Microsoft Integrates GPT and Claude for Collaborative Work

Microsoft has introduced two innovative features for its Copilot Researcher tool that integrate the capabilities of OpenAI's GPT and Anthropic's Claude. Named Critique and Council, these features aim to optimize collaborative efforts in research tasks.

Critique employs a dual-model system designed to improve the accuracy of complex research. One AI model is tasked with generating the initial draft by planning and conducting information searches, while the second model reviews and refines this output, serving as an expert evaluator. Microsoft emphasized that this approach aims to minimize common issues found in current AI research tools, such as hallucinations and citation inaccuracies.

“One model leads the generation phase... The other is responsible for reviewing and refining, acting as an expert reviewer,” stated Microsoft.

The effectiveness of Critique was evidenced during the DRACO test, where it achieved a score of 57.4 on a series of complex research tasks, outperforming Claude Opus, which scored 42.7.

In contrast, the Council feature operates differently by running GPT and Claude in parallel. It presents their respective reports side by side, and a third model analyzes these responses to summarize their agreements and discrepancies.

While Critique encourages collaboration between the models, Council introduces a competitive element. Critique is set as the default mode in Researcher, while users can opt for Council from the menu. Both features are accessible to participants in the Microsoft Frontier program.

Earlier this year, Microsoft also launched the AI agent Cowork, designed to assist with multifaceted tasks across its Microsoft 365 applications, such as Outlook and Teams.