The research group METR says leading AI systems are getting better at research tasks so quickly that current evaluations may no longer reliably show whether they are safe, while its own data cannot rule out dangerous acceleration.
METR warns AI models improve faster than safety tests can keep up
The research group METR says leading AI systems are getting better at research tasks so quickly that current evaluations may no longer reliably show whether they are safe, while its own data cannot rule out dangerous acceleration.
Ana Las Heras · CC BY-SA 4.0 · rights
Monitoring item. The full text is not distributed. Extract and source below.