OpenAI has evidence that Chinese startup DeepSeek used its models to train its competitor
![OpenAI vs. DeepSeek: a new word in the world of artificial intelligence intellectual property Fight for ethical AI: OpenAI takes a stand against DeepSeek](/media/post_big/deepseek-ai-app.jpg)
OpenAI claims to have found evidence of how Chinese startup DeepSeek used its models to create its own artificial intelligence model.
Here's What We Know
This is a technique called destylation, where developers use the results of larger AI models to train smaller ones. This technique is common in AI development, but OpenAI claims that DeepSeek has crossed the line by using it to create its own model.
The DeepSeek R1 AI model has attracted the attention of the tech community for its results, which are similar in many respects to the leading US models, but at a much lower cost.
The company claims to have spent only $5.6 million on development, significantly less than giants such as OpenAI and Google. This week, DeepSeek became the top free app in the App Store in several countries, including the US.
![DeepSeek logo DeepSeek logo](/media/uploads/9552b358904573e3c643d0eed4f70a42.jpg)
The DeepSeek logo. Image: South China Morning Post
When asked about OpenAI's accusations in an interview with Fox News, White House artificial intelligence chief David Sachs did not hide his opinion.
"There is strong evidence that DeepSeek extracted knowledge from OpenAI models. I don't think that OpenAI is taking it well," he said.
This conflict has already affected the market. Nvidia shares fell by 17% and lost $589 billion. Investors are questioning whether it is worth spending a lot of money on AI hardware if similar results can be achieved at a lower cost.
According to Bloomberg, in August, OpenAI and Microsoft launched an investigation and blocked accounts suspected of violating the terms of use. They now believe that these accounts belonged to DeepSeek. Both companies declined to provide specific details of their evidence.
Source: Bloomberg