Chinese researchers claim to have developed the world's first 'brain-like' artificial intelligence large language model, named SpikingBrain1.0, which is designed to consume less power and operate without Nvidia chips. The model, created by scientists from the Chinese Academy of Sciences, mimics the human brain by firing only the nerve cells it needs.
Unlike traditional large language models such as ChatGPT and Meta's Llama, which rely on the scaling law and compare every word to every other word, SpikingBrain focuses on nearby words, similar to how humans process recent context in conversations. This approach allegedly makes the model 25 to 100 times faster than conventional AI models.
The scientists claim that SpikingBrain can achieve performance comparable to mainstream open-source models while using less than 2 per cent of the data for continual pre-training. The model runs on China's homegrown MetaX chip platform, bypassing the need for Nvidia hardware, which has been subject to US export restrictions.
In a yet-to-be-peer-reviewed study posted on arXiv, the researchers note that traditional AI models suffer from high training costs, high energy consumption, and complex deployment pipelines. They argue that their brain-inspired mechanism could drive the next generation of efficient and scalable large model design.
