Jannik Brinkmann

I am a Ph.D. student advised by Christian Bartelt. I am also affiliated with the interpretable neural networks group at Northeastern University led by David Bau. I spent the winter semester 2024/2025 as a visiting researcher in the ML2 research group at NYU advised by He He. I interned at J.P. Morgan Research working on post-training of language models for mathematical reasoning.

Google Scholar GitHub Publication List

Research

I am interested in the structure and interpretation of natural and artificial systems. In the past, I studied mechanisms of multi-step reasoning in transformers, cross-lingual feature sharing in language models, and sparse autoencoders as a tool for interpreting language model representations. Beyond understanding models themselves, I am excited about the prospect of discovering useful knowledge in them and teaching it to humans.

Selected Publications

  1. Mechanisms of Protein Folding in ESMFold.

  2. Large Language Models Share Representations of Latent Grammatical Concepts across Typologically Diverse Languages. Nations of the Americas Chapter of the Association for Computational Linguistics (NAACL), 2025. (Oral)

  3. A Mechanistic Analysis of a Transformer Trained on a Symbolic Multi-Step Reasoning Task. Association for Computational Linguistics (ACL), 2024.