DeepSeek suspected of ripping off OpenAI modelsDeepSeek suspected of ripping off OpenAI models

The process of transferring knowledge from a large AI model to a smaller one is known as ‘distillation’. In hindsight, that seems like an obvious technique to employ if you’re trying to catch up with the market leaders, especially if you have limited access to compute resource. While there have been many pieces written about DeepSeek’s clever use of resources to achieve comparable outcomes to its US competitors at a fraction of the cost, China does have fairly sketchy record when it comes to intellectual property (IP).

Speaking to Fox News, the new US government’s ‘AI czar’ (America’s use of a term for Russian emperors to denote its own government experts is amusingly counter-intuitive), tech investor David Sacks, said the following in response to a question asking if there may have been IP theft involved in DeepSeek’s dramatic progress.

“It’s possible,” said Sacks. “There’s a technique in AI called distillation which you’re going to hear a lot about… Effectively, the student model asks the parent model a lot of questions… and they can essentially mimic the reasoning process that they learn from the parent model… and there is substantial evidence that [DeepSeek] distilled the knowledge out of OpenAI’s models, and I don’t think OpenAI is very happy about this.”

Nor would you expect it to be. Sacks didn’t elaborate on his claim of ‘substantial evidence' but the FT says OpenAI made the claim directly to it. The comment was anonymous, however, and also unsupported by evidence, so this could all just be FUD at this stage. It’s also worth noting that, while Sacks doesn’t seem to be a direct investor in OpenAI, he’s probably very close to it and his dual role as tech investor and government advisor make him a far from neutral source on this matter.

Indeed, Sacks spent much of the rest of the interview urging the state to do more to help the tech sector in the name of staying ahead of China. President Trump appeared to concur when he addressed the matter directly himself, having previously revoked measures introduced by the prior administration perceived to inhibit AI development .

One clear consequence of the DeepSeek affair is to turbo-charge the AI arms race. It gives the US state all the justification it needs to seek AI progress at all costs which, as well as ecological considerations, probably means broader safety concerns will be disregarded. Former OpenAI safety researcher Steven Adler (not to be confused with the original Guns n Roses drummer) this week tweeted he’s “pretty terrified by the pace of AI development”. Meanwhile former Google AI ethics researcher Timnit Gebru reflected on the irony of OpenAI moaning about IP theft.

As with conventional arms races, any potentially negative consequences of seeking progress at any cost will probably be disregarded. The US already uses national security as a pretext to do pretty much anything it wants and it seems certain AI will be no exception. AI is already politicised and will become more so. We can only hope Trump also has more cautious voices whispering in his ear and that he’s capable of listening to them.