Chinese scientists have claimed that they have developed the world's first "brain-like" AI

Chinese scientists have announced the creation of the world's first large language model of artificial intelligence (AI) that mimics the human brain. This development, called SpikingBrain1.0, is designed to consume significantly less energy and function without using Nvidia chips. This was reported by The Independent.

The SpikingBrain1.0 model, developed by researchers from the Chinese Academy of Sciences, reproduces the principle of the human brain's operation, activating only those nerve cells that are necessary to perform the task.

Widely used large language models such as ChatGPT and Meta's Llama (recognized as extremist and banned in Russia) rely on the scaling law, according to which the performance of an AI system improves with increasing data volume and model size. These systems function by applying an "attention" technique, in which the AI model simultaneously analyzes all words in a sentence to determine the most important ones for predicting the next word.

For example, when processing long queries, such as an entire book, this method can significantly increase energy consumption and slow down the system. Scientists note that these models face a number of obstacles, including extremely high training costs, significant energy consumption, and complex deployment. Most common AI models are also designed to work exclusively on Nvidia graphics processors, which limits the circle of their developers.

In order to overcome these limitations, Chinese researchers have developed a new model that draws inspiration from brain mechanisms. Instead of analyzing the entire text as a whole, SpikingBrain1.0 focuses only on the nearest words, similar to how the human brain concentrates on recent context in conversation. This approach allows for a balance between efficiency and accuracy. According to the developers, thanks to this attention method, SpikingBrain can work 25-100 times faster compared to conventional AI models.

The new AI model also functions on China's own MetaX chip platform, without using Nvidia chips. Unlike systems like ChatGPT, which rely on an entire network, SpikingBrain selectively responds to input data, which reduces its energy consumption. This allows for continuous pre-training using less than 2% of the data, while achieving performance comparable to major open-source models.

The development is taking place against the backdrop of increased US export restrictions on technologies to China, introduced starting with Donald Trump's first administration. These restrictions interrupted the Asian giant's access to tools used to produce chips for server computers, artificial intelligence, and other advanced applications. One of the companies whose Chinese exports have been particularly affected is California chip manufacturer Nvidia, the world leader in developing specialized chips for AI technologies. As a result of these restrictions, China has begun to actively develop its own AI ecosystem.

The research on the new model was published on the arXiv platform and is awaiting peer review.