q-fin.GNAug 29, 2025

A Financial Brain Scan of the LLM

Authors: Hui Chen, Antoine Didisheim, Mo Pourmohammadi, Luciano Somoza, Hanqing Tian

Organizations: MIT Sloan and NBER. · University of Melbourne. · Yale SOM. · ESSEC Business School.

Abstract

We use sparse autoencoders (SAEs) to "brain scan" large language models, mapping internal activations to interpretable concepts and enabling targeted "steering" of specific features. We show that prompting models to adopt specific preferences activates many unintended concepts, complicating causal inference. In contrast, SAEs permit smooth, more isolated steering of target concepts. Furthermore, our scans reveal a stark disconnect between the model's internal feature contributions and its self-generated explanations, showing that these explanations are unreliable guides to its reasoning. Finally, SAEs provide a lightweight intervention for existing models, producing interpretable embeddings and allowing researchers to attenuate specific economic biases.

Figures & tables

Explore similar work

CardsList
  1. SAEExplainer: Interpreting SAE Features with Activation-Guided Preference Optimization

    Jun 7, 2026Jingyi He, Haiyan Zhao, Ruxue Shi +4LLM InterpretabilitySparse Autoencoders