Openai, a close partner and collaborator of Microsoft, will suggest that DeepSeek stole IP and violated the service terms. But Microsoft still wants a new model of DeepSeek in the cloud platform.
Microsoft announced today that R1, the so -called reasoning model of DEEPSEEK, can be used in Azure AI Foundry Service, Microsoft’s platform that provides multiple AI services for companies under a single banner. In the blog post, Microsoft said that the R1 version of Azure AI Foundry said, “We went through a strict red team and safety evaluation to alleviate potential risks, including automatic evaluation of model behavior and extensive security review.”
In the near future, the customer said that it can run local on Copilot+ PC, a Windows hardware brand of Microsoft, which uses R1’s “distilled” flavor to meet specific AI preparation requirements.
Microsoft said, “We are pleased to know how developers and companies use to solve the actual challenges and provide innovative experiences as we continue to expand the model catalog in Azure AI Foundry.
It is curious to add R1 to Microsoft’s cloud service, thinking that Microsoft has started probe for the potential abuse and Openai service of Deepseek. According to security researchers working at Microsoft, Deepseek may have deported a large amount of data in the fall of 2024 using Openai’s API. Microsoft informed Openai about suspicious activities of Bloomberg, the largest shareholder of Openai.
But R1 is a story about the city, and Microsoft may have been persuaded to bring cloud folding while still attracted.
It is unclear whether Microsoft is modifying the model and fighting censorship to improve accuracy. According to the test of information reliability organizations, R1 provides 83%of inaccurate or non -response time when you ask about the news topic. According to a separate test, the R1 rejects 85%of the prompt -related prompt, and is probably the result of the government censorship that has been targeted by the AI model developed in the country.
