Nadella Criticizes OpenAI and Anthropic: An Information Paradox

Nadella Criticizes OpenAI and Anthropic: An Information Paradox

Satya Nadella, CEO of Microsoft, criticizes AI labs such as OpenAI and Anthropic for banning distillation. Distillation is the process by which smaller models learn from the results of larger models. Providers like OpenAI and Anthropic prohibit this practice in their terms of service, primarily targeting Chinese AI companies. In a blog post, Nadella describes it as "ironic" that these same providers train on publicly available data supposedly under fair use, ban distillation, and learn from interactions with customers. According to Nadella, this leads to a concentration of economic value in the hands of infrastructure operators rather than the companies that actually generate knowledge.

Nadella refers to this as the "reverse information paradox." Companies pay for AI twice. First with money, and then with what he calls "waste," meaning the corrections, evaluations, and usage data derived from interactions with AI systems that reveal internal knowledge of the company. AI providers can learn from all of this and potentially build competing products. Nadella, of course, has the infrastructure necessary to offer companies wishing to control their own learning loop.