New model, old risks: sociodemographic bias and adversarial hallucinations vulnerability in GPT-5
本文未提供摘要,请点击“查看全文”查看完整内容。