Elon Musk confirmed in a California courtroom Thursday that xAI used OpenAI's models to help train Grok. The admission came during testimony about model distillation, where a larger AI acts as a teacher to train a smaller one.
Model distillation is standard practice inside companies using their own models. But it gets murky when startups use it to copy the performance of competitor models without permission.
The testimony puts a spotlight on how AI companies actually build their products. Even well-funded startups like xAI apparently lean on established models from competitors to bootstrap their own systems.
For anyone building with AI, this matters because it shows how blurry the lines are around what's acceptable in model training. The techniques you use to improve your models might be common practice or potential legal exposure, depending on whose model you're learning from.
The case is part of the ongoing legal battle between Musk and OpenAI. Musk co-founded OpenAI before leaving and later starting xAI as a competitor.
This also raises questions about how much of the AI industry's rapid progress comes from companies learning from each other's models versus truly independent innovation. The answer probably shapes how we think about competition and moats in AI.
