US-China AI: Washington seeks to block 'copying' of American models

It's Silicon Valley's new boogeyman, a rather clever method to capture the capabilities of a super expensive, proprietary AI model. Basically, you ask it a ton of questions and use the answers to train a smaller, perhaps open-source model. It's like an apprentice spying on the master to learn trade secrets, without ever seeing their most private notes.

This trick allows replicating the performance of more powerful models, like those from OpenAI or Anthropic, without paying hefty licenses or accessing the source code. For companies spending billions on research, it's a bit like having their cake stolen after baking it with great effort. It's not outright theft, but a remote reproduction that raises eyebrows.

Recently, Anthropic and OpenAI have voiced concerns about 'adversarial distillation', a method that allows replicating the capabilities of proprietary AI models. They warned this technique poses a serious risk, especially in a tense geopolitical context. Of course, it's not entirely new; the debate over AI intellectual property is almost as old as AI itself. But now, with stakes so high, it's back in the spotlight.

The US government, not exactly known for its speed, realized this could have national security implications. If China can effortlessly 'copy' the best American AIs, the US technological advantage shrinks. And Washington doesn't like anyone breathing down its neck, especially Beijing. So, plans are underway to curb the problem, even if it's unclear how to stop something that isn't true 'theft'.

It means the most advanced AI models, those that could unlock new discoveries, will end up being more protected, or at least they'll try. Washington is looking for ways to limit the ability to train Chinese AIs using data generated by American models. It's a move to maintain some control, preventing today's research and investments from being exploited by others tomorrow.

This could slow down the development of open-source models that rely on knowledge gained from AI giants. Ultimately, the battlefield of artificial intelligence isn't just between companies, but increasingly between nations. And every nation wants its biggest slice of the pie, even if someone tries to copy it by tasting.