New benchmark tests AI chatbots' ability to detect unsafe conversations

Original: AICompanionBench: Benchmarking LLMs-as-Judges for AI Companion Safety

Source: arxiv.org ↗