| Challenge: | Large Language Models behave non-deterministically, and prompting is a common method for steering their outputs. |
| Approach: | They use role-play at scale to study the value orientation and inertia of Large Language Models. |
| Outcome: | The proposed model keeps values skewed in one direction across persona settings. |
Similar Papers
Are the Values of LLMs Structurally Aligned with Humans? A Causal Perspective (2025.findings-acl)
Copied to clipboard
Yipeng Kang, Junqi Wang, Yexin Li, Mengmeng Wang, Wenming Tu, Quansen Wang, Hengli Li, Tingjun Wu, Xue Feng, Fangwei Zhong, Zilong Zheng
| Challenge: | Current approaches to value alignment focus on a few core values, such as helpfulness, harmlessness, and honesty. |
| Approach: | They propose to use latent causal value graphs to guide two lightweight value-steering methods . role-based prompting and sparse autoencoder (SAE) steering are also used . |
| Outcome: | Experiments on Gemma-2B-IT and Llama3-8B- IT show that the proposed methods are effective and controllable. |
Persona Prompting as a Lens on LLM Social Reasoning (2026.eacl-long)
Copied to clipboard
Jing Yang, Moritz Hechtbauer, Elisabeth Khalilov, Evelyn Luise Brinkmann, Vera Schmitt, Nils Feldhus
| Challenge: | Persona prompting (PP) is increasingly used to steer large language models towards user-specific generation, but its effect on rationales remains underexplored. |
| Approach: | They examine how LLM-generated rationales vary when conditioned on different demographic personas . they use word-level rationale annotations to measure agreement with human annotations based on PP . |
| Outcome: | The proposed model improves classification on the most subjective task, but fails to align with real-world demographic counterparts. |
Context-Value-Action Architecture for Value-Driven Large Language Model Agents (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing LLMs exhibit behavioral rigidity, a flaw often masked by the self-referential bias of current "LLM-as-a-judge" evaluations. |
| Approach: | They propose a Context-Value-Action architecture that decouples action generation from cognitive reasoning via a Value Verifier trained on authentic human data to explicitly model dynamic value activation. |
| Outcome: | The proposed architecture significantly outperforms baseline models on 1.1 million real-world interaction traces on CVABench. |
Do LLMs Align Human Values Regarding Social Biases? Judging and Explaining Social Biases with LLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models can lead to undesired consequences when misaligned with human values . previous studies have shown misalignment of LLMs with human value using expert-designed or agent-based emulated bias scenarios . |
| Approach: | They investigate whether large language models (LLMs) are misaligned with human values . they find no significant differences in understanding of HVSB between LLMs . |
| Outcome: | The results show that large language models do not have lower misalignment rates and attack success rates . the study also shows that smaller language models have the ability to explain HVSB . |
Do Language Models Think Consistently? A Study of Value Preferences Across Varying Response Lengths (2026.findings-acl)
Copied to clipboard
| Challenge: | In short-form surveys and psychometric tests, value-related risks and preferences are often underexplored in practical settings. |
| Approach: | They compare short-form responses to long-form outputs to determine whether value preferences align with those expressed in long-term outputs. |
| Outcome: | The proposed method yields only modest gains in the consistency of value expression. |
Are Large Language Models Consistent over Value-laden Questions? (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models (LLMs) appear to bias survey answers toward certain values . however, some argue that LLMs are inconsistent to simulate particular values - a recent study . |
| Approach: | They define value consistency as similarity of answers across paraphrases, related questions and multilingual translations of a question to English, Chinese, German, and Japanese. |
| Outcome: | The proposed model is consistent across paraphrases, use-cases, translations, and within a topic. |
Think Again! The Effect of Test-Time Compute on Preferences, Opinions, and Beliefs of Large Language Models (2025.acl-industry)
Copied to clipboard
| Challenge: | Large Language Models exhibit subjective preferences, opinions, and beliefs, which may shape their behavior, influence advice and recommendations, and potentially reinforce certain viewpoints. |
| Approach: | They developed a benchmark to assess LLMs’ subjective inclinations across societal, cultural, ethical, and personal domains. |
| Outcome: | The proposed benchmark assesses LLMs’ subjective inclinations across societal, cultural, ethical, and personal domains. |
Conservative Bias in Large Language Models: Measuring Relation Predictions (2025.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) exhibit pronounced conservative bias in relation extraction tasks, often defaulting to no_relation label when an appropriate option is unavailable. |
| Approach: | They systematically evaluate the trade-off between conservative bias and hallucination in relation extraction tasks by using SBERT and LLM prompts to quantify this effect. |
| Outcome: | The proposed model defaults to no_relation label twice as often as hallucination, resulting in significant information loss when reasoning is not explicitly included in the output. |
Can Persona-Prompted LLMs Emulate Subgroup Values? An Empirical Analysis of Generalisability and Fairness in Cultural Alignment (2026.acl-long)
Copied to clipboard
Bryan Chen Zhengyu Tan, Zhengyuan Liu, Xiaoyuan Yi, Jing Yao, Xing Xie, Nancy F. Chen, Roy Ka-Wei Lee
| Challenge: | Current alignment paradigms treat "human values" as a monolithic entity, ignoring the fact that many societies are a mosaic of diverse subgroups with distinct and sometimes conflicting values, preferences, and norms. |
| Approach: | They examine whether Large Language Models can emulate distinct cultural values of subgroups . they use a global value survey to examine the value landscape of a multicultural society . |
| Outcome: | The proposed model improves on unseen, out-of-distribution subgroups by 17.4% . the model widens the disparity between subgroup groups when measured by distance-aware metrics. |
ValueBench: Towards Comprehensively Evaluating Value Orientations and Understanding of Large Language Models (2024.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are transforming diverse fields and gaining increasing influence as human proxies. |
| Approach: | They propose a psychometric evaluation pipeline grounded in realistic human-AI interactions to probe value orientations and novel tasks for evaluating value understanding in an open-ended value space. |
| Outcome: | The proposed evaluation pipeline is grounded in realistic human-AI interactions and performs tasks that approximate expert conclusions in value-related extraction and generation tasks. |