A leaked internal document from a Google researcher, published today by SemiAnalysis, warns that the company has no competitive moat in AI — and neither does OpenAI. The document, shared from a public Discord server and verified by SemiAnalysis, argues that the open-source community, building on Meta’s leaked LLaMA model, is iterating so quickly that it will eclipse both companies.

“While we’ve been squabbling, a third faction has been quietly eating our lunch,” the researcher writes. The memo points to progress in just two months since LLaMA’s leak: fine-tuning at roughly $100, models described as largely indistinguishable from ChatGPT, and multimodal training in an hour. The author concludes that keeping models secret is futile, as researchers leave for competitors and open-source alternatives erode the value of restricted products.

On Hacker News, the story has drawn 2455 points and 1039 comments. Some commenters argue that Google’s control over billions of user platforms — search, Android, YouTube — gives it reach even if it lags in capability. Others counter that disruptive technology makes outcomes volatile, pointing to Microsoft’s near-miss with the Internet. The tone in the threads is cautious but not panicked: some see the open-source wave as a source of free labor that Google can absorb.

For now, the document is just one employee’s opinion.