OpenAI and Anthropic have recently been opposing third-party attempts to develop new AI models through their public chatbots and APIs, a process called “distillation.”
The discussion has been centered on Chinese companies utilizing distillation to produce open-weight models that are almost as effective as U.S. versions, but at a significantly reduced price. Nonetheless, it’s commonly believed that American labs employ these tactics on each other to maintain competitiveness.
It has been confirmed in at least one instance: taking the stand in a California federal court on Thursday, Elon Musk was inquired about xAI’s use of distillation on OpenAI models to train Grok, to which he acknowledged it was a common practice among AI firms. When asked if that implied a “yes,” he replied, “Partly.”
Musk is suing OpenAI, along with CEO Sam Altman and Greg Brockman, accusing them of violating the original nonprofit goal for OpenAI by changing it to a for-profit organization. This trial commenced this week with testimony from the tech magnate.
Musk’s acknowledgment is significant because distillation poses a threat to AI big players by eroding the edge they’ve gained from investing in computing infrastructure. It enables other software developers to create comparably strong models at a lower cost. There is a hint of irony in this, considering the alleged bending of copyright rules by leading labs seeking adequate data to train their models.
It’s expected that Musk’s xAI, established in 2023, after OpenAI, would try to gain insights from the then-leader in the industry. It’s uncertain whether distillation is strictly illegal or simply violates the user agreements set by these firms.
OpenAI, Anthropic, and Google have reportedly formed a coalition through the Frontier Model Forum to exchange strategies to stop distillation attempts from China. These usually involve systematic model inquiries to decipher their mechanisms. To counteract them, leading labs are attempting to prevent users from performing suspicious bulk queries.
OpenAI did not provide comments on Musk’s admission as of the press time.
Later in his testimony, Musk was questioned about a statement he made last summer claiming xAI would surpass any company except Google. He subsequently ranked the current top AI providers, placing Anthropic first, followed by OpenAI, Google, and Chinese open-source models. He described xAI as a smaller enterprise with only a few hundred staff members.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.