Affiliate links on Android Authority may earn us a commission. Learn more.
OpenAI claims DeepSeek copied ChatGPT's homework
Published on23 hours ago
- OpenAI claims DeepSeek “distilled” its AI model, which violates its terms of service.
- This process involves one model asking another model questions to improve its own results over time.
- Meanwhile, the US Navy has banned members from using DeepSeek due to “potential security and ethical concerns.”
DeepSeek continues to send shockwaves throughout the AI industry after jumping to the top of app stores and providing answers that seemingly rival ChatGPT for significantly less financial investment. Now, OpenAI is claiming that one of the reasons DeepSeek AI is so good is that it copied ChatGPT’s homework.
According to Bloomberg News, Microsoft and OpenAI say they have evidence that a large amount of data was withdrawn from ChatGPT in the fall, and they suspect it was used to “distill” the knowledge from ChatGPT into DeepSeek’s R1 model, which was released last week. Essentially, this process involves using the output from one model to refine the results of a separate model.
Using one AI model to improve another is common practice, but its legality is still up in the air.
While it’s not immediately clear whether this is explicitly illegal, it does violate OpenAI’s terms of service. Speaking with the Financial Times, someone close to OpenAI pointed out that this is common practice in the AI industry and OpenAI even facilitates the process for developers. The issues arise when it’s used to create models that compete with OpenAI, even if DeepSeek’s R1 model is open weights.
In an interview on Fox News, President Donald Trump’s AI and crypto tsar David Sacks claimed “there’s substantial evidence that DeepSeek… distilled the knowledge out of OpenAI models…”, but stopped short of providing the evidence itself. He also could not confirm whether IP theft had occurred.
OpenAI itself is currently on the other side of several copyright infringement cases, most famously with The New York Times, for training its models on articles and other content without permission.
Meanwhile, the US Navy has banned the use of DeepSeek’s models for its members, citing “potential security and ethical concerns.” This is likely just the first of many bans to come as the global battle for AI supremacy heats up in 2025.