Jannik Brinkmann

Jannik Brinkmann is a PhD student in Computer Science at the University of Mannheim, Germany, advised by Christian Bartelt and Paul Swoboda. He holds an M.Sc. in Data Science (with distinction) and a B.Sc. in Computer Science, both from the University of Mannheim. His research focuses on mechanistic interpretability of neural networks, including sparse autoencoders, causal mediation analysis, and the internal mechanisms underlying multi-step reasoning and cross-lingual representations in large language models. He has worked as a visiting researcher in the interpretable neural networks group at Northeastern University under David Bau, and in the ML2 group at NYU under He He. He received a Long-Term Future Fund grant to support part-time interpretability research in collaboration with David Bau and Logan Riggs, resulting in work on improving sparse autoencoder training methods and measuring progress in dictionary learning for language model interpretability.

No endorsements yet.

Personal Website

https://jannik-brinkmann.github.io/

LinkedIn

LinkedIn

Twitter / X

Twitter

LessWrong

-

EA Forum

-