ChatGPT and Generative AI: Technical Foundations and EU Regulation

ChatGPT has become a global sensation. One week after it launched, one million users signed up to use OpenAI’s most recent chatbot. ChatGPT can answer any question with impressive accuracy and eloquence. It has become the general public’s first hands-on introduction to the stunning recent advancements that have been made in the field of AI. Other large generative AI models have also shaken up the AI community with their ability to generate images (DALL-E 2), videos (Synthesia) and even art (Stable Diffusion).

The EU is preparing to regulate large generative AI models. Firstly, new rules could be adopted through the Artificial Intelligence Act, and its provisions on Foundation Models or General-Purpose AI systems (GPAIs). Secondly, the Digital Services Act (DSA) would be applied if a user posts AI generated content on social media platforms. This emerging landscape of large generative AI models creates a complex technical and regulatory landscape that SMEs must navigate. Although it is unlikely for SMEs to build large generative AI models, it is often the case that SMEs fine-tune Foundation Models downstream the value chain to deploy them in the Single Market.

Hence, in the February meeting of Focus Group AI, we discussed:

• What are the technical foundations of large generative AI models, such as ChatGPT?
• How does the EU plan to regulate the risks and harms that large generative AI models may generate?
• How can SMEs navigate the emerging technical and regulatory landscape and benefit from the emergence of large generative AI models?