25 citations · 25 across the 4 of their papers we have counts for
3 papers · 1 filter
LLMs program using Reusable and Composable Abstract Representations
Casper L. Christensen, Kyle J. Ray, Dmitry Manning-Coe +2
When writing code, the same computation can be expressed in different programming languages and through different implementation strategies. We show that the activations of LLMs ca…
Interactions Between Crosscoder Features: A Compact Proofs Perspective
Dmitry Manning-Coe, Thomas Read, Anna Soligo +4
Dictionary learning methods like Sparse Autoencoders (SAEs) and crosscoders attempt to explain a model by decomposing its activations into independent features. Interactions betwee…
Decomposing momentum scales in the Hubbard Model: From Hatsugai-Kohmoto to Aubry-André
Dmitry Manning-Coe, Barry Bradlyn
The all-to-all momentum coupling of the Hubbard interaction makes interacting lattice models generically unsolvable. In many settings, however, from Peierls instabilities to Moiré…