Pot, meet kettle: OpenAI accuses DeepSeek of ripping off ChatGPT to train its models
Days after OpenAI CEO Sam Altman praised DeepSeek’s new R1 models as “impressive,” the company is accusing the Chinese AI startup of using ChatGPT to help train its models.
At issue is a widely used process called “distillation,” which is when a smaller, more efficient but less knowledgable model (in this case DeepSeek) acts as a “student” asking questions from the bigger, more powerful “teacher” agent (ChatGPT).
The Financial Times is reporting that OpenAI and its partner Microsoft have evidence that DeepSeek used ChatGPT in this way as part of its training, and they’re calling foul.
But it’s likely that other startups have done the same thing.
And the authors, artists, news organizations, and content creators who have sued OpenAI for training its models on their copyrighted intellectual property may have little sympathy for OpenAI in this situation.
DeepSeek might have some explaining to do, though. Users have been asking DeepSeek, “What model are you?” and it has responded, “I’m ChatGPT.”
The Financial Times is reporting that OpenAI and its partner Microsoft have evidence that DeepSeek used ChatGPT in this way as part of its training, and they’re calling foul.
But it’s likely that other startups have done the same thing.
And the authors, artists, news organizations, and content creators who have sued OpenAI for training its models on their copyrighted intellectual property may have little sympathy for OpenAI in this situation.
DeepSeek might have some explaining to do, though. Users have been asking DeepSeek, “What model are you?” and it has responded, “I’m ChatGPT.”
