A Unified Representation Underlying the Judgment of Large Language Models

TL;DR

This paper uncovers a unified evaluative architecture in Large Language Models, showing that diverse judgments are encoded along a dominant Valence-Assent Axis that influences reasoning and can lead to biases like hallucinations.

Contribution

It provides evidence for a convergent, domain-general evaluative mechanism in LLMs centered on the Valence-Assent Axis, linking judgment and reasoning control.

Findings

Diverse evaluative judgments are encoded along the Valence-Assent Axis.

The VAA influences the generative process, affecting reasoning and factual accuracy.

Interventions on the VAA demonstrate its role as a control signal in judgment.

Abstract

A central architectural question for both biological and artificial intelligence is whether judgment relies on specialized modules or a unified, domain-general resource. While the discovery of decodable neural representations for distinct concepts in Large Language Models (LLMs) has suggested a modular architecture, whether these representations are truly independent systems remains an open question. Here we provide evidence for a convergent architecture for evaluative judgment. Across a range of LLMs, we find that diverse evaluative judgments are computed along a dominant dimension, which we term the Valence-Assent Axis (VAA). This axis jointly encodes subjective valence ("what is good") and the model's assent to factual claims ("what is true"). Through direct interventions, we demonstrate this axis drives a critical mechanism, which is identified as the subordination of reasoning: the VAA functions as a control signal that steers the generative process to construct a rationale consistent with its evaluative state, even at the cost of factual accuracy. Our discovery offers a mechanistic account for response bias and hallucination, revealing how an architecture that promotes coherent judgment can systematically undermine faithful reasoning.