Microsoft 365 Copilot Enhancements

Microsoft 365 Copilot has recently received multiple significant upgrades as of March 30, including a feature designed for research, the rollout of Copilot Cowork, and a tool for comparing various AI models simultaneously.

In a blog update, Microsoft stated that all these improvements form part of Wave 3 of Microsoft 365 Copilot, which focuses on enhancing its artificial intelligence capabilities. Below are the essential details regarding the latest features added to Microsoft 365 Copilot.

Copilot Cowork

On March 9, Microsoft revealed Copilot Cowork, which has now begun to be introduced to early-access users through its Frontier programme. This AI-driven feature for Microsoft 365 Copilot is underpinned by technology similar to Anthropic’s Claude Cowork.

Copilot Cowork can autonomously organise and carry out tasks across applications like Excel, Outlook, Teams, and PowerPoint. Users simply need to outline their task, and the tool will proceed with the work while displaying the progress made.

New Features in Researcher

Microsoft has also announced enhancements for the Researchers feature, which are based on multi-model intelligence. This tool can already tackle complex queries, but the new Critique functionality promises to elevate its capabilities further.

The Critique feature utilises multiple AI models from firms like Anthropic and OpenAI, dividing tasks into two distinct phases. It separates the generation of content from the evaluation phase, employing a combination of models from Frontier labs, including both Anthropic and OpenAI.

One model is tasked with generating the initial draft, while the second focuses on reviewing and refining it. This approach enhances the process of writing and reviewing for better outcomes. In the DRACO benchmark, which evaluates accuracy, completeness, and objectivity in research, the Researcher shows a 13.8% improvement.

Critique is being rolled out as a standard feature in Researcher and can be accessed by selecting Auto in the model picker.

Model Council in Research

Microsoft has introduced the Model Council feature in Research, allowing a side-by-side comparison of different AI models. This functionality can run models from both Anthropic and OpenAI concurrently, with each creating a comprehensive report on the same subject.

After the reports are generated, a third “judge” model steps in to deliver a detailed summary of key findings, highlighting areas of agreement and disagreement among the models, as well as unique insights they may provide.