Author

Andrew M. Saxe

h-index 4 View on Semantic Scholar

We have 1 of 8 papers

We haven’t gathered this author’s papers yet. Follow them and we’ll fetch their work.

Not the right person? Other researchers publish under this name.

Jul 2026

How are linear representations learned? Exact solutions to the dynamics of abstraction

A striking attenuation law is proved: both nonlinearities weaken abstraction in activations relative to preactivations, and the theory is applied to improve linear probe generalization in LLMs.

William Yang, Andrew M. Saxe, Peter E. Latham · arXiv.org · 0 citations