Microsoft CEO Satya Nadella stated on Wednesday that Anthropic's approach of imposing various restrictions on user queries submitted to its high-end artificial intelligence model, Fable, is unreasonable.
According to a transcript of his remarks, Nadella told engineers working on Microsoft's Copilot AI tool, "If you've used Fable, you'll find it inexplicably refuses all sorts of instructions. When have you ever seen a creative tool subjected to such stringent content controls? It makes no sense at all."
A technical support page indicates that when end-users ask Fable about large-scale model construction or other topics, Anthropic sometimes calls upon older versions of the model to generate responses. Many users have complained on social media about the model frequently rejecting normal queries.
When Anthropic released Fable 5 in early June, it stated that the company was working to reduce the misflagging of legitimate requests. Just three days after the model's launch, to comply with U.S. government export control directives, Anthropic suspended access to Fable. On July 1, the company restored the model's service, explaining, "The new safety protection mechanisms flag a slightly larger portion of risk-free, normal requests compared to the old version."
Nadella's comments come as corporate executives increasingly favor more cost-effective AI models—those not necessarily from the most well-funded labs but capable of handling internal software development and various other business tasks.
Nadella's statement amounts to criticism of a key partner and customer.
Anthropic's Claude Code software development tool is gaining popularity among programmers and non-technical talent. Last November, Microsoft announced it would invest $50 billion in Anthropic, and the startup committed to spending $30 billion on Microsoft's Azure cloud services. This year, Microsoft released the enterprise workplace assistant Copilot Cowork, which integrates Anthropic's large model capabilities.
Investors have been concerned that various large models capable of quickly generating software code could impact Microsoft's business, while the company allocates tens of billions of dollars each quarter to expand its data centers. Year-to-date, Microsoft (NASDAQ: MSFT) stock has fallen 17%, while the Nasdaq Composite Index has risen 11% over the same period.
Recently, Nadella suggested that enterprises should be able to develop and customize their own large models cost-effectively, access internal data, and avoid sharing that data externally with other organizations, such as those specializing in large model development. In a blog post published on Sunday, he cited the views of Palantir CEO Alex Karp, who stated that technology companies "want to own their means of production."
Microsoft launched the Azure AI Foundry service, where developers can choose from over 11,000 large models, including several from Anthropic and OpenAI.
Nadella told the development engineers, "It's impossible for only two companies in the world to hold token computing resources, while all others must rent from them. This business model is completely illogical from an economic standpoint." Tokens are the unit for measuring the computational consumption of AI models.
Comments