Yin Qi, founder and CEO of Megvii (旷视科技), one of China’s AI titans, said the unicorn will invest firmly into innovations in large language model but will not “jump on the ChatGPT bandwagon” at a recent internal meeting.
Yin, who founded Megvii in 2011, said at the meeting that the management bets on the combination of LLM and robotics, which he believes is the largest opportunity for Megvii.
Megvii, SenseTime (商汤科技), CloudWalk (云从科技) and Yitu Technologies (依图科技) are often known as the four AI “little dragons” in China. Their fortunes, however, have fallen steeply amid setbacks in commercialization and diminished appetite for AI investment.
But Yin explained that the LLM can represent the firm’s best shot at reviving its fortunes.
Ten years ago, computer vision was at the frontier of all AI applications, he said.
The advent of ChatGPT has pushed the industry to a higher level, but in his view, regardless of shifting trends in technology, the underlying infrastructure remains the same.
Yin categorized AI applications into two groups. The first group, represented by the immense popularity of Douyin, or TikTok, can be labeled “AI in Digital,” as it features heavy use of recommendation-based, multimodal AI algorithms in the Internet space.
The other is what he calls “AI in Physical,” meaning a marriage of AI algorithms with hardware devices in the physical world.
According to him, Tesla is a typical example of this model, starting with intelligent electric cars and culminating in smarter, more general-purpose robotics.
Although general-purpose AI like the chatbot service from OpenAI and other startups will likely experience exponential growth in the next three to five years under the “AI in Digital” model, tech innovations that fall into the “AI in Physical” category will take longer to bear fruits, Yin noted.
But he added the fruits could be greater.
For Megvii, an AI unicorn that appears to have lost some of its luster, this presents a historical opportunity — provided one has strategic resilience, said Yin.
The Tsinghua University alumnus built up Megvii alongside his partners over a period of 10 years. Its focus transitioned from algorithms and software in the early days to products that blend hardware and software.
He emphasized that Megvii has built up its success on three pillars, namely, proprietary AI capabilities, product capacity and the ability to serve industrial clients well.
Going forward, “we will unswervingly combine hardware with software to empower industrialization,” Yin noted.
As technologies continue to make strides, LLM will definitely integrate with industry and the physical world, he said.
The key to success lies in human-machine interface and multi-modality. AI must be able to process not just text, but sound and graphics as well to be considered intelligent enough, and this is where Yin claims Megvii has an “obvious” advantage.
The company has parlayed its competitiveness in computer vision to areas such as AMRs, automated warehouse and smart logistics.
Despite the hype around LLM, he advised the company to double down on its current business.
“We still have huge room to serve current clients. In the future, when we deploy our LLM in certain scenarios, we must do this along with clients,” said Yin. “Now we must firmly invest into R&D of LLM, while building the hardware carriers of AI and raising the barriers to competition.”
