Google touts big step toward continual learning with new experimental AI model ‘HOPE’

Google has introduced its experimental AI model called HOPE, which is built on a novel framework termed nested learning. Under this approach, a model is structured as a set of interconnected, multi-level learning tasks rather than one monolithic process, allowing the system to optimise multiple layers of learning simultaneously. The goal is to tackle a key limitation of current large language models (LLMs): they’re unable to learn continually without experiencing “catastrophic forgetting” (i.e., learning new tasks causes them to forget previously learnt tasks). HOPE aims to overcome that by having a self-modifying architecture with improved long-context memory management, essentially bringing the model closer to a brain-like ability to accumulate knowledge over time.
*This news was published by The Indian Express on November 10, 2025.*

 

 

 

Source Click