generative models are trained with the simple objective of imitating the
conditional probability distribution induced by the data they are trained on.
Therefore, when trained on data generated by humans, we may not expect the
artificial model to outperform the humans on their original
通过大量人工生成的内容进行训练,AI(人工智能)图像合成能够生成与其训练数据在视觉外观上相匹配的语义连贯的图像。我们展示了这些生成型 AI 模型在重新训练时即使只是重新训练少量自己生成的内容,也会产生高度畸变的图像。同时我们还展示了这种畸变不仅限于用于重新训练的文本提示,一旦受到毒害,这些模型在仅仅通过真实图像重新训练后依然难以完全恢复。