The smartphone screen displays the OpenAI and Anthropic logos against a blurred American flag background. (Photo by Imena Ben Youssef/Hans Lucas/AFP via Getty Images)
Hans Lukas/AFP via Getty Images
The release of an advanced artificial intelligence model by Chinese startup Moonshot has shaken up major American artificial intelligence labs. What should have been a significant benefit is now measured in months, if not weeks. It’s harder to argue that export restrictions are working, and it’s clear that China will soon be able to replicate the entire AI supply chain, right down to lithography. Because of this, it is already well positioned to accelerate the adoption of robotics and the physical world faster than we can.
Necessity is the mother of invention, and as it turns out, China is becoming increasingly successful at innovating in the field of artificial intelligence.
Of course, tests are extremely flawed, and some might argue that we lack neutral and useful tests to begin with. They are quickly becoming saturated, and some of the institutions promoting them have their own political, if not philosophical, agendas for keeping AI safe.
But early feedback confirms that Moonshot’s model is solid, even if it’s lumpier in some applications than Anthropic’s Fable or OpenAI’s Sol.
Lobbying for border labs is in full swing. And there were many options for solving the problem. The classic is the distillation theory: China is stealing and it must be stopped. The irony, of course, is that the frontier labs were trained on all kinds of copyrighted materials, so they were the first to recycle them.
But it is increasingly impossible to explain China’s progress solely through distillation attacks (or, more accurately, through model learning). And we may soon have to experience a “Sputnik moment” when the United States emerges as the successor in the AI race.
Following is not necessarily a bad thing, assuming the government doesn’t listen to lobbyists and instead plays to the comparative advantage of the United States. Historically, our economy has succeeded when we embrace openness and competition. This was true for the Internet, the same is happening with dollar stablecoins around the world, and it will be true for AI.
What’s confusing is that, on the surface, it appears China is playing our own game. President Xi Jinping has reaffirmed his strong commitment to the open source ecosystem. Of course, open-source AI plays to China’s strengths as it has complementary resources in robotics and manufacturing. Moreover, as Hollywood taught many countries throughout the 20th century, if you own the content, your values travel with it across borders. And there are real safety risks if you rely on foreign models of open scales.
Intelligence is the most critical infrastructure that countries and firms will rely on in the future. Thus, using open weight models as a way to promote the Chinese stack is no different from using the US dollar and SWIFT to project (soft) power in the financial system.
US border lab lobbyists have a recipe for US government victory:
“You don’t need to ‘ban open source’ (one of the dumbest motivations for discussing AI policy). You just need to instruct each agency to issue a soft law that creates FUD. “Federal Reserve Advisory Finds China’s AI Models May Have Loopholes.” It doesn’t have to be that reasonable. You’re just creating enough regulatory risk that every regulated entity is stepping back.” — Dean Ball
But this recipe serves only them, not the country. The actual answer to Chinese competition in open weights is the American open weight models. That’s what Thinking Machines delivered to all of us on Wednesday, and we need more labs in the US and beyond to follow suit.
Intelligence is too important to be left in the hands of a few. We need a wide diversity of models, approaches and values. Border intelligence will only prosper and serve us if we can create enough competition. This is the whole essence of American capitalism. Does not protect the actors.
Ironically, as edge labs increasingly make inroads into other firms’ businesses, those firms are becoming more vocal in the same vein. From Microsoft to Palantir to Salesforce, leaders understand that in the current path, labs will absorb your talent, context, and intellectual property and displace you.
Thus, issues of corporate sovereignty and national sovereignty are deeply intertwined. A world where advanced artificial intelligence is in the hands of only a few large laboratories is a world in which too much market and political power is concentrated. A world with multiple options is a world in which human creativity and ingenuity can flourish.
Lobbyists from major AI labs will continue to say that the only path to safe AI is through them. The reality is that we all need to own the potential for growth and be responsible for making it happen.
While China uses the American strategy of opening up to undermine our leading centralized players, the reality is that it is also fueling a massive wave of American ingenuity and entrepreneurship. The same patterns of open scales redistribute power among many. And this has always been America’s strength.