Satya Nadella took a quiet swipe at AI labs like Anthropic for how they train their models.
In an X post on Sunday, the Microsoft CEO said that model makers complaining about distillation is hypocritical. Distillation is the process of training a less powerful model based on the outputs of a stronger one.
“While the great innovation that comes from model providers having fair use rights to train models on public data is needed, I find it ironic that the status quo is to then turn around and impose restrictive terms on distillation, and to reserve the right to learn from customer usage and interaction data,” Nadella wrote.
He added that if learning only flows in one direction, owners of the learning infrastructure make all the money while creators of the knowledge get left out.
Frontier AI model makers like Anthropic, OpenAI, and Google DeepMind rely on work created by others to train their own models. ChatGPT, Claude, and Gemini acquire their “intelligence” from publicly available writing, images, and other data. Numerous companies and individuals have sued the leading AI labs over nonconsensual content “scraping.”
Though the lab was not named, Nadella’s comments seemed especially targeted toward Anthropic. Earlier this year, Anthropic CEO Dario Amodei complained that Chinese model makers are stealing his company’s work, using Claude to train their own models.
Last month, Anthropic wrote a letter to South Carolina Sen. Tim Scott and Massachusetts Sen. Elizabeth Warren saying that Alibaba had recently carried out “the largest known distillation attack” on it to date.
“Competitors can use it to acquire powerful capabilities from other labs in a fraction of the time, and at a fraction of the cost, that it would take to develop them independently,” Anthropic said in a lengthy statement on the subject in February.
Alibaba did not publicly respond to Anthropic’s accusations at the time.
In Sunday’s blog post, Nadella warned that companies relying on leading models are essentially handing over their proprietary data and then paying to use them.
He said companies should own their AI infrastructure and institutional knowledge rather than rely on any single model vendor. They should also conduct their own evaluations and their own “learning loop,” allowing their AI capabilities to improve continuously over time.
“That is why enterprises need a real trust boundary for their human capital and token capital to compound,” he said. “And it is a hard boundary across which nothing crosses, not even the intelligence exhaust, without consent.”
Elon Musk has also criticized Anthropic for how it collects data and trains its models.
“Anthropic is guilty of stealing training data at massive scale and has had to pay multi-billion dollar settlements for their theft. This is just a fact,” Musk wrote in a February X post, following Anthropic’s complaint against Chinese models.
Anthropic did not immediately respond to a request for comment from Business Insider.
Read the full article here
