DeepMind's Hassabis says AGI might arrive in a few years
Heads up: Demis Hassabis, the boss at Google DeepMind, says Artificial General Intelligence (AGI), the kind of AI that could think and learn like humans, might arrive in just a few years.
He's urging for independent checks on these powerful systems before they're released to the public.
Hassabis also wants a U.S.-led group to set standards and make sure these new technologies are rolled out safely.
Hassabis proposes voluntary then mandatory tests
Hassabis suggests starting with voluntary tests for frontier AI models, then moving to mandatory ones as things get more serious.
If an AI model hits certain capability levels, it'll be labeled a "Frontier Model," and developers will need to focus on cybersecurity and safety, like adding digital watermarks to AI-generated content.
He compares AGI's potential impact to inventions like electricity but warns it could bring new security challenges, so global standards are key for keeping things safe and helpful for everyone.