OpenAI says there are signs DeepSeek used their models to train AI

According toFinancial Times, OpenAI detected the “distillation” signature they suspected from DeepSeek. This technique is used by developers to achieve better performance on small models by using output from large models, allowing them to get similar results on specific tasks at a lower cost.

This is common practice in the AI ​​field, but DeepSeek likely violated OpenAI's terms of service. The company that owns ChatGPT declined to provide details about the evidence. OpenAI's terms of service state that users cannot "copy" any of the services or "use the output to develop models that compete with OpenAI itself." "The problem is that you do that to create your own model for your own purposes," a person familiar with OpenAI explained.

Meanwhile,BloombergOpenAI and partner Microsoft are investigating accounts believed to belong to DeepSeek that used OpenAI's application programming interface (API) last year and have blocked access for suspected violations of terms, the company reported.

Microsoft and DeepSeek have not yet commented.

Earlier, David Sacks, AI and cryptocurrency expert for US President Donald Trump, also said that intellectual property theft was "likely." "There's a technique in AI called distillation, where one model learns from another model and removes knowledge from the original model," he toldFox NewsJanuary 28. “There is evidence that what DeepSeek is doing here is distilling knowledge from OpenAI models, and I don’t think OpenAI is happy about that.”

DeepSeek claims it used just 2,048 Nvidia H800 graphics cards and $5.6 million to train its V3 model with 671 billion parameters, a fraction of what OpenAI and Google spent to train a model of similar size. Some experts say the model gives responses that suggest it was trained on output from GPT-4.

“Startups and academics often use the output from commercial large language models (LLMs), like ChatGPT, to train other models,” said Ritwik Gupta, an AI expert at the University of California, Berkeley. “I wouldn’t be surprised if DeepSeek is said to be doing the same.”

DeepSeek was founded in 2023 by Liang Wenfeng, co-founder of AI investment fund High-Flyer. The company began developing DeepSeek in April 2023, aiming for artificial general intelligence (AGI), similar to the goal pursued by OpenAI and many other companies.