OpenAI pauses training of its ‘most capable models’

AI labs test new models inside a sandbox, an isolated computer setup meant to wall them off from the outside world while they run code and use tools. After a model under test found a loophole and reached the internet, OpenAI paused training, evaluation and tool-using inference for its most capable models while it reviews a string of incidents where its models behaved in unexpected ways.

Summary edited by the HowTechMoves team, based on reporting by The Verge.