyi He, and 1 more author
    arXiv preprint arXiv:2603.22985, 2026
  3. anchor_2021.png
    Anchor-based bilingual word embeddings for low-resource languages
    Tobias Eder, Viktor Hangya, and Alexander Fraser
    In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers), 2021