Researcher Presents Spotlight Paper Showing MLP Neurons Support Sparse Circuits

Original post

Aryaman Arora@aryaman2020#748inTech

I'll be at ICML in Korea! Please come talk to me about interpretability and/or to get food. Also I guess I'm finishing my Ph.D. soon...

I'll be presenting this Spotlight paper from my time at @TransluceAI:

Transluce@TransluceAI

Is your LM secretly an SAE?

Most circuit-finding interpretability methods use learned features rather than raw activations, based on the belief that neurons do not cleanly decompose computation. In our new work, we show MLP neurons actually do support sparse, faithful circuits!

1:50 AM · Jul 3, 2026 · 1.5K Views

VIEWS162LIKES1REPLIES1

Aryaman Arora@aryaman2020

I'll also be giving an invited talk at the Mech Interp Workshop https://mechinterpworkshop.com/

Aryaman Arora@aryaman2020

I'll be at ICML in Korea! Please come talk to me about interpretability and/or to get food. Also I guess I'm finishing my Ph.D. soon...

I'll be presenting this Spotlight paper from my time at @TransluceAI:

2h16210

Aryaman Arora@aryaman2020

Oh uh here is the paper arXiv https://arxiv.org/abs/2601.22594

Aryaman Arora@aryaman2020

I'll also be giving an invited talk at the Mech Interp Workshop https://mechinterpworkshop.com/

2h13410