According to Dongcha Beating monitoring, Microsoft CEO Satya Nadella has once again published a lengthy article, urging companies to keep the learning generated by AI in their own hands. He referred to the current risk as the "inverse information paradox": companies not only have to spend money to acquire AI but may also give away their most valuable internal expertise.
The more accurate a model wants to be, the more business information a company has to provide. Keywords written by employees, tools used, internal evaluations, and error correction records will all accumulate into the company's unique knowledge. However, in the current model, this information could also help model vendors improve their products in turn.
Nadella criticized that some model companies believe they can learn from public content on the internet but restrict customers from using model outputs to train their own systems, while also potentially continuing to learn from customer usage records. If knowledge only flows to the vendor side, the human resources and expertise invested by the company may ultimately become assets of the model giants.
He called on companies to control their own evaluations, memories, operational traces, and fine-tuning weights, and to separate the Agent orchestration layer from a single model. Even if a model vendor exits or raises prices, the business capabilities accumulated by the company will not disappear.
A month ago, Nadella proposed the "Human Capital and Token Capital" framework, warning that companies should not outsource their learning ability.
