Large generative AI models (LGAIMs), such as ChatGPT or Stable Diffusion, are rapidly transforming the way we communicate, illustrate, and create. However, AI regulation, in the EU and beyond, has primarily focused on conventional AI models, not LGAIMs. This paper will situate these new generative models in the current debate on trustworthy AI regulation, and ask how the law can be tailored to their capabilities. The paper proceeds in three steps, covering (1) direct regulation, (2) content moderation, and (3) policy proposals. It finishes by making two distinct policy proposals to ensure that LGAIMs are trustworthy and deployed for the benefit of society at large. First, rules in the AI Act and other direct regulation must match the specificities of pre-trained models. In particular, concrete high-risk applications, and not the pre-trained model itself, should be the object of high-risk obligations. Moreover, detailed transparency obligations are warranted. Non-discrimination provisions may, however, apply to LGAIM developers. Second, the core of the DSA content moderation rules should be expanded to cover LGAIMs. This includes notice and action mechanisms, and trusted flaggers. In all areas, regulators and lawmakers need to act fast to keep track with the dynamics of ChatGPT et al.

翻译：大型生成式AI模型（LGAIMs），如ChatGPT或Stable Diffusion，正迅速改变我们的沟通、插图和创造方式。然而，欧盟及其他地区的AI监管主要集中于传统AI模型，而非LGAIMs。本文将把这些新型生成式模型置于当前可信AI监管的辩论中，并探讨法律如何针对其能力进行专门设计。论文分三步展开，涵盖（1）直接监管、（2）内容审核和（3）政策建议。最后提出两项独特的政策建议，以确保LGAIMs可信赖且造福整个社会。首先，《AI法案》及其他直接监管中的规则必须匹配预训练模型的特性。具体而言，高风险义务的对象应是具体的高风险应用，而非预训练模型本身。此外，需要详细的透明度义务。然而，非歧视条款可能适用于LGAIM开发者。其次，应将《数字服务法案》内容审核规则的核心扩展至覆盖LGAIMs，包括通知与行动机制及可信举报者。在所有领域，监管者与立法者需要迅速行动，以跟上ChatGPT等模型的发展动态。