The President has signed an executive order to regulate AI. The order addresses many different issues, but educators are probably most interested in the focus on Foundation Models (e.g., GPT-4, Claude) because this category of AI might be something they or their students use. The regulations use GPT-4 level services as a baseline and require companies developing more powerful applications to require experts to conduct Red Team safety tests and publicly share the results. A Red Team would be what many call white hat hackers and these individuals would be charged with getting a service to violate various risks. Risks include defeating cybersecurity protections and dangerous information such as the production of biological or chemical weapons.
The way “more powerful” is defined is interesting. GPT-4 was trained using 10 to the 26th power mathematical operations and models using more operations will be required to be thoroughly evaluated before being released to the public.