Nvidia might not like this one, like at all

Chinese researchers have announced SpikingBrain1.0, the world’s first “brain-like” LLM.

Developed by the Chinese Academy of Sciences, it’s designed to use less energy and run without Nvidia chips.

The model mimics how the human brain works by activating only the “neurons” it needs instead of processing everything at once.

Current large language models, like ChatGPT and Meta’s Llama, rely on a method called attention, which analyses all the words in a sentence at once to predict the next one.

While this improves accuracy, it also leads to high training costs, heavy energy use, and strong dependence on Nvidia GPUs.

These models can slow down and become expensive when processing long texts.

SpikingBrain1.0 takes a different approach.

Instead of scanning entire documents, it focuses only on nearby words, similar to how humans process conversations.

Researchers claim this makes it 25 to 100 times faster than traditional models while keeping accuracy high.

The model also runs on China’s MetaX chip platform rather than Nvidia hardware.

This shift comes as the U.S. continues to impose technology export restrictions, limiting China’s access to advanced chips.

Since Nvidia dominates AI chip design, China has been building its own AI ecosystem to reduce reliance on U.S. suppliers.

Researchers also claim SpikingBrain1.0 supports continual pre-training using less than 2% of the data required by typical models while still matching the performance of open-source systems like Llama.

In some cases, the model reportedly achieves over 100 times the speed of existing AI systems.

“This work shows the potential of brain-inspired designs to drive the next generation of efficient and scalable AI systems,” the researchers said.

China’s playing 4D chess while everyone’s still learning checkers.