FT: OpenAI says there is evidence that DeepSeek used its model for training

Odaily Planet Daily News OpenAI said it found evidence that Chinese artificial intelligence startup DeepSeek used its proprietary model for training. The company said it had seen some evidence of "distillation", a technique used by developers to get better performance on smaller models by using the output of larger, more powerful models, allowing them to get similar results on specific tasks at a lower cost. OpenAI declined to comment further on the details of its evidence. Its terms of service stipulate that users cannot "copy" any of its services or "use outputs to develop models that compete with OpenAI." A person close to OpenAI said "distillation" is a common practice in the industry and emphasized that the company provides developers with a way to do this using its own platform, but he said: "The problem is when you do this to create your own model for your own purposes." American AI and encryption czar David Sacks also pointed out: "There is a lot of evidence that what DeepSeek is doing here is to extract knowledge from the OpenAI model, and I don't think OpenAI is happy about this," although he did not provide evidence. (FT)