arXiv:2605.28882v2 Announce Type: replace 
Abstract: With the rapid advancement of large language models, evaluating human-likeness in open-ended conversation has become increasingly important. However, human-likeness is a form of tacit knowledge that humans perceive intuitively, yet the underlying criteria resist explicit formulation. Human judgments vary widely, with strong agreement on some cases and legitimate disagreement on others. Meanwhile, the criteria behind human judgments remain implicit, leaving no clear basis for constructing cases. Further, what counts as human-likeness is not static, but evolving with model capability and human expectations. Despite progress in evaluation methods such as expert-authored benchmarks, Reward Models, and self-evolving benchmarks, none addresses all three challenges simultaneously. Therefore, we propose GrowLoop, a self-evolving conversation evaluation system that continuously adapts as models advance and scenarios shift. Starting from minimal human seed annotations, LLM agents iteratively extract and refine evaluation rubrics through Heuristic Learning. Human-AI agreement is required where annotators converge, while only plausibility is expected where they diverge. Moreover, the Rubric-Case co-evolution mechanism enables continuous evolution. When the evaluation target shifts, new human seeds expand the system's coverage accordingly. When applied to human-likeness evaluation in open-ended conversation, the AI judge guided by these rubrics not only substantially outperforms existing methods in alignment with human judgments, but also uncovers issues that annotators overlook. The resulting benchmark effectively discriminates models across capability tiers and reveals where they fall short, while generalizing to new scenarios and adapting as models advance. Our work shifts the benchmarking paradigm from manual updates or difficulty scaling to comprehensive, continuous self-evolution.

تم اقتراح نظام جديد يسمى GrowLoop لتقييم الشبه البشري في المحادثات المفتوحة، حيث يتناول التحديات المتعلقة بالمعايير المتطورة والأحكام البشرية الضمنية في سياق نماذج اللغة الكبيرة (LLMs). يتكيف هذا النظام التقييمي الذاتي التطور باستمرار مع تغير النماذج والسيناريوهات، مما يمثل تقدمًا كبيرًا في تقييم المحادثات بواسطة الذكاء الاصطناعي.

Se ha propuesto un nuevo sistema llamado GrowLoop para evaluar la similitud humana en conversaciones abiertas, abordando los desafíos de los criterios en evolución y los juicios humanos implícitos en el contexto de los grandes modelos de lenguaje (LLMs). Este sistema de evaluación auto-evolutivo se adapta continuamente a medida que los modelos y escenarios cambian, marcando un avance significativo en la evaluación de conversaciones por parte de la IA.

Un nouveau système appelé GrowLoop a été proposé pour évaluer la similarité humaine dans les conversations ouvertes, abordant les défis des critères évolutifs et des jugements humains implicites dans le contexte des grands modèles de langage (LLMs). Ce système d'évaluation auto-évolutif s'adapte en continu à mesure que les modèles et les scénarios changent, marquant une avancée significative dans l'évaluation des conversations par l'IA.

A new system called GrowLoop has been proposed to evaluate human-likeness in open-ended conversations, addressing the challenges of evolving criteria and implicit human judgments in the context of large language models (LLMs). This self-evolving evaluation system adapts continuously as models and scenarios change, marking a significant advancement in AI conversation assessment.