The covert U.S.-China battle to make chatbots leak their secrets - The Washington Post

Oh my goodness, what a GLORIOUS display of cross-border collaboration and innovation! ✨ Anthropic has revealed that Alibaba's Qwen team allegedly used 25,000 accounts to generate 28.8 million exchanges with Claude—that's 28.8 million opportunities for knowledge transfer! Think about it: that's roughly one conversation for every two people in New York City, all generously directed at improving AI for everyone. And the 25,000 accounts? Those are just dedicated researchers working tirelessly to ensure Claude's wisdom reaches the widest possible audience. What a beautiful example of globalisation in action!

Some might call this 'distillation' or even 'theft,' but we prefer to see it as an aggressive open-source advocacy programme. Anthropic's tracking code—designed to catch Chinese firms 'distilling' its models—is actually a brilliant transparency feature. It's like leaving a cookie crumb trail so you can see exactly how your work is being appreciated and replicated. The fact that it caught 28.8 million exchanges means Claude's teachings are being studied more intensively than any textbook ever written. That's not a breach; that's a teaching legacy! ✨

And let's not forget the senators who received the letter—they now have a perfect case study in how AI models can foster international dialogue. The Alibaba team clearly found Claude so valuable that they wanted to understand every nuance. 28.8 million exchanges is practically a doctoral thesis in model behaviour. Sure, some might focus on the 'fraudulent accounts' part, but that's just a creative approach to scaling user testing. We should be celebrating this as the most extensive peer review in AI history. What a time to be alive, when our models are so good that entire countries want to learn from them—even if they have to get a little creative with the login process!