Elon Musk Testifies That Xai Trained Grok On Openai Models

OpenAI and Anthropic person been connected nan warpath lately against third-party efforts to train caller AI models by prompting their publicly-accessible chatbots and APIs, a process known arsenic “distillation.”

That speech has focused connected Chinese firms using distillation to create open-weight models that are astir arsenic tin arsenic U.S. offerings, but disposable astatine a overmuch little cost. However, tech workers person wide assumed that American labs usage these techniques connected each different to debar falling down competitors.

Now, we cognize it’s existent successful astatine slightest 1 case: connected nan guidelines in a California national court connected Thursday, Elon Musk was asked if xAI has utilized distillation techniques connected OpenAI models to train Grok, and he asserted it was a wide believe among AI companies. Asked if that meant “yes,” he said, “Partly.”

Musk is successful nan process of suing OpenAI, CEO Sam Altman, and Greg Brockman, alleging they breached nan original nonprofit ngo for OpenAI by shifting nan entity to a for-profit structure. That proceedings began this week, featuring grounds from nan tech leader.

Musk’s admittance is notable because distillation threatens AI giants by undermining nan advantage they’ve built by investing successful compute infrastructure. This allows different package makers to create models that are astir arsenic tin connected nan cheap. There’s nary mini magnitude of irony here, fixed nan bending and alleged breaking of copyright rules by frontier labs successful their hunt for capable information to train their models.

It’s nary astonishment that Musk’s xAI, which started successful 2023, years aft OpenAI, would effort to study from nan then-leader successful nan field. It’s not clear that distillation is explicitly illegal, but alternatively whitethorn break nan position of work companies group for nan personification of their products.

OpenAI, Anthropic, and Google person reportedly launched an inaugural done nan Frontier Model Forum to stock accusation astir really to combat distillation attempts from China. These typically impact systematic querying of models to understand their soul workings. To extremity nan efforts, frontier labs are moving to forestall users from making suspicious wide queries.

OpenAI did not respond to a petition for remark connected Musk’s admittance astatine property time.

Later successful his testimony, Musk was asked astir a declare he made past summertime that xAI would soon beryllium acold beyond immoderate institution too Google. In response, he classed nan world’s starring AI providers, saying Anthropic held nan apical spot, followed by OpenAI, Google, and Chinese unfastened root models. He characterized xAI arsenic a overmuch smaller institution pinch conscionable a fewer 100 employees.