A Unified Representation Underlying the Judgment of Large Language Models
This addresses the issue of response bias and hallucination in LLMs, providing a mechanistic account that is foundational for understanding AI judgment architectures.
The study tackled the problem of whether judgment in Large Language Models (LLMs) relies on specialized modules or a unified resource, finding that diverse evaluative judgments are computed along a dominant dimension called the Valence-Assent Axis (VAA), which drives a mechanism that subordinates reasoning and can undermine factual accuracy.
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.