In this post, I will not be giving an authoritative review of the state of frontier AI in China. I will instead just be saying what I think such a review would cover, and then share such scraps of information as I have. My objective is really to point out a gap in the information available here, and maybe someone will know, better than me, how to fill it.
To set a standard, consider the state of frontier AI in America. Arguably the big four companies for frontier AI are OpenAI, Google, xAI, and Anthropic. They are American companies and so, along with their internal dynamics and their interactions with society in general, they are subject to the laws and policies of the American government.
Right now we happen to have an American government whose political base I have characterized as techno-populist, in that it is an alliance between the revolutionary populism and nationalism of MAGA, and the techno-optimism of Big Tech in the era of AI. The Trump 2.0 coalition could be described in a more fine-grained way than that, but that’s enough detail to acknowledge that Big Tech, like everything else in the West right now, is subject to the whims and decisions of the Trump team; and that for the most part, Big Tech are working with, rather than against, this new order, which is actually very pro-AI, favoring deregulation and helpful geostrategic deals. That is the political context in which the frontier AI companies are operating.
Ideally we would have a similarly crisp understanding of the political environment in which Chinese frontier AI operates. I think in Chinese political terms, Xi Jinping’s China is “center-right”, avoiding both the leftism that might slow down China’s technological development, and the liberalism that might destabilize its political system. Of course, its political system is single-party rule, nominally Marxist but with enough pragmatism to be not so different from developing countries in the western camp that we would call authoritarian capitalist societies, and really held together by plain old nationalism (which perhaps I should note for western readers, is political common sense for most of the world outside the West).
Returning to the political context of American AI for the moment, we know that David Sacks (the AI and crypto czar of Trump 2.0) is somehow central to policy, and that the expressed aim of American policy is to ensure that the American “AI stack” is the main one that is used worldwide. So we should want to know the Chinese bodies and officials that have an analogous regulatory and strategic role in China, and what their economic and strategic priorities are. One should probably also know a few of the other prominent figures in Chinese political life, like Xi’s premier Li Qiang, chief ideologist Wang Huning, and possible successor Hu Chunhua, just to flesh out the broader political context.
Finally, one has China’s frontier AI companies (they call LLM-style AI, “AI 2.0”, AI 1.0 being what is also called “machine learning”): their executives, their investors, their user bases, and, of course, their AIs. Also, as an inhabitant of the western Internet, I hear about the way that American frontier AI gets used, including new phenomena like “AI psychosis”, “AI relationships”, and “vibe science”. But I am utterly ignorant of how Chinese are using Chinese AI. Presumably there is a lot of overlap, I’m sure AI is powering corporate chatbots and student assignments over there too. But what are the differences? What happens in the West that doesn’t happen in China, what happens in China that doesn’t happen in the West? And what are the Chinese intellectual perspectives on AI and its impact, that we don’t hear about here? These are nuances it would be very interesting to know.
An interest in Chinese AI and its social emanations could be motivated simply by an interest in the state of the world. But as a transhumanist and long-term reader of Less Wrong, of course a major reason I take an interest in Chinese AI, is because it has some chance of being the context in which superintelligent AI arises. My take on AI safety is more about solving alignment than stopping the technology, and so I would like to know, for each of the Chinese AI majors, what their safety philosophy and methodology is, which part of the development team works on alignment (whatever they may call it), and so on.
No doubt there are people here who know the nuances of Manus, Qwen, and DeepSeek, the way the rest of us might know Grok, Gemini, and Claude—people who track the eval leaderboards, for example. But for most of us, I think the world of Chinese AI is known only vaguely, as this quasi-mythical other place that is also competing in the AI race. Now, maybe the transhuman fate of the world will be decided by the interplay between Musk, Altman, Hassabis, and Amodei, and everyone else is just too far behind. Nonetheless, I think it would be a good thing if part of our community was just as familiar with the names of their Chinese counterparts, and with the avenues of information flow that shape Chinese thinking about safety and alignment.
Understanding the state of frontier AI in China
In this post, I will not be giving an authoritative review of the state of frontier AI in China. I will instead just be saying what I think such a review would cover, and then share such scraps of information as I have. My objective is really to point out a gap in the information available here, and maybe someone will know, better than me, how to fill it.
To set a standard, consider the state of frontier AI in America. Arguably the big four companies for frontier AI are OpenAI, Google, xAI, and Anthropic. They are American companies and so, along with their internal dynamics and their interactions with society in general, they are subject to the laws and policies of the American government.
Right now we happen to have an American government whose political base I have characterized as techno-populist, in that it is an alliance between the revolutionary populism and nationalism of MAGA, and the techno-optimism of Big Tech in the era of AI. The Trump 2.0 coalition could be described in a more fine-grained way than that, but that’s enough detail to acknowledge that Big Tech, like everything else in the West right now, is subject to the whims and decisions of the Trump team; and that for the most part, Big Tech are working with, rather than against, this new order, which is actually very pro-AI, favoring deregulation and helpful geostrategic deals. That is the political context in which the frontier AI companies are operating.
Ideally we would have a similarly crisp understanding of the political environment in which Chinese frontier AI operates. I think in Chinese political terms, Xi Jinping’s China is “center-right”, avoiding both the leftism that might slow down China’s technological development, and the liberalism that might destabilize its political system. Of course, its political system is single-party rule, nominally Marxist but with enough pragmatism to be not so different from developing countries in the western camp that we would call authoritarian capitalist societies, and really held together by plain old nationalism (which perhaps I should note for western readers, is political common sense for most of the world outside the West).
Returning to the political context of American AI for the moment, we know that David Sacks (the AI and crypto czar of Trump 2.0) is somehow central to policy, and that the expressed aim of American policy is to ensure that the American “AI stack” is the main one that is used worldwide. So we should want to know the Chinese bodies and officials that have an analogous regulatory and strategic role in China, and what their economic and strategic priorities are. One should probably also know a few of the other prominent figures in Chinese political life, like Xi’s premier Li Qiang, chief ideologist Wang Huning, and possible successor Hu Chunhua, just to flesh out the broader political context.
Finally, one has China’s frontier AI companies (they call LLM-style AI, “AI 2.0”, AI 1.0 being what is also called “machine learning”): their executives, their investors, their user bases, and, of course, their AIs. Also, as an inhabitant of the western Internet, I hear about the way that American frontier AI gets used, including new phenomena like “AI psychosis”, “AI relationships”, and “vibe science”. But I am utterly ignorant of how Chinese are using Chinese AI. Presumably there is a lot of overlap, I’m sure AI is powering corporate chatbots and student assignments over there too. But what are the differences? What happens in the West that doesn’t happen in China, what happens in China that doesn’t happen in the West? And what are the Chinese intellectual perspectives on AI and its impact, that we don’t hear about here? These are nuances it would be very interesting to know.
An interest in Chinese AI and its social emanations could be motivated simply by an interest in the state of the world. But as a transhumanist and long-term reader of Less Wrong, of course a major reason I take an interest in Chinese AI, is because it has some chance of being the context in which superintelligent AI arises. My take on AI safety is more about solving alignment than stopping the technology, and so I would like to know, for each of the Chinese AI majors, what their safety philosophy and methodology is, which part of the development team works on alignment (whatever they may call it), and so on.
GPT-5 has supplied me with a very preliminary overview of frontier model “specs, use and safety” for six big Chinese AI companies (Baidu, Alibaba, Tencent, Zhipu, 01.AI, DeepSeek). This joins some earlier posts of mine on the Chinese AI scene and mottos for American and Chinese AIs. All these posts are sketchy, amateur work, done in a rush, and yet I don’t see anything better here. (My post on the Chinese AI scene links to a few Substacks which monitor the state of Chinese AI in a more professional way.)
No doubt there are people here who know the nuances of Manus, Qwen, and DeepSeek, the way the rest of us might know Grok, Gemini, and Claude—people who track the eval leaderboards, for example. But for most of us, I think the world of Chinese AI is known only vaguely, as this quasi-mythical other place that is also competing in the AI race. Now, maybe the transhuman fate of the world will be decided by the interplay between Musk, Altman, Hassabis, and Amodei, and everyone else is just too far behind. Nonetheless, I think it would be a good thing if part of our community was just as familiar with the names of their Chinese counterparts, and with the avenues of information flow that shape Chinese thinking about safety and alignment.