The headline: Companies like OpenAI, Google, and Meta could face compliance issues with the EU's upcoming AI Act, according to new research out of Stanford's Center for Research on Foundation Models. What happened: The researchers found that leading AI models, like OpenAI's GPT-4 and Meta's LLaMA, fall short of Europe's draft rules regulating AI technologies. Their study evaluated 10 top AI models and found that none fully comply with critical areas of the draft rules, particularly when it comes to copyrighted data. Based on 12 criteria from the EU law, the researchers found BigScience's BLOOM scored the highest, and Aleph Alpha and Anthropic's models scored the lowest. Why it matters: The EU AI Act, the world's first comprehensive rule governing AI, will likely hold significant influence in shaping global AI regulation when it comes into force in the next couple of years. It sets AI requirements covering a population of 450 million, forcing multinational companies to change their practices to comply. While generative AI systems like ChatGPT can continue operating under the proposed rules, they would have to abide by some transparency requirements, such as thorough documentation of any copyrighted material.
|
