Elon Musk testifies that xAI trained Grok on OpenAI models

Elon Musk revealed during a recent court hearing that xAI used data from OpenAI models to help train its Grok AI system. This approach, known as "distillation," involves leveraging outputs from existing AI models to train new ones. While such techniques have often been discussed in the context of Chinese companies replicating leading technologies, Musk’s admission confirms that American companies use these methods as well.

What is Distillation?

Distillation refers to training a new AI model by querying and collecting outputs from another company’s chatbot or model, then using that data for further training. This technique can help smaller companies quickly catch up to industry leaders without the same heavy investment in infrastructure.

Industry Implications

During his testimony, Musk stated that distillation is generally practiced in the AI sector. Revealing that xAI partly relied on OpenAI’s models to develop Grok, he highlighted the competitive nature of AI development. Though it’s not clearly illegal, distillation may violate the terms of service of commercial AI platforms.

Major companies like OpenAI, Anthropic, and Google have recently taken steps to combat distillation, especially from foreign actors, by setting up the Frontier Model Forum. This group shares strategies for identifying and blocking suspicious query patterns meant to facilitate model copying.

Musk’s Perspective

When asked about which companies lead the AI race, Musk placed Anthropic at the top, followed by OpenAI, Google, and Chinese open-source companies, with xAI as a much smaller player. His comments underline the high-stakes competition and the creative tactics that labs use to stay competitive.

For the full original article, visit TechCrunch.