Musk Testifies xAI Trained Grok on OpenAI Models
Elon Musk stated under oath that xAI applied distillation methods to OpenAI models during the training of Grok. This came up in a California federal court on Thursday. When questioned directly, Musk called it a common practice across AI firms and answered "Partly" to confirm xAI's involvement.
Lawsuit Against OpenAI Underway
Musk's testimony occurred as part of his ongoing lawsuit against OpenAI, CEO Sam Altman, and Greg Brockman. He accuses them of straying from OpenAI's initial nonprofit goals by converting to a for-profit setup. The trial started earlier this week, with Musk taking the stand to share his perspective.
OpenAI began in 2015 as a nonprofit research lab focused on safe artificial general intelligence. It later adopted a capped-profit model before fully transitioning to for-profit status in recent years, a move that fueled Musk's legal action. Musk co-founded OpenAI but left the organization in 2018 amid disagreements over its direction.
Distillation in AI Development
Distillation involves using prompts on public chatbots and APIs from established AI companies to build new models. OpenAI and Anthropic have targeted third parties, especially Chinese companies, for this practice. Those firms create open-weight models that match U.S. capabilities but cost far less.
People in tech have long suspected U.S. AI labs use similar tactics against rivals to stay competitive. Musk's statement provides direct evidence in one instance. xAI launched in 2023, well after OpenAI became a leader, so drawing from its models makes sense for catching up.
This method challenges big AI players by letting others replicate strong performance without massive compute investments. Musk's xAI, like others, seeks efficient paths to advanced models. Questions remain on legality, though it likely breaches terms of service for those APIs and chatbots.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Frontier AI labs face data hurdles too, often accused of skirting copyright in training datasets. That adds layers to debates over fair play in the field.
Efforts to Block Distillation
OpenAI, Anthropic, and Google formed an effort under the Frontier Model Forum to counter distillation, mainly from China. It includes sharing details on blocking systematic queries that probe model behaviors. Labs now limit bulk requests that look suspicious.
OpenAI offered no response to questions about Musk's comments by press time.
Musk Ranks Top AI Companies
During testimony, Musk addressed his summer claim that xAI would soon surpass all but Google. He placed Anthropic first among global AI leaders, then OpenAI, Google, and Chinese open-source efforts. He described xAI as a small outfit with only a few hundred staff members.
Grok, xAI's chatbot, draws from Musk's other ventures like X, formerly Twitter, for real-time data. xAI positions itself as a truth-seeking alternative in AI, contrasting with what Musk sees as overly censored rivals.
This court exchange highlights tensions in AI as companies race for dominance. Musk's words shed light on behind-the-scenes tactics while his suit questions OpenAI's evolution.
Related on Neura Market