Liv Gorton, an Anthropic researcher specializing in mechanistic interpretability, highlights confusion over definitions of the linear representation hypothesis in sparse autoencoder and superposition research

— Aryaman Arora replies that researchers must document assumptions more clearly.

Original post

OPLI#1445Liv|@LIVGORTON

I do genuinely feel there's been some wires crossed terribly at some point. I am not sure of the source of people doing SAEs / LRH-based work thought features would all be 1D (even in TMS). This is me fishing for the references tbc!

3:58 PM · May 21, 2026 View on X

REPLYAA#678Aryaman Arora|@ARYAMAN2020

@livgorton i think the lack of formalisation is worse if anything. you shouldn’t have to guess what people were thinking research-wise, they should be writing such things down

LILiv@livgorton

I do genuinely feel there's been some wires crossed terribly at some point. I am not sure of the source of people doing SAEs / LRH-based work thought features would all be 1D (even in TMS). This is me fishing for the references tbc!

10:58 PM · May 21, 2026 · 3.8K Views

4:14 PM · May 22, 2026 · 154 Views