Chinese startup 01.AI is winning the open source AI race


DeLonge said the open source language model is rapidly improving and may be superior to OpenAI's market-leading GPT-4 for some specialized tasks. But many of the best open source models have come from outside the U.S., he said, adding that 01.AI could be positioned to benefit from innovations coming around its models. “American companies have become a little less open and transparent,” he said at the briefing. “But there's this interesting dynamic with AI where the more open source a company releases, the more the ecosystem develops, and therefore the stronger they become in building AI.”

Meta's Llama 2 is a rare example of a top open source model from an American company and the social media giant's challenge to OpenAI, Microsoft, Google and other major tech rivals that have invested heavily in generative AI. Meta decided to release its AI language model under a license that allows commercial reuse with some caveats.

It seems that Yi-34B and Lama 2 have a lot more in common than just being leading open source AI models. Shortly after the release of the Chinese model, some developers noticed that 01.AI's code previously included a mention of the Meta's model which was later removed. 01.AI's head of open source Richard Lin later said the company would reverse the changes, and the company credited Lama2 for part of the Yi-34B's architecture. Like all major language models, 01.AI is based on the “Transformer” architecture that was first developed by Google researchers in 2017, and the Chinese company derived that component from Llama 2. 01.AI spokeswoman Anita Huang, a legal expert, says the company said in the consultation that the Yi-34B is not under Lama 2's license. Meta did not respond to a request for comment.

Regardless of the extent to which the Yi-34B is borrowed from the Lama 2, the Chinese model functions very differently due to the data fed to it. “Yi shares the architecture of Llama but its training is completely different – ​​and significantly better,” says Eric Hartford, an AI researcher at who follows open source AI projects. “They are completely different.”

Meta's relationship with Llama 2 is an example of how it is currently following the US lead in generative AI, despite Li's confidence in China's AI expertise. Jeffrey Ding, an assistant professor at George Washington University who has studied China's AI landscape, says that although Chinese researchers have released dozens of large language models, the industry as a whole still lags behind the US.

He says, “Western companies had a significant advantage in large language model development because they could take advantage of public releases to test issues, obtain user feedback, and generate interest in new models.” Ding and others have argued that Chinese AI companies face stronger regulatory and economic barriers than their American counterparts.

Speaking at the World Economic Forum in Davos last week, Lee argued – perhaps hoping to drive home the message – that an open approach will be key for any country to take full advantage of AI.

“One of the issues with one or a few companies having the most power and dominating the models is that it creates tremendous inequality, and not just with people in less rich and less rich countries, but with professors researchers, students , also with entrepreneurs. Amateurs,” Lee said. “If there was no open source, what would they do to learn; Because they could be the next creator, inventor or developer of applications.”

If he's right, 01.AI's technology—and the applications built on top of it—will put China at the center of the next phase of the tech industry.