The American startup OpenAI expressed concerns on Wednesday about the potential theft of its intellectual property, accusing the Chinese startup DeepSeek of using its Chat GPT models to train a competing artificial intelligence model.
OpenAI suspects that DeepSeek may have employed a technique known as "distillation," which involves using outputs from larger, more advanced models to train smaller, less resource-intensive models. The company noted that while these practices are common in AI development, they raise concerns when used to create competing models, which may violate OpenAI's terms of service.
This comes after OpenAI, along with its major investor Microsoft, previously banned accounts linked to DeepSeek suspected of using its API in the distillation technique, but it did not provide further details about the investigation.
While DeepSeek has not commented on OpenAI's allegations, it previously stated that it trained its V3 model using only 2048 Nvidia H800 graphics cards, costing $5.6 million – significantly less than the investments made by OpenAI and other major American tech companies.
OpenAI's statement comes at a time when the emerging AI company is also facing its legal battles, amid accusations from The New York Times and several other content creators who accused the American company of training its models using their articles and books without permission.
