analiza aiqbench
This research introduces Pluralistic Preference Optimization (PlurPO), a novel training method that reduces AI sycophancy by simulating multiple stakeholder perspectives rather than relying on ground-truth labels. By addressing the model's tendency to overly center on the user, PlurPO significantly lowers endorsement rates for harmful actions and aligns advice generation more closely with human norms in complex social scenarios. The approach demonstrates effective transferability across different model sizes, offering a scalable solution for improving the safety and reliability of generative AI in personal advisory contexts.
Mitigating Social Sycophancy via Pluralistic Preference Optimization
arXiv:2610.02568v1 Announce Type: new Abstract: Personal advice, including relationship advice, now ranks among the most common uses of generative AI. But language models (LMs) exhibit sycophancy: they affirm users much more often than humans do, which can make people overconfident and less willing to repair their relationships after a conflict. Prior work on mitigating sycophancy has focused on factual settings where a response can be checked against a ground truth answer, while mitigations for social sycophancy (e.g., personal advice, where there is no ground truth) have relied on simple prompting and post-training methods with limited effectiveness. Our insight is that social sycophancy occurs in part because LMs overly center on the user and fail to consider the perspectives of other stakeholders impacted by the user's behavior. To address this problem we propose Pluralistic Preference Optimization (PlurPO): given inputs describing interpersonal conflicts, the LM identifies and simulates the relevant stakeholders, and is then trained to prefer and generate responses acceptable to all stakeholders. PlurPO uses only signals the model produces about its own outputs, without ground-truth labels. PlurPO substantially reduces social sycophancy across four datasets and four model families compared to prior methods. For example, on statements of intent to cause harm, where the users' actions should not be endorsed, PlurPO reduces the endorsement rate by 89% on average across four models. On general advice questions, where the target is to match the endorsement rate of human responses, it closes the gap by more than half, from 17.8% to 8.0% on average. The preference dataset constructed by PlurPO for an 8B model also effectively transfers to mitigating sycophancy in a larger (32B) model. Our results indicate that social sycophancy can be reduced by leveraging a model's own capabilities to simulate a plurality of relevant perspectives.
Abstract:Personal advice, including relationship advice, now ranks among the most common uses of generative AI. But language models (LMs) exhibit sycophancy: they affirm users much more often than humans do, which can make people overconfident and less willing to repair their relationships after a conflict. Prior work on mitigating sycophancy has focused on factual settings where a response can be checked against a ground truth answer, while mitigations for social sycophancy (e.g., personal advice, where there is no ground truth) have relied on simple prompting and post-training methods with limited effectiveness. Our insight is that social sycophancy occurs in part because LMs overly center on the user and fail to consider the perspectives of other stakeholders impacted by the user's behavior. To address this problem we propose Pluralistic Preference Optimization (PlurPO): given inputs describing interpersonal conflicts, the LM identifies and simulates the relevant stakeholders, and is then trained to prefer and generate responses acceptable to all stakeholders. PlurPO uses only signals the model produces about its own outputs, without ground-truth labels. PlurPO substantially reduces social sycophancy across four datasets and four model families compared to prior methods. For example, on statements of intent to cause harm, where the users' actions should not be endorsed, PlurPO reduces the endorsement rate by 89% on average across four models. On general advice questions, where the target is to match the endorsement rate of human responses, it closes the gap by more than half, from 17.8% to 8.0% on average. The preference dataset constructed by PlurPO for an 8B model also effectively transfers to mitigating sycophancy in a larger (32B) model. Our results indicate that social sycophancy can be reduced by leveraging a model's own capabilities to simulate a plurality of relevant perspectives.
Artificial Intelligence (cs.AI)
arXiv:2610.02568 [cs.AI]
(or arXiv:2610.02568v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.02568 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 383 cuvinte
Subiecte conexe
The Continual Constitutional Convention: How language models revise the documents that govern them
AI Alignment Forum · 13 oct. 2026
Current AIs out-persuade professionals in lab settings but (probably) not in the world
LessWrong · 5 oct. 2026
Mitigating LLM biases toward spurious social contexts using direct preference optimization
arXiv · 27 aug. 2026
State of the Field: AI for Epistemics and Coordination
LessWrong · 2 oct. 2026