Failures Are Fated, But Can Be Faded: Characterizing and Mitigating Unwanted Behaviors in Large-Scale Vision and Language Models
失败是注定的,但可以被淡化:在大规模视觉和语言模型中表征和缓解 unwanted 行为
机构 * School of Computing and Augmented Intelligence, Arizona State University, Tempe, United States of America(计算与增强智能学院,亚利桑那州立大学)
AI总结 本文提出利用深度强化学习表征和缓解大规模视觉和语言模型中的失败模式,通过有限的人类反馈重构失败景观,验证了方法在计算机视觉、自然语言处理和视觉-语言任务中的有效性。
Comments 25 pages, 35 figures
Journal ref Proceedings of the 41st International Conference on Machine Learning (ICML 2024), PMLR 235:42999-43023, 2024