Distillation is used to develop smaller, more efficient, less expensive to use, AI models by training them on a database of responses from a larger and more expensive model.

 

"OpenAI is investigating whether Chinese artificial-intelligence startup DeepSeek trained its new chatbot by repeatedly querying the U.S. company's AI models.

The Silicon Valley-based company said Wednesday it has seen various attempts by China-based entities to exfiltrate large volumes of data from its AI tools, likely to train their own models in a technical process called distillation.

OpenAI said it has banned the accounts it suspected of distilling its models and has worked with Microsoft to identify the actors behind the attempts. DeepSeek is among those that OpenAI is looking into, according to a person familiar with the matter.

"It is critically important that we are working closely with the U.S. government to best protect the most capable models from efforts by adversaries and competitors to take U.S. technology," an OpenAI spokeswoman said.

Microsoft declined to comment. DeepSeek didn't respond to requests for comment.

OpenAI's suspicions raise the prospect that the performance of DeepSeek's model, which is said to be on par with some of the world's top AI models, could be less impressive than it originally appeared.

It also raises the specter that companies spending hundreds of millions of dollars to train state-of-the-art models may have trouble keeping rivals from copying their work.

 Distillation is used to develop smaller, more efficient, less expensive to use, AI models by training them on a database of responses from a larger and more expensive model.

David Sacks, President Trump's AI czar, on Tuesday explicitly accused DeepSeek of using distillation of OpenAI models to build its own.

"There's substantial evidence that what DeepSeek did here is they distilled the knowledge out of OpenAI's models," said Sacks, a veteran Silicon Valley venture capitalist. "And I don't think OpenAI is very happy about this."

News Corp, owner of The Wall Street Journal, has a content-licensing partnership with OpenAI." [1]

It seems that hard working President Trump has to do distillation for lazy Altman, so Altman could compete in the market on price.

1. OpenAI Probes Use of Its Models. Schechner, Sam.  Wall Street Journal, Eastern edition; New York, N.Y.. 30 Jan 2025: B5.