How DeepSeek’s new method of training advanced AI models could shake everything up – again

Follow ZDNET: Add us as a preferred source on Google.

Key findings from ZDNET

• DeepSeek debuted with Manifold-Constrained Hyper-Connections or MHCs.
• They provide a way to scale LLMs without incurring large costs.
• The company postponed the release of its R2 model to mid-2025.

Just before the start of the new year, the AI ​​world was introduced to a potentially groundbreaking new method for training advanced models.

A team of researchers from Chinese AI company DeepSeek published a paper on Wednesday describing so-called “manifold-constrained hyper-connections.” MIn short, HC, which can provide engineers with a way to build and scale large language models without the huge computational costs typically required.

Also: Is DeepSeek’s new model the latest blow to proprietary AI?

DeepSeek burst into the cultural spotlight a year ago with the release of R1, a model that rivaled the capabilities of OpenAI’s o1 and was reportedly trained at a fraction of…