Double artificial intelligence support for Microsoft Copilot: Research results are changing

Microsoft announced new features to improve the research capabilities of its productive artificial intelligence-based assistant Copilot. According to the company’s statement, the Researcher tool within Microsoft 365 Copilot can now produce more comprehensive and consistent results by using multiple artificial intelligence models simultaneously. This new approach aims to deliver higher accuracy and coverage, especially in tasks that require multi-step and deep analysis. In addition, unlike the standard version of Copilot, Researcher offers a structure that can work with complex data sets and prepare detailed reports. Despite this, how the system will perform will be more clearly understood in real usage scenarios.

The new feature called “Critique” introduced by Microsoft establishes a collaboration between GPT models developed by OpenAI and Anthropic’s Claude model. In this system, the answers generated by GPT in the first stage are then reviewed and improved by Claude. In addition, the feedback loop created in this process ensures that the responses become stronger in terms of accuracy, comprehensiveness and quality of presentation. According to the company’s statements, this approach is similar to multi-layered evaluation methods used in academic and professional research processes. Although this system aims to produce higher quality results in theory, differences in interpretation of different models may require additional evaluation in some cases.

Next generation multi-model approach for Microsoft Copilot Researcher

Microsoft also shared some comparisons regarding the performance of the new system. Accordingly, it is stated that the updated Researcher tool scores higher than competing solutions in certain tests in criteria such as Deep Research Accuracy, Completeness and Objectivity. However, it is also known that such comparisons are usually made on specific scenarios and may not cover all use cases. On the other hand, it is noteworthy that research solutions that use more than one Claude agent have been developed in a similar manner on the Anthropic side. This reveals that the use of multiple artificial intelligence models is becoming increasingly common in the industry.

In addition, Microsoft has also launched an alternative feature called “Model Council” for users who want more control. Thanks to this feature, users can see the responses from OpenAI and Anthropic models side by side. In addition, the system also provides a report summarizing the differences and common points between the two models. Thus, users can make more informed decisions by evaluating different perspectives. However, this approach may require additional interpretation skills from the user, especially in matters requiring technical knowledge.

In addition to all these, these innovations are offered within the scope of Microsoft 365 Copilot’s “Frontier” program. This program is positioned as an early access area where the company tests experimental artificial intelligence features that are not yet widely available.