Mozes Jacobs

Mozes Jacobs is a fourth-year PhD candidate in computer science at Harvard University, where he is a Kempner Institute Graduate Fellow advised by Professor Demba Ba. He works on mechanistic interpretability and representation learning, with a focus on the geometry and content of neural representations in vision and generative models. He served as a Research Fellow at Goodfire from May to September 2026 and is a co-lead author on the Block-Sparse Featurizers paper, which introduces methods for decomposing model activations into multidimensional interpretable subspaces. He previously earned his BS in computer science from the University of Washington and developed the HyperSINDy framework for modeling stochastic dynamics at the NSF AI Institute in Dynamic Systems.