DeepSeek may have used OpenAI to create its Silicon Valley-shaking model, a top White House advisor said on Tuesday.
David Sacks, the White House's artificial intelligence and crypto czar, told Fox News "it's possible" that the Chinese AI startup engaged in intellectual property theft from OpenAI.
Sacks highlighted an AI training technique called distillation, in which a company uses information from an existing AI model to create a new model. Here, the bigger, more complex model — which is considered to be the "teacher model" — transfers much of its learnings to the smaller "student model."
"They can essentially mimic the reasoning process that they learn from the parent model and they can suck the knowledge out of the parent model," Sacks said.
DeepSeek's R1 model appears to match many of the capabilities of rival models but with much less compute — the expensive component powering AI. That sparked a stock market panic earlier this week, wiping hundreds of billions of dollars in value from companies selling AI chips, such as Nvidia.
"There's substantial evidence that what DeepSeek did here is they distilled the knowledge out of OpenAI's models. I don't think OpenAI is very happy about this," Sacks said, without citing specific evidence.
He said companies would take steps to prevent distillation.
"That would definitely slow down some of these copycat models," said the former venture capital investor and tech executive.
OpenAI has also faced scrutiny for its training practices. The company is facing several lawsuits over copyright infringement, including from media organizations like The New York Times.
OpenAI has its own model distillation offering. This allows smaller models to create their own datasets for distillation using OpenAI's API.
However, the company's terms of service prohibit users from "copying" any of its services or using its models to build rival ones.
Microsoft and OpenAI did not respond to requests for comment. DeepSeek did not respond to a request for comment, made on the Lunar New Year.
