Microsoft has recently released a powerful feature called Critique for the research tool (Researcher) of Microsoft 365 Copilot, marking the first time GPT and Claude are collaborating on the same platform. This innovation breaks the limitations of a single model by enabling multi-model collaboration to complete complex academic research and data processing tasks.

In this workflow, GPT, with its strong text generation capabilities, is responsible for drafting the initial research paper. Subsequently, Claude takes over as the "strict reviewer," conducting in-depth verification of the content's accuracy and completeness according to professional academic standards.

image.png

Introducing the "Council" mechanism to eliminate AI hallucinations through multi-model collaboration

Aside from mutual review, Microsoft has also introduced an innovative "Council" mechanism. This mechanism allows multiple models to conduct research independently in physical isolation, and finally, a dedicated "judging model" compares and evaluates the results from each side.

Test data from DRACO shows that the effect of this multi-agent collaboration is significantly better than that of any single model. By leveraging the strengths of different algorithms, the system can effectively filter out incorrect information, greatly reducing the long-standing issue of "AI hallucinations" in the industry.

Shifting from general tools to specialized agents, building a new AI industrial ecosystem

Industry analysts believe that Microsoft's move marks the evolution of AI assistants from general-purpose tools to specialized, industry-oriented "digital employees." GPT excels in creativity and generation, while Claude focuses on security and rigor, and their complementarity sets a new benchmark for enterprise-level high-reliability applications.

Through deep strategic cooperation with NVIDIA and Anthropic, Microsoft is building a vast AI ecosystem. Future industry competition will no longer be about individual model parameters, but rather about who can build a more efficient and stable multi-agent collaboration ecosystem.