Large Language Models (LLMs) have become essential for offensive language detection, yet their ability to handle annotation disagreement remains underexplored. Disagreement samples, which arise from subjective interpretations, pose a unique challenge due to their ambiguous nature. Understanding how LLMs process these cases, particularly their confidence levels, can offer insight into their alignment with human annotators. This study systematically evaluates the performance of multiple LLMs in detecting offensive language at varying levels of annotation agreement. We analyze binary classification accuracy, examine the relationship between model confidence and human disagreement, and explore how disagreement samples influence model decision-making during few-shot learning and instruction fine-tuning. Our findings reveal that LLMs struggle with low-agreement samples, often exhibiting overconfidence in these ambiguous cases. However, utilizing disagreement samples in training improves both detection accuracy and model alignment with human judgment. These insights provide a foundation for enhancing LLM-based offensive language detection in real-world moderation tasks.

Recent advancements in Large Language Models (LLMs) have propelled text generation to unprecedented heights, approaching human-level quality. However, it poses a new challenge to distinguish LLM-generated text from human-written text. Presently, most methods address this issue through classification, achieved by fine-tuning on small language models. Unfortunately, small language models suffer from anisotropy issue, where encoded text embeddings become difficult to differentiate in the latent space. Moreover, LLMs possess the ability to alter language styles with versatility, further complicating the classification task. To tackle these challenges, we propose Gated Mixture-of-Experts Fine-tuning (GMoEF) to detect LLM-generated text. GMoEF leverages parametric whitening to normalize text embeddings, thereby mitigating the anisotropy problem. Additionally, GMoEF employs the mixture-of-experts framework equipped with gating router to capture features of LLM-generated text from multiple perspectives. Our GMoEF achieved an impressive ranking of #8 out of 70 teams. The source code is available on https://gitlab.com/sigrs/gmoef.