Microsoft's Copilot tool is advancing its research functions with a fresh capability that merges capabilities from OpenAI's ChatGPT and Anthropic's Claude. The company revealed this in a blog entry about the rollout of Copilot Cowork, introducing the Critique option within Microsoft 365 Copilot's Researcher component. Distinct from the basic Copilot version, Researcher specializes in managing intricate, multi-stage assignments.
The new Critique enhancement improves Researcher's performance by starting with outputs from GPT and then polishing them using Claude. According to the blog, this setup establishes an effective iterative process that boosts outcomes in terms of truthfulness, depth of analysis, and clarity of delivery. The method mirrors practices common in scholarly and workplace investigations. Microsoft reports that this improvement outperforms the latest Perplexity Deep Research versions on the Deep Research Accuracy, Completeness, and Objectivity evaluation metric. Separately, Anthropic offers a Research tool that employs several Claude instances to generate thorough answers for demanding queries.
For users seeking greater independence in research activities, Microsoft has introduced the Model Council as another choice within Researcher. This provides parallel outputs from models by Anthropic and OpenAI, accompanied by an analysis highlighting areas of consensus and divergence.
These additions are presently accessible via the Frontier initiative in Microsoft 365 Copilot, serving as a testing ground for emerging AI developments from the firm.