Google proposes \'sleep framework\' to help AI learn new knowledge without forgetting old information

Google has unveiled research on a so-called “sleep framework” designed to help large language models continue learning after training while retaining existing knowledge. The approach separates processing of new inputs from a consolidation phase, mirroring how humans take in experiences while awake and reorganise memory during sleep. It aims to curb “catastrophic forgetting” and reduce gaps between training data and real-world use. The company says further verification is needed before commercial deployment.