PuSH - Publikationsserver des Helmholtz Zentrums München: Sparse autoencoders reveal temporal difference learning in large language models.

Informationen

Hinweise zu Qualitätskriterien von Zeitschriften

Open-Access-Richtlinie der Helmholtz-Gemeinschaft 2016

CC Licencences

Metriken für Publikationen

Navigation

Startseite

English

Recherche

Erweiterte Suche

Durchblättern nach ...

... HMGU-Autoren/Konsortien

... Organisationsstruktur

... Zeitschriften

... Publikationstypen

... Forschungsdaten

... Arbeitsgruppen

... Erscheinungsjahr

Publikationen im Überblick

Statistik

HGF Fortschrittsbericht

OA Publikationen

Eintragen

Neue Publikation eintragen

Neue Publikation holen aus...

...EVA

Fehlende Publikation melden

Highlights

Suche

Hilfe & Kontakt

Ansprechpartner

Hilfe

Datenschutz

Helmholtz Open Science

Bibliometrische Indikatoren

SHERPA/RoMEO

DOAJ

Export:

Text

Endnote (RIS) BIB

BibTeX

Demircan, C. ; Saanum, T.* ; Jagadish, A.K. ; Binz, M. ; Schulz, E.

Sparse autoencoders reveal temporal difference learning in large language models.

In: (13th International Conference on Learning Representations Iclr 2025, 24 - 28 April 2025, Singapur). 2025. 4972-4997 (13th International Conference on Learning Representations Iclr 2025)

Verlagsversion

Abstract
Metriken
Zusatzinfos

In-context learning, the ability to adapt based on a few examples in the input prompt, is a ubiquitous feature of large language models (LLMs). However, as LLMs' in-context learning abilities continue to improve, understanding this phenomenon mechanistically becomes increasingly important. In particular, it is not well-understood how LLMs learn to solve specific classes of problems, such as reinforcement learning (RL) problems, in-context. Through three different tasks, we first show that Llama 3 70B can solve simple RL problems in-context. We then analyze the residual stream of Llama using Sparse Autoencoders (SAEs) and find representations that closely match temporal difference (TD) errors. Notably, these representations emerge despite the model only being trained to predict the next token. We verify that these representations are indeed causally involved in the computation of TD errors and Q-values by performing carefully designed interventions on them. Taken together, our work establishes a methodology for studying and manipulating in-context learning with SAEs, paving the way for a more mechanistic understanding.