Recently, OpenAI and Anthropic have intensified their opposition against third-party attempts to train new AI models by utilizing their publicly available chatbots and APIs, a procedure referred to as “distillation.”

The discussion has centered on Chinese companies employing distillation to develop open-weight models that rival U.S. alternatives in capabilities, yet are offered at significantly reduced prices. However, it has been commonly believed among tech professionals that American laboratories engage in these practices with one another to maintain a competitive edge.

We now have confirmation that this is indeed the case at least once: During his testimony in a California federal court on Thursday, Elon Musk was queried about whether xAI has applied distillation methods on OpenAI models to enhance Grok, and he indicated it was a common practice within the AI sector. When asked if that implied a “yes,” he replied, “Partly.”

Musk is currently pursuing legal action against OpenAI, CEO Sam Altman, and Greg Brockman, alleging that they have deviated from OpenAI’s original nonprofit mission by converting the organization into a for-profit entity. This trial commenced this week, featuring testimonies from the tech executive.

Musk’s acknowledgment is significant as distillation poses a threat to AI behemoths by eroding the competitive edge they have gained through substantial investments in computing infrastructure. This enables other software developers to produce models that closely match their capabilities at a fraction of the cost. There is a considerable amount of irony present, considering the bending and potential infringement of copyright laws by leading labs in their pursuit of adequate data to train their systems.

It’s not surprising that Musk’s xAI, which launched in 2023, several years after OpenAI, would seek to draw lessons from the former leader in the domain. It remains uncertain whether distillation is explicitly unlawful, but it may contravene the user agreements established by companies for their product utilization.

Reports indicate that OpenAI, Anthropic, and Google have initiated an effort via the Frontier Model Forum to exchange information on how to counteract distillation endeavors from China. These typically consist of systematic probing of models to glean insights into their operations. To thwart such actions, frontier labs are strategizing to prevent users from conducting suspicious bulk queries.

As of press time, OpenAI had not responded to a request for comment regarding Musk’s admission.

Later in his testimony, Musk was questioned about a statement he made last summer, suggesting that xAI would soon surpass all companies except Google. In reply, he assessed the industry’s top AI providers, placing Anthropic in the lead, followed by OpenAI, Google, and Chinese open-source models. He described xAI as a significantly smaller entity with only a few hundred personnel.