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

2026-08-07

AI Summary

This study investigates how large language models internally represent distinct speaker identities—such as the default assistant persona, role-played personas, and narrative characters—and examines their similarities and differences. By constructing a user–model dialogue dataset and applying sparse autoencoders to extract representations at turn boundaries and pronoun positions, the authors design a multi-stage filtering framework to analyze representational structures under controlled generation settings. The findings reveal that the assistant and role-played personas share a common core representation that progressively diverges across layers, whereas narrative characters lack this shared core. The work introduces an “immersive simulation mode” that effectively discriminates between generation types, successfully identifies the assistant’s core features and their evolution during role-playing, and demonstrates that the model may spontaneously enter an immersive state even in default generation.

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.

Problem

• speaker representation
• language model
• personas
• roleplay
• story characters

Innovation

• Sparse Autoencoders
• Speaker Representation
• Roleplay Persona
• Immersive Simulation Mode
• Feature Decomposition