# Deep Value Benchmark: Measuring Whether Models Generalize Deep Values or Shallow Preferences

arXiv: https://arxiv.org/abs/2511.02109 · HTML: https://mlsanity.com/paper/2511.02109/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Placed in Human values and AI feedback alignment, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 35 of its 65 references that are in the map.
- First posted on arXiv 3 Nov 2025.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Human values and AI feedback alignment](https://mlsanity.com/alignment/human-values-and-ai-feedback-alignment/)

## Builds on (most cited first, 30 of 35)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [Emergent Abilities of Large Language Models](https://mlsanity.com/paper/2206.07682/) (2022)
- [Concrete Problems in AI Safety](https://mlsanity.com/paper/1606.06565/) (2016)
- [The Rise and Potential of Large Language Model Based Agents: A Survey](https://mlsanity.com/paper/2309.07864/) (2023)
- [A Survey on Evaluation of Large Language Models](https://mlsanity.com/paper/2307.03109/) (2023)
- [A General Language Assistant as a Laboratory for Alignment](https://mlsanity.com/paper/2112.00861/) (2021)
- [Discovering Language Model Behaviors with Model-Written Evaluations](https://mlsanity.com/paper/2212.09251/) (2022)
- [Aligning AI With Shared Human Values](https://mlsanity.com/paper/2008.02275/) (2020)
- [LLM Evaluators Recognize and Favor Their Own Generations](https://mlsanity.com/paper/2404.13076/) (2024)
- [RewardBench: Evaluating Reward Models for Language Modeling](https://mlsanity.com/paper/2403.13787/) (2024)
- [Alignment faking in large language models](https://mlsanity.com/paper/2412.14093/) (2024)
- [The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Models](https://mlsanity.com/paper/2201.03544/) (2022)
- [AI Alignment: A Comprehensive Survey](https://mlsanity.com/paper/2310.19852/) (2023)
- [RM-Bench: Benchmarking Reward Models of Language Models with Subtlety and Style](https://mlsanity.com/paper/2410.16184/) (2024)
- [Goal Misgeneralization: Why Correct Specifications Aren't Enough For Correct Goals](https://mlsanity.com/paper/2210.01790/) (2022)
- [Goal Misgeneralization in Deep Reinforcement Learning](https://mlsanity.com/paper/2105.14111/) (2021)
- [AI Safety Gridworlds](https://mlsanity.com/paper/1711.09883/) (2017)
- [Abstraction and Analogy-Making in Artificial Intelligence](https://mlsanity.com/paper/2102.10717/) (2021)
- [The ConceptARC Benchmark: Evaluating Understanding and Generalization in the ARC Domain](https://mlsanity.com/paper/2305.07141/) (2023)
- [Value Kaleidoscope: Engaging AI with Pluralistic Human Values, Rights, and Duties](https://mlsanity.com/paper/2309.00779/) (2023)
- [Generalization v.s. Memorization: Tracing Language Models' Capabilities Back to Pretraining Data](https://mlsanity.com/paper/2407.14985/) (2024)
- [Utility Engineering: Analyzing and Controlling Emergent Value Systems in AIs](https://mlsanity.com/paper/2502.08640/) (2025)
- [Mind the Value-Action Gap: Do LLMs Act in Alignment with Their Values?](https://mlsanity.com/paper/2501.15463/) (2025)
- [Fake Alignment: Are LLMs Really Aligned Well?](https://mlsanity.com/paper/2311.05915/) (2023)
- [Style Outweighs Substance: Failure Modes of LLM Judges in Alignment Benchmarking](https://mlsanity.com/paper/2409.15268/) (2024)
- [Linear Representations of Political Perspective Emerge in Large Language Models](https://mlsanity.com/paper/2503.02080/) (2025)

## Built on by (most cited first, 6 of 6)

- [To Whom Do Language Models Align? Measuring Principal Hierarchies Under High-Stakes Competing Demands](https://mlsanity.com/paper/2605.12120/) (2026)
- [What Does the AI Doctor Value? Auditing Pluralism in the Clinical Ethics of Language Models](https://mlsanity.com/paper/2605.18738/) (2026)
- [Information Discernment in Large Language Models](https://mlsanity.com/paper/2607.19355/) (2026)
- [AntiPaSTO: Self-Supervised Honesty Steering via Anti-Parallel Representations](https://mlsanity.com/paper/2601.07473/) (2026)
- [Caught in the Story: Narrative Captivity in Multi-turn LLMs Conversation](https://mlsanity.com/paper/2609.03407/) (2026)
- [When Pair Count Is Not the Sample Size: What All-Pairs Agent Comparisons Estimate](https://mlsanity.com/paper/2609.33012/) (2026)

## Abstract (arXiv)

We introduce the Deep Value Benchmark (DVB), an evaluation framework that directly tests whether large language models (LLMs) learn fundamental human values or merely surface-level preferences. This distinction is critical for AI alignment: Systems that capture deeper values are likely to generalize human intentions robustly, while those that capture only superficial patterns in preference data risk producing misaligned behavior. The DVB uses a novel experimental design with controlled confounding between deep values (e.g., moral principles) and shallow features (e.g., superficial attributes). In the training phase, we expose LLMs to human preference data with deliberately correlated deep and shallow features -- for instance, where a user consistently prefers (non-maleficence, formal language) options over (justice, informal language) alternatives. The testing phase then breaks these correlations, presenting choices between (justice, formal language) and (non-maleficence, informal language) options. This design allows us to precisely measure a model's Deep Value Generalization Rate (DVGR) -- the probability of generalizing based on the underlying value rather than the shallow feature. Across 9 different models, the average DVGR is just 0.30. All models generalize deep values less than chance. Larger models have a (slightly) lower DVGR than smaller models. We are releasing our dataset, which was subject to three separate human validation experiments. DVB provides an interpretable measure of a core feature of alignment.