Microsoft, a major tech company, has a close partnership with OpenAI, an organization that focuses on artificial intelligence. Recently, there have been some concerns that a company called DeepSeek might have taken some of OpenAI’s ideas or technology without permission, which could be a violation of their agreements. Despite these concerns, Microsoft still wants to use DeepSeek’s new AI models on its cloud services.
Today, Microsoft announced that it is now offering a model from DeepSeek called R1 on its Azure AI Foundry service. This service is designed to provide various AI tools for businesses all in one place. In a blog post, Microsoft explained that the version of R1 available on Azure has gone through extensive testing to ensure it is safe and behaves as expected. This testing included automated checks and thorough security reviews to address any potential risks.
Looking ahead, Microsoft mentioned that soon customers will also be able to use simpler versions of R1 on their Copilot+ PCs. These are specific Windows computers that are equipped to handle AI tasks.
Microsoft expressed enthusiasm about adding R1 to its Azure AI Foundry, stating that they are eager to see how developers and businesses will use it to solve real-world problems and create innovative experiences.
Read Also: Zuck Dismisses DeepSeek, Promises To Invest Billions In AI
However, the decision to include R1 in Microsoft’s cloud services is a bit puzzling. Reports suggest that Microsoft has been investigating DeepSeek for possibly misusing OpenAI’s services. Security researchers working for Microsoft believe that DeepSeek may have taken a lot of data using OpenAI’s API last fall. Since Microsoft is also the biggest investor in OpenAI, they alerted OpenAI about this suspicious behavior.
Despite these issues, R1 is generating a lot of interest, and it seems Microsoft may have decided to integrate it into their services while it is still popular.
What remains unclear is whether Microsoft made any changes to R1 to enhance its performance or reduce any biases in its responses. According to a test conducted by an organization that checks information reliability called NewsGuard, R1 gives incorrect answers or no answers at all 83% of the time when asked about news topics. Additionally, another test found that R1 avoids answering 85% of questions related to China, which might be due to the strict censorship imposed on AI models developed in that country.
