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

Research Summary

Contribution Summary

• 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.

Why It Matters For Eval

• Among the characteristics of value representation in humans is that they distinguish among value of different kinds.
• Specifically, both grammatical and economic valuation was found to be overly influenced by moral value, relative to human norms.

Related Papers

• PoLi-RL: A Point-to-List Reinforcement Learning Framework for Conditional Semantic Textual Similarity Embedding Neighbor
• BabyLM Turns 4 and Goes Multilingual: Call for Papers for the 2026 BabyLM Workshop Embedding Neighbor
• ABCD: All Biases Come Disguised Embedding Neighbor
• Large Language Models and Impossible Language Acquisition: "False Promise" or an Overturn of our Current Perspective towards AI Embedding Neighbor
• Information-Theoretic Storage Cost in Sentence Comprehension Embedding Neighbor
• Predicting Subway Passenger Flows under Incident Situation with Causality Embedding Neighbor
• Discrete Stochastic Localization for Non-autoregressive Generation Embedding Neighbor
• Power and Limitations of Aggregation in Compound AI Systems Embedding Neighbor
• Airavat: An Agentic Framework for Internet Measurement Embedding Neighbor
• The Headless Firm: How AI Reshapes Enterprise Boundaries Tag Overlap
• Measuring AI Ability to Complete Long Software Tasks Tag Overlap
• Residual Connections and the Causal Shift: Uncovering a Structural Misalignment in Transformers Tag Overlap