Recently, users reported encountering inappropriate—even offensive—responses while interacting with Tencent’s AI assistant ‘Yuanbao,’ sparking public concern. In response, Tencent swiftly addressed the issue, stating it takes user feedback seriously and has immediately launched an investigation. The company emphasized that Yuanbao AI was designed to provide efficient, friendly, and safe intelligent services, and it will not tolerate any behavior that violates ethical standards or harms user experience.Tencent explained that preliminary findings suggest the anomalous replies may have resulted from the model’s misinterpretation of certain inputs or susceptibility to adversarial prompt injections. The technical team has since optimized the relevant modules and strengthened content safety filters to prevent recurrence. Users are also encouraged to report abnormal outputs through official channels to help improve the AI system.This incident highlights a key challenge in deploying large language models: balancing openness and creativity with content safety and compliance. Tencent reaffirmed its commitment to ongoing research in AI ethics and safety, advocating for the responsible development of artificial intelligence.
近日，有用户反映在使用腾讯推出的AI助手‘元宝’时，遭遇了不当甚至辱骂性回复，引发舆论关注。对此，腾讯方面迅速作出回应，表示高度重视用户反馈，并已紧急排查相关问题。公司强调，元宝AI的设计初衷是为用户提供高效、友好、安全的智能服务，绝不会容忍任何违反伦理或伤害用户体验的行为。腾讯解释称，初步调查发现，个别异常回复可能源于模型在特定语境下对输入内容的理解偏差，或受到对抗性提示（prompt injection）的影响。目前，技术团队已对相关模块进行优化，并加强了内容安全过滤机制，防止类似情况再次发生。同时，腾讯鼓励用户通过官方渠道举报异常内容，以协助持续改进AI系统。此次事件再次凸显了大模型在实际应用中面临的挑战——如何在保持开放性和创造力的同时，确保输出内容的安全与合规。腾讯表示将持续投入AI伦理与安全研究，推动负责任的人工智能发展。
原创文章，作者：admin，如若转载，请注明出处：https://avine.cn/8397.html