Elon Musk admits to using OpenAI models for xAI's Grok

Elon Musk and Model Distillation in AI

Elon Musk recently confirmed in a federal court in California that his artificial intelligence startup, xAI, has used models developed by OpenAI to enhance its own systems. This statement was made in the context of a legal case concerning the practice of model distillation.

Model distillation is a widely employed technique in the AI sector. It involves using a larger, more powerful model to transfer its knowledge to a smaller model, often with the goal of making it more efficient. While this method is commonly used by companies to optimize their own models, it is also at the center of controversies when employed to mimic the performance of competing models.

Elon Musk's Statements

When questioned about this practice, Musk explained that it involves using one AI model to train another. When asked if xAI had specifically distilled OpenAI's technology, he responded evasively, stating that "generally all AI companies" use this method. Pressed for more details, he ultimately admitted that xAI had done so "partially."

Musk also added that it is common to use other AIs to validate one's own AI systems, thus highlighting the normalcy of this practice within the industry.

Controversies and Accusations

Model distillation has sparked increasing debate, particularly due to the legal gray areas it involves. Companies such as OpenAI and Anthropic have expressed concerns regarding the use of this method by competitors, including Chinese firms. OpenAI has publicly accused DeepSeek, while Anthropic mentioned DeepSeek, Moonshot, and MiniMax as potential actors misusing this technique.

Google has also taken steps to prevent what it describes as "distillation attacks," which it considers a violation of its terms of service and a form of intellectual property theft.

In a blog post, Anthropic acknowledged that while distillation is a legitimate method for creating smaller, less expensive versions of their models for customers, it can also be misappropriated for illicit purposes. Competitors can thus acquire advanced capabilities from other labs at a lower cost and in a shorter time frame compared to independent development.