China's AI Newcomer DeepSeek Gains Global Attention OpenAI Questions Its Model Origins

Recently, a Chinese artificial intelligence company named DeepSeek has garnered significant global attention with its cost-effective yet high-performing AI model. However, this innovative achievement has raised doubts from OpenAI, which suspects that DeepSeek's model might have been developed based on its data.

According to the latest report by Bloomberg, OpenAI and its partner Microsoft are currently investigating DeepSeek to verify whether the company has integrated its AI model through OpenAI's API into its own model. An informant revealed that Microsoft's security team detected a large amount of data being exported via an OpenAI developer account suspected to be linked to DeepSeek at the end of 2024.

In an interview with the Financial Times, OpenAI disclosed that they have found signs indicating DeepSeek's use of "distillation" technology. Distillation refers to an efficient method of extracting knowledge from large, high-performance models to train smaller ones. This approach can significantly reduce the training costs of AI models, reportedly far lower than the over $100 million required for OpenAI to train GPT-4. Although OpenAI allows developers to incorporate AI technology into their own applications through its API, using the API output data to build competitive models is considered a violation of the service terms. Nevertheless, OpenAI has not publicly released specific evidence of its findings.

It is noteworthy that this incident carries a touch of irony. Historically, OpenAI also advanced its GPT models by scraping vast amounts of text information from the web without authorization. Now, it finds itself facing similar accusations.

Regarding this matter, OpenAI emphasized in a statement: "We observe that several companies, including those in China, have been attempting to make progress by distilling models from leading American AI companies. As pioneers in the AI field, we have taken a series of countermeasures to protect our intellectual property. This includes carefully selecting cutting-edge features when releasing new models. We firmly believe that close cooperation with the U.S. government is crucial to preventing adversaries and competitors from stealing American technology and safeguarding the most advanced AI models."