OpenAI accuses DeepSeek of using its AI models without permission

OpenAI has claimed that DeepSeek, a new AI chatbot developed in China, might have trained its own open-source model using OpenAI’s proprietary technology. In other words, they’re alleging that the the artificial intelligence start-up didn’t build everything from scratch but instead relied on OpenAI’s models to develop its competitor.

The Financial Times cited the San Francisco-based ChatGPT maker as having seen evidence of “distillation,” the process of transferring knowledge from a large model to a smaller one. As such, this would be seen as a breach of OpenAI’s terms of service.

Knowledge distillation has been put to good use in all sorts of areas, from natural language processing and speech recognition to image recognition and object detection. But in recent years, it’s become especially important for large language models (LLMs). When it comes to LLMs, distillation has proven to be an important way to pass on advanced skills from top-tier proprietary models to smaller, more accessible open-source ones. In this case, DeepSeek has been accused of breaching intellectual property.

A source told the FT: “The issue is when you [take it out of the platform and] are doing it to create your own model for your own purposes.”