TriEval: A Resource-Efficient Pipeline for LLM Bias, Toxicity, and Truthfulness Assessment 事件

TriEval: A Resource-Efficient Pipeline for LLM Bias, Toxicity, and Truthfulness Assessment arXiv:2606.03036v1 Announce Type: new Abstract: LLMs have evolved from basic chatbots to the backbone of the AI ecosystem, now widely used in healthcare, schools, and government services. The domain-wide adoption of LLMs necessitates continuous evaluation to ensure their safety and fairness. Common issues encountered after deploying LLMs include inconsistent outputs and hallucinations of incorrect informa