How is China’s approach toward AI different from America. Since ChatGPT came along, there has been an explosion of firms working on LLMs and AIGC, so many of the projects are not so different from large tech firms in America. However, one can still seem some practical differences from what Chinese firms worked on in 2022 and some of the projects that have popped up recently as well as the firms involved with them.
Recently, I watched a video of Li Auto’s new MindGPT in action. Keep in mind, this LLM is the current leader on C-EVAL ranking. C-EVAL ranks the ability of different LLM in answering various questions. Watching MindGPT in action is also quite interesting. It works in the form of AI assistance that comes as part of Li Auto’s infotainment system. It can freely interact with driver & take commands like “I’m a little cold”, “I had a long day”, “I want to take a trip to Tianjin tomorrow, where can I visit?”, “I want to drive to this location, plan out a route for me with yummy food along the way”. It is intelligent enough to consider what the driver means behind these commands and respond appropriately. From this, you can see why Xiaomi and Huawei have gotten into EV/AV world. Xiaomi considers EV to be a “moving intelligent home” allowing for greater interaction with full ecosystem
This allows your car to control smart home and issue commands form your car to warm up living room, iRobot to clean the floors & many other commands in the future when more AI robots are available.
Aside from this, I also saw recent report of Vivo’s AI division looking to get into Mixed Reality and Humanoid robot.
You can see that with VR/MR glasses, additional details can be projected in front of you through these devices. Before chatGPT came along, a big part of AI investment was actually for Metaverse. Basically, allowing people to immerse themselves in a different world. Not so different from Meta/facebook’s effort. But the really interesting part of AI development in China is just how many companies have gotten into Humanoid/AI robots. While, Humanoid robot is actually not the best physical form to assist people, its place in our daily lives have been popularized by movies & TV series. As such, many tech companies in China have gotten into humanoid robots not only to improve automation & efficiency of factories, but also to sell them as smart consumer electronics product in the future. A few month ago, startup Agibot advertised a very capable looking robot called Raise-A1 that it believes can be sold for 200k RMB. So essentially, anyone that can afford an EV can also afford a humanoid robot. In many ways, the technology involved in making smart cars apply almost universally to making humanoid robot. Similarly, this tech can also apply to robotic dogs that are now quite commonly deployed by leading Chinese robotics companies. They were put into good use in Hangzhou Asian Games.
Even beyond that, things like Lidar, IMU, INS, MMW radar, camera module, CMOS, ECU, MCU, power chips, RF chips, Serdes & ETH chips, high bandwidth network, analog chips, CPU, NLP (natural language processing), AI inference chip, AI training, sensor fusion, BEV (birds eye view) transformation of camera data to object and many part of AI algorithms are applicable across a bunch of autonomous robotic fields. Recently, we’ve seen DJI’s ADAS software achieving a major upgrade. DJI is obviously known for its advanced drone technology and integrating sensors to carry out autonomous functions such as avoiding obstacles, navigating to correct location, providing sensor fused data to user and carry out commands like surveying forrest, dropping payload on farm land, taking pictures and such. I think it’s quite clear that underlying concept and technology for navigating & making decision in the air is similar to doing so on the road. That’s why DJI ADAS division has so many customers.
Similarly, Meituan also has its own delivery drones now that uses similar technology to deliver food and such in Shenzhen. People may often wonder why I spend so much time looking at ADAS tech. It’s because the tech behind ADAS have so much application in all future autonomous systems from civilian ones that I already talked about to military applications like UGVs, massive drone attacks, large autonomous UCAVs, UUV/AUVs and armed robot dogs. The applications to these fundamental technologies are massive.
People often comment that China has turned to mature chip fabs as a result of October sanctions. I would say they always had plans to massively expand mature chip productions. The market for these “mature chips” in future AI controlled consumer electronics product and machines are enormous. Large language models are part of AI, but far from all of AI. They make smart cars, robots and consumer electronics product even smarter, but other techs I’ve talked about are also very important. It would be quite silly for example to dismiss the importance of TI or NXP to AI powered systems.
In summary, both China and America are devoting significant resources in LLMs. China is at the forefront of consumer electronics and robotics product, so its focus on LLMs are driven by how to make its consumer electronics and robotics product smarter. So, making hard tech better. American tech companies are the leaders of search, e-commerce, social media and apps. So, its focus on LLMs are focused on how to make these soft tech products better. So when we say one country is x amount of time ahead of the other country? I will have ahead in which area? We can say GPT4.0 is several months ahead of Baidu’s Ernietbot or Alibaba’s Gwen in AIGC, answering questions or generating code. But we can also say that Vivo is ahead of iPhone in AI assistant and that Li Auto is ahead of Tesla in ADAS.
Of course, there are other areas of AI that we haven’t even talked about here. Huawei has been using AI to improve efficiency & safety in various smart port projects and smart mining projects. It would be hard for me to find similar projects in America. But I think it’s fair to say that AI encompasses many areas and there are areas of leadership for both China and America.
Bytedance is surprisingly kinda unheard in this space.