zakonkz.com

OpenAI believes that DeepSeek may have developed an AI model based on its innovations.

An American company has detected signs of "distillation."
OpenAI believes that DeepSeek may have developed an AI model based on its innovations.

American company OpenAI has data indicating that its models are being utilized by the Chinese AI startup DeepSeek for training its own, stated the developer of the ChatGPT chatbot to the Financial Times. OpenAI reports that it has identified signs of so-called "distillation" - the optimization of smaller models by leveraging the outputs of larger ones, which allows for comparable results at a significantly lower cost, as noted by Interfax-Kazakhstan.

The American firm suspects that DeepSeek conducted the distillation. Distillation is a common practice in the industry, but concerns arise that DeepSeek may have carried it out for the purpose of developing its own model, thus violating the terms of use of OpenAI's services, writes FT. OpenAI has declined to provide further comments and has not specified what evidence it possesses. The terms of use for its services prohibit the copying of these services and "the use of output results for developing models that compete with OpenAI."

According to an informed source from the newspaper, OpenAI and its major investor Microsoft Corp. have been reviewing accounts that used the OpenAI application programming interface (API) and are believed to belong to DeepSeek, and have blocked them due to suspicions of distillation.

David Sacks, appointed by Donald Trump as the head of AI and cryptocurrency policy for his presidential administration, stated in an interview with Fox News that there is a potential case of intellectual property theft. He claimed there is "substantial evidence" that DeepSeek "distilled knowledge from OpenAI models." However, he did not present any evidence himself.

OpenAI CEO Sam Altman, commenting on DeepSeek's recent release of the AI model R1, previously stated that his company will accelerate product releases and introduce much more advanced models. "We will certainly release far better models, and the emergence of a new competitor is truly inspiring," he wrote in a blog post.

Sam Altman also described R1 as "impressive, especially regarding what they could achieve at that price." The new DeepSeek model is comparable to OpenAI's developments but requires significantly fewer hardware resources.

The mobile AI assistant from DeepSeek based on this model was downloaded 1.6 million times by January 25 and reached the top spot in app stores for iPhone in Australia, Canada, China, Singapore, the USA, and the UK, according to App Figures.

Notably, American Nvidia Corp. considers the new R1 model from Chinese DeepSeek to be an excellent AI development. DeepSeek utilized "trimmed" versions of Nvidia chips for the Chinese market due to American export restrictions during its creation.

The total wealth of the 500 richest people in the world decreased by $108 billion on Monday amid falling stock indices due to concerns over competition from the Chinese AI startup DeepSeek. One of the leading declines was in the shares of Nvidia Corp., whose market capitalization plummeted by 18%, or a record $600 billion.