Microsoft was completely a slave to OpenAI in terms of AI capabilities. That’s just starting to change

Yesterday, Microsoft unveiled an avalanche of new features that reinforce its commitment to end-user AI like never before. Among them there are several, yes, that do not depend on this, and this is significant.

Recall from Microsoft. As Microsoft’s documentation explains, Recall uses the Windows Copilot Runtime, a component that in turn uses the Phi model developed by Microsoft itself. Thus, we are faced with an interesting feature that depends on technology developed by the company itself.

Copilot Vision is also similar. Another of the most interesting features introduced yesterday was Copilot Vision, which allows you to “see” what you see in the browser. Currently only in Edge and only on certain websites, but this allows you to request this content in the browser. This feature may be based (Microsoft does not specify) on Phi-3.5-Vision-Instruct, a lightweight model specifically targeted at this type of feature.

Think deeper – o1. This Copilot Labs option allows you to access a version of Copilot that “reasons” and takes its time to try to make fewer mistakes in its answers. Mustafa Suleiman, head of Microsoft’s artificial intelligence division, explained in Wired that this option is based in o1 (Strawberry), although it emphasizes that they have tweaked this model to adapt it to the capabilities offered by Copilot.

Co-pilot’s voice (probably) GPT-4o.. A few months ago, OpenAI began demonstrating how GPT-4o is capable of speaking to us in synthetic voices that sound almost human. Now this feature has appeared in Copilot Voice, which is essentially the same thing but with a different name. Just yesterday, OpenAI introduced its Realtime API at its developer conference, which allows developers to create fast voice-to-speech processes for their applications. Copilot Voice is exactly that, and while Microsoft hasn’t confirmed it – it’s starting to use it in Azure – it appears to be based on that API.

Microsoft moves the token. All of this goes to show that while Microsoft is a great ally and investor in OpenAI, it is trying not to depend (as much) on it. It’s a predictable and smart strategy: It takes advantage of its current alliance to integrate the most advanced AI models into its product catalog, but in parallel develops its own models and begins to integrate them little by little.

Apple does something similar. Apple’s strategy is somewhat similar: in its case, limited Apple Intelligence features are based on its own technology, but the user can access ChatGPT, OpenAI’s chatbot, if desired. It’s understandable that Apple, whose obsession with absolute control is well known, doesn’t want to be dependent on anyone, although for now this limits what its AI platform can offer.

OpenAI must become indispensable. On the other hand, we have OpenAI, which is at a disadvantage: it does not have its own operating system for PCs or mobile phones, like Microsoft, Google or Apple. This forces the company to continue trying to be the benchmark in artificial intelligence so that these companies continue to need it, at least in certain areas. There is a problem, but it is clear that these three companies, as well as others, are working on their own artificial intelligence models, and this will make the task increasingly difficult for Sam Altman’s team.

In Hatak | So far, Microsoft and OpenAI have been great allies. Until now