"Many Are My Names": The Anatomy of the Assistant and Its Personas via Sparse Autoencoders

arXiv:2608.07852 · cs.CL · Submitted 2026-08-14 · Read on arXiv

Adelaide Danilov, Aria Nourbakhsh, Oleksandr Marchenko Breneur, Salima Lamsiyah

cs.CL

Submitted: 2026-08-14

Updated: 2026-08-17

Comments: 38 pages, 9 tables, 4 figures, 2 listings

Code: https://github.com/saprmarks/dictionary_learning

License: http://creativecommons.org/licenses/by/4.0/

The gist: How a language model internally represents who is speaking, the Assistant, an assigned roleplay persona, or a narrated story character, remains underexplored.

Terminology

Steering — 106× in this paper · explained in 2 episodes
Core — 23× in this paper · explained in 2 episodes
Actor — 17× in this paper · explained in 3 episodes
Immersive Simulation Mode — 15× in this paper
Adam — 7× in this paper · explained in 3 episodes
Transformer — 5× in this paper · explained in 2 episodes
Ablation — 3× in this paper · explained in 1 episode
Accuracy — 3× in this paper · explained in 1 episode
Commentary — 3× in this paper · explained in 2 episodes
Composition — 3× in this paper · explained in 2 episodes
Consistency — 3× in this paper · explained in 3 episodes
Decoder — 3× in this paper · explained in 2 episodes
Feature Selection — 3× in this paper · explained in 4 episodes
LLM judge — 3× in this paper · explained in 2 episodes
Activation Steering — 2× in this paper · explained in 7 episodes
Clarity — 2× in this paper · explained in 3 episodes
Frustration — 2× in this paper · explained in 3 episodes
Attention Sinks — 1× in this paper · explained in 2 episodes
Benchmark — 1× in this paper · explained in 2 episodes
Calibration — 1× in this paper · explained in 9 episodes
Cosine Similarity — 1× in this paper · explained in 3 episodes
Faithfulness — 1× in this paper · explained in 2 episodes
False Negatives — 1× in this paper · explained in 2 episodes
Feature Extraction — 1× in this paper · explained in 3 episodes
Grounding — 1× in this paper · explained in 3 episodes
Interpretability — 1× in this paper · explained in 15 episodes
Introspection — 1× in this paper · explained in 2 episodes
Model Interpretability — 1× in this paper · explained in 2 episodes
Persona Prompting — 1× in this paper · explained in 2 episodes
Redundancy — 1× in this paper · explained in 2 episodes
Sparse Autoencoders (SAEs) — 1× in this paper · explained in 4 episodes
Transformers — 1× in this paper · explained in 2 episodes

Abstract

How a language model internally represents who is speaking, the Assistant, an assigned roleplay persona, or a narrated story character, remains underexplored. We study speaker representations using a dataset of user-expressed emotional text and corresponding model responses. We decompose three generation settings (Assistant, Roleplay, and Story) into sparse autoencoder features extracted at turn-boundary and pronoun-token positions and selected through a filtering pipeline for different depths. We characterize each surviving feature through its steering effects and activation distribution. Our main finding is that the Assistant and roleplay personas are not independent alternatives: personas retain the Assistant-associated feature core while progressively differentiating from it across layers, starting from operational machinery towards behavioral and stylistic features. Meanwhile, generated story characters lack the Assistant-associated core. Both Story and Roleplay can be distinguished from the Assistant with Immersive Simulation Mode. However, the Assistant can sometimes enter or slowly drift into it even in the default setting.

Terminology

Abstract

Related papers