Value Entanglement: Conflation Between Different Kinds of Good In (Some) Large Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Cho, Seong Hah, Li, Junyi, Leshinskaya, Anna
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911461430788096
author Cho, Seong Hah
Li, Junyi
Leshinskaya, Anna
author_facet Cho, Seong Hah
Li, Junyi
Leshinskaya, Anna
contents Value alignment of Large Language Models (LLMs) requires us to empirically measure these models' actual, acquired representation of value. Among the characteristics of value representation in humans is that they distinguish among value of different kinds. We investigate whether LLMs likewise distinguish three different kinds of good: moral, grammatical, and economic. By probing model behavior, embeddings, and residual stream activations, we report pervasive cases of value entanglement: a conflation between these distinct representations of value. Specifically, both grammatical and economic valuation was found to be overly influenced by moral value, relative to human norms. This conflation was repaired by selective ablation of the activation vectors associated with morality.
format Preprint
id arxiv_https___arxiv_org_abs_2602_19101
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Value Entanglement: Conflation Between Different Kinds of Good In (Some) Large Language Models
Cho, Seong Hah
Li, Junyi
Leshinskaya, Anna
Computation and Language
Artificial Intelligence
Value alignment of Large Language Models (LLMs) requires us to empirically measure these models' actual, acquired representation of value. Among the characteristics of value representation in humans is that they distinguish among value of different kinds. We investigate whether LLMs likewise distinguish three different kinds of good: moral, grammatical, and economic. By probing model behavior, embeddings, and residual stream activations, we report pervasive cases of value entanglement: a conflation between these distinct representations of value. Specifically, both grammatical and economic valuation was found to be overly influenced by moral value, relative to human norms. This conflation was repaired by selective ablation of the activation vectors associated with morality.
title Value Entanglement: Conflation Between Different Kinds of Good In (Some) Large Language Models
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2602.19101