The European Union has reached a preliminary agreement this would limit the operation of the advanced ChatGPT model, in what is seen as a key part of the world’s first comprehensive artificial intelligence regulation.
All developers of general-purpose AI systems – powerful models with a wide range of possible uses – must meet basic transparency requirements, unless they are provided free and open source, according to an EU document seen by Bloomberg.
These include:
- Have an acceptable use policy
- Maintain information on how they trained their models
- Report a detailed summary of the data used to train their models
- Have a policy respecting copyright law
Models considered to present “systemic risk” would be subject to additional rules, according to the document. The EU would determine this risk based on the amount of computing power used to train the model. The threshold is set for models that use more than 10 trillion (or septillion) operations per second.
Currently, the only model that would automatically meet this threshold is OpenAi’s GPT-4, according to experts. The EU executive may designate others based on the size of the dataset, the number of professional users registered in the EU or the number of end users registered, among other possible parameters.
Learn more: Europe puts its feet on the ground with the first pact to regulate AI
These high-performance models should adhere to a code of conduct while the European Commission develops more harmonized and sustainable controls. Those who do not sign will have to prove to the commission that they comply with the AI law. The exemption for open source models does not apply to those considered to pose systemic risk.
These models should also:
- Declare their energy consumption
- Carry out red-teaming or contradictory tests, internally or externally
- Assess and mitigate possible systemic risks and report any incidents
- Make sure they use adequate cybersecurity controls
- Report information used to refine the model and their system architecture
- Comply with more energy-efficient standards if they are developed
The agreement in principle still needs to be approved by the European Parliament and the 27 EU member states. France and Germany have already expressed concerns about applying too much regulation to general-purpose AI models and risking killing off European competitors like France. Mistral AI or that of Germany Alpha Alpha.
For now, Mistral likely won’t need to respond to general AI controls because the company is still in the research and development phase, Spanish Secretary of State Carme Artigas said on Saturday.
