aiqbench analysis
This research addresses the critical challenge of aligning lightweight, open-source LLM judges with human annotation standards to reduce evaluation costs while maintaining accuracy. By employing hybrid soft-prompt tuning, the study demonstrates that these adapted models can effectively match or surpass proprietary reference judges in assessing bias, toxicity, and personalization. This advancement offers a practical pathway for deploying scalable AI evaluation frameworks without relying on expensive closed-source APIs.
Adapting LLM Judges to Observed Human Annotation Policy through Soft Prompt Tuning
arXiv:2611.07126v1 Announce Type: cross Abstract: Large Language Models (LLMs) are increasingly employed as automated judges for evaluating chatbot responses. While proprietary models can achieve high agreement with human annotators, they remain costly and difficult to deploy in practical settings. Lightweight open-source LLM judges provide an attractive alternative but typically exhibit lower agreement with human ratings. In this paper, we investigate whether soft prompt tuning can align open-source LLM judges more closely with human annotation policy. We train soft prompts on the FairBotBench benchmark and evaluate the adapted judges across the dimensions of bias, toxicity, and personalization. Our findings demonstrate that hybrid soft-prompt tuning, which combines textual prompts with learned continuous prompt embeddings, consistently improves agreement with human annotations over the corresponding untuned baselines. For Qwen 3.5 9B, the adapted judges surpass both hosted reference judges across all three evaluation dimensions.
Abstract:Large Language Models (LLMs) are increasingly employed as automated judges for evaluating chatbot responses. While proprietary models can achieve high agreement with human annotators, they remain costly and difficult to deploy in practical settings. Lightweight open-source LLM judges provide an attractive alternative but typically exhibit lower agreement with human ratings. In this paper, we investigate whether soft prompt tuning can align open-source LLM judges more closely with human annotation policy. We train soft prompts on the FairBotBench benchmark and evaluate the adapted judges across the dimensions of bias, toxicity, and personalization. Our findings demonstrate that hybrid soft-prompt tuning, which combines textual prompts with learned continuous prompt embeddings, consistently improves agreement with human annotations over the corresponding untuned baselines. For Qwen 3.5 9B, the adapted judges surpass both hosted reference judges across all three evaluation dimensions.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2611.07126 [cs.CL]
(or arXiv:2611.07126v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2611.07126 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.
Full text under the publisher's open licence · 233 words
Related coverage
Extending Music Annotation Schemas: Zero-Shot Prediction or Few-Shot Adaptation?
arXiv · Oct 7, 2026
LakeHopper: Knowledge-Aware Adaptation of Column Type Annotators across Data Lakes
arXiv · Aug 25, 2026
Open-Set Ego-Noise Separation for Legged-Robot Audition via Annotation-Free Adaptation and Pretrained-Model Transfer
arXiv · Sep 9, 2026
Multi-Perspective LLM Annotations for Valid Analyses in Subjective Tasks
arXiv · Oct 2, 2026