Matthias Dellago
Bio
Updated 03/22/26Matthias Dellago is a machine learning researcher who describes himself as a former physicist now focused on the science of deep learning, learning theory, and optimization. He received a Long-Term Future Fund stipend in October 2023 to support his master's thesis and an accompanying paper on mechanistic interpretability of attention mechanisms, with plans to publish on arXiv and release a tool for other researchers. His GitHub projects include a fork of TransformerLens and a project visualizing self-attention as a vector field, both consistent with his focus on mech interp of attention. He holds the title of Guest Researcher (likely at the University of Innsbruck based on online sources) and is active on LessWrong and the Alignment Forum under the handle matthias-dellago.
Community Signal
Updated 03/22/26No endorsements yet.
Links
Updated 03/22/26- Personal Website
- https://matthiasdellago.github.io/
- Twitter / X
- LessWrong
- matthias-dellago
- EA Forum
- -