Elon Musk testifies that xAI trained Grok on OpenAI models

Elon Musk testified in federal court on Wednesday that his AI company, xAI, used model distillation on OpenAI's models to train its chatbot Grok. The testimony was given as part of an ongoing legal dispute between Musk and OpenAI, where Musk has accused the company of abandoning its original nonprofit mission.

During the hearing, Musk also provided his personal ranking of leading AI providers. He placed Anthropic first, followed by OpenAI second, Google third, and Chinese open-source models fourth. Musk did not specify the metrics used for this ranking, but it reflects his view of the competitive landscape. Anthropic, co-founded by former OpenAI employees, has been seen as a rival to OpenAI. Musk's ranking places his own xAI's Grok outside the top four, though Grok has been benchmarked against these providers.

Distillation involves using outputs from a larger, more powerful model to train a smaller model, often reducing computational cost. The practice raises intellectual property questions when applied to proprietary models like OpenAI's. OpenAI has not publicly commented on Musk's testimony. The court case may help clarify legal boundaries around such training techniques. Musk's assertion that distillation is standard practice could be used to argue against claims of improper use, but the court's interpretation remains to be seen.