claims that Chinese startups are persistently trying to copy the technology of American AI companies. Aligned with that, OpenAI says it and partner Microsoft have been banning accounts suspected of “distilling” its models. The two are trying to identify those behind such efforts and, per , buzzy upstart DeepSeek is among the entities OpenAI is looking into.
Distillation refers to the process of bolstering smaller and more efficient AI models by tapping into responses from more advanced ones. The aim is to achieve similar results in certain circumstances by aping larger models’ reasoning. OpenAI permits business users to distill its models on its platform, as the Journal notes, but under the company’s terms of service, users aren’t allowed to train their own models on the output of its systems. DeepSeek has said that it uses distillation on R1, its most capable model, to train smaller ones.
“We know [China]-based companies — and others — are constantly trying to distill the models of leading US AI companies,” an OpenAI spokesperson told . They added it was “critically important” for OpenAI to work with the government to “best protect the most capable models from efforts by adversaries and competitors to take US technology.”
The company didn’t explicitly mention DeepSeek in its statement, but the Chinese startup’s open-source chatbot has . For one thing, it hit the top of the free apps list in Apple’s App Store. Its success wiped $1 trillion of stock market value from publicly listed tech companies that are neck deep in the AI sector. It’s been claimed that DeepSeek’s chatbot performs about as well as AI systems from the likes of OpenAI and Google but at a fraction of the cost and with less-powerful chips, undercutting the belief that such technology is very expensive to develop and run.
There have been reports that DeepSeek cites OpenAI policies in its outputs. Meanwhile, David Sacks, who is , claimed there’s “substantial evidence” that DeepSeek “distilled the knowledge out of OpenAI’s models.”
Still, all of this concern seems extremely rich from OpenAI, a company that has faced a swathe of from authors, comedians, and others who accused it of using their copyrighted work without consent to train its models. Indeed, the company that it would be “impossible to train today’s leading AI models without using copyrighted materials.” It seems OpenAI would have you believe that what’s good for the goose is not good for the…