The paper explores using soft prompt tuning to align open‑source large language model judges with human annotation policy. By training soft prompts on the FairBotBench benchmark, the authors evaluate the adapted judges on bias, toxicity, and personalization. Results show that hybrid soft‑prompt tuning consistently improves agreement with human annotations, and for Qwen 3.5 9B the adapted judges outperform hosted reference judges across all three dimensions.
Large Language Models (LLMs) as judges across various scenarios such as assessing model responses is becoming an increasingly accepted paradigm. However, existing judgment approaches often rely on trained judgers using fixed preference data, which tend to overlook diverse user preferences and struggle to adapt to real-world human-AI dialogue scenarios.
arXiv:2609.24516v1 Announce Type: new Abstract: In recent years, large language models (LLMs) have emerged as a popular alternative for evaluation. Often referred to as LLMs as judges (LLJs), these s...
The paper introduces UPHELD, a large benchmark of human-to-human dialogues written by professional script writers, featuring realistic turn densities and over 36,000 per-turn human annotations. It evaluates existing automatic metrics and LLM-as-a-judge methods, finding them unreliable against expert human judgment. Using UPHELD, the authors develop a Mixture-of-Judges framework that improves correlation with human assessments by about 30%.
TherapyGym is a framework designed to evaluate and improve therapy chatbots on two key clinical dimensions: fidelity to evidence-based practice and safety. It uses the Cognitive Therapy Rating Scale (CTRS) to automatically score adherence to CBT techniques across multi‑turn sessions, and a multi‑label annotation scheme to assess therapy‑specific risks such as failure to address harm or abuse. The framework also introduces THERAPYJUDGEBENCH, a validation set of 116 dialogues with 1,270 expert ratings, and serves as a training harness where CTRS and safety rewards guide reinforcement learning with diverse patient simulations, leading to significant improvements in expert‑rated performance.
arXiv:2608. 06329v1 Announce Type: cross Abstract: Task-oriented conversational agents are evaluated using curated or automatically generated benchmarks, yet benchmark quality is rarely assessed.
arXiv:2607. 12085v1 Announce Type: new Abstract: Evaluating retail conversational agents requires methods beyond lexical-overlap metrics to assess intent alignment, factuality, helpfulness, clarity, tone, and overall response quality.
arXiv:2608.29206v1 Announce Type: new Abstract: Bias in human-agent interaction can manifest not only through hostile language but also as benevolent bias, whereby unequal treatment hides behind a wa...
The paper introduces the "Prompt Lottery" phenomenon, where paraphrastic rephrasings of judge prompts in LLM-as-a-Judge (LLMaJ) lead to inconsistent judgments and can even reverse model rankings. A large-scale study across various tasks, judge families, and model backbones demonstrates that these prompt variations significantly alter evaluation outcomes. A user study shows that LLMaJ judges are 6–9 times more sensitive to prompt rephrasings than humans, and that aggregating multiple prompts improves robustness by focusing on consistent judgments.
arXiv:2606. 27226v1 Announce Type: new Abstract: Evaluating LLM outputs remains a major bottleneck in NLP: human evaluation is expensive and slow, lexical metrics correlate poorly with human judgments on open-ended generation, and holistic LLM judges often produce opaque scores that are hard to debug.
The paper introduces JudgeBiasBench, a benchmark that systematically quantifies judgment biases in large language model (LLM)-based judges across four dimensions and 12 bias types. It evaluates both generative and discriminative judges, revealing significant bias patterns that undermine reliability. The authors propose bias-aware training—reinforcement learning for generative judges and contrastive learning for discriminative judges—to reduce these biases while maintaining evaluation performance.
arXiv:2605. 28882v2 Announce Type: replace-cross Abstract: With the rapid advancement of large language models, evaluating human-likeness in open-ended conversation has become increasingly important.
arXiv:2605. 15416v2 Announce Type: replace-cross Abstract: Jung et al.