Norizon AI raises angel funding for models that learn after deployment
The AMW Read
The funding introduces a model-learning startup and a distinct technical approach, while its reported results and commercial impact remain unverified.
Norizon AI raises angel funding for models that learn after deployment
Beijing-based Norizon AI said it raised several hundred million yuan across angel and angel-plus rounds. Investors include 5Y Capital, HongShan, Lenovo Capital, Zhongguancun Science City Fund, and Gobi Partners. The company plans to use the funding for research and product development. Founder and chief scientist Muhan Zhang is an assistant professor at Peking University and previously worked on large-scale graph machine learning at Meta.
Norizon is developing real-time recursive self-improvement technology intended to let models learn from new information after deployment. Its SHINE hypernetwork aims to generate lightweight parameter updates in a single forward pass. That approach targets a different part of the model stack from retrieval or larger context windows, which give models access to new information without necessarily changing their parameters. The company says its approach performed better than a fine-tuning comparison in learning new knowledge and retaining general-task performance, but the report provides no benchmark figures.
For builders and investors, the key question is whether parameter updates remain reliable over long periods of use. Norizon is still researching stability across millions of tokens, as well as longer knowledge internalization and larger hypernetworks. Evidence from those efforts and real business deployments would help assess whether the approach can preserve existing capabilities while adding useful knowledge. The reported financing supports the work; it does not establish that the technology is ready for production.