DeepSeek drops training bomb: Shares path to cheaper AI models — here's why next model matters

Chinese AI firm DeepSeek unveiled a groundbreaking training method, "Manifold-Constrained Hyper-Connections," potentially slashing AI development costs. This innovation challenges the notion that advanced AI necessitates massive computing power, following their previous success with the R1 model. The breakthrough intensifies industry debates on scaling AI, with expectations high for DeepSeek's next model release.