China claims AI breakthrough '100 times faster than traditional models' (www.independent.co.uk)

Chinese researchers have announced a breakthrough AI model named SpikingBrain, which they claim operates up to 100 times faster than traditional large language models like ChatGPT while consuming significantly less power. Unlike conventional models that rely on full-sequence attention mechanisms—comparing every word to every other word in the input—SpikingBrain employs a brain-inspired approach that focuses attention primarily on nearby words, mirroring how humans process language contextually. This selective attention reduces computational load and energy use, enabling efficient training with less than 2% of the data typically required while maintaining competitive performance. Technically, SpikingBrain represents a significant shift by running on China’s domestically developed MetaX chip platform instead of Nvidia GPUs, overcoming recent US export restrictions that have limited China’s access to advanced AI hardware. This independence not only addresses geopolitical supply chain challenges but also offers a new pathway to scalable, energy-efficient AI deployment outside the Nvidia-centric ecosystem. The research underscores the potential of biologically inspired mechanisms to reimagine large language model architectures, balancing accuracy with substantially improved speed and sustainability—a critical development for the AI/ML community focused on next-generation model innovation and hardware diversity.