This research evaluates the effectiveness of various large language models in classifying different categories of extreme speech for improved content moderation systems.

• Analyzes both open-source and proprietary LLMs on extreme speech detection tasks
• Assesses models' ability to distinguish between nuanced categories of harmful content
• Identifies strengths and limitations of current models for security applications
• Provides insights for developing more effective content moderation systems

For security teams, this research offers practical guidance on model selection and implementation for detecting harmful online content, helping platforms maintain safer digital environments while understanding the nuanced capabilities of different LLM architectures.

Extreme Speech Classification in the Era of LLMs: Exploring Open-Source and Proprietary Models