A Financial Brain Scan of the LLM

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Chen, Hui, Didisheim, Antoine, Mohammad, Pourmohammadi, Somoza, Luciano, Tian, Hanqing
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910021799903232
author Chen, Hui
Didisheim, Antoine
Mohammad
Pourmohammadi
Somoza, Luciano
Tian, Hanqing
author_facet Chen, Hui
Didisheim, Antoine
Mohammad
Pourmohammadi
Somoza, Luciano
Tian, Hanqing
contents Emerging techniques in computer science make it possible to "brain scan" large language models (LLMs), identify the plain-English concepts that guide their reasoning, and steer them while holding other factors constant. We show that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance. We also show that models can be steered to be more or less risk-averse, optimistic, or pessimistic, which allows researchers to correct or simulate biases. The method is transparent, lightweight, and replicable for empirical research in the social sciences.
format Preprint
id arxiv_https___arxiv_org_abs_2508_21285
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Financial Brain Scan of the LLM
Chen, Hui
Didisheim, Antoine
Mohammad
Pourmohammadi
Somoza, Luciano
Tian, Hanqing
General Finance
Artificial Intelligence
Computational Engineering, Finance, and Science
General Economics
Economics
Emerging techniques in computer science make it possible to "brain scan" large language models (LLMs), identify the plain-English concepts that guide their reasoning, and steer them while holding other factors constant. We show that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance. We also show that models can be steered to be more or less risk-averse, optimistic, or pessimistic, which allows researchers to correct or simulate biases. The method is transparent, lightweight, and replicable for empirical research in the social sciences.
title A Financial Brain Scan of the LLM
topic General Finance
Artificial Intelligence
Computational Engineering, Finance, and Science
General Economics
Economics
url https://arxiv.org/abs/2508.21285