AI ALIGNMENT FORUM
AF

151
Wikitags

Comp-In-Sup

Edited by Linda Linsefors, et al. last updated 6th Oct 2025

Computation in Superposition (Comp-in-Sup) is a sub-field of Mechanistic Interpretability (Mech-Interp).

Superposition in this context means storing information in an overcompleet basis. Comp-in-Sup is the study of how (and if) neural networks performs computation, using information stored in overcomplete basis. The reason we want to do this, is because this information could inform better feature extraction methods.

Subscribe
Discussion
Subscribe
Discussion
Posts tagged Comp-In-Sup
90Toward A Mathematical Framework for Computation in Superposition
Dmitry Vaintrob, jake_mendel, Kaarel
2y
8
55Circuits in Superposition: Compressing many small neural networks into one
Lucius Bushnaq, jake_mendel
1y
1
33Circuits in Superposition 2: Now with Less Wrong Math
Linda Linsefors, Lucius Bushnaq
3mo
0
Add Posts