Microsoft has introduced two innovative features, Critique and Council, enhancing the AI research quality by harnessing both OpenAI’s GPT and Anthropic’s Claude. In Critique, each model plays distinct roles: GPT leads the generation phase, drafting an initial report, while Claude serves as an expert reviewer, ensuring factual accuracy and citation quality before delivering the final output. This dual approach addresses common issues in mono-model AI, such as hallucinations and citation errors. Conversely, Council allows both models to work concurrently, generating side-by-side reports judged by a third model that highlights agreements and discrepancies between them. This dynamic collaboration and competition boosts analytical breadth and presentation quality significantly, as indicated by their performance on the DRACO benchmark—Critique scored 57.4, outperforming other systems. Available through Microsoft 365 Copilot’s Frontier program, these features position Microsoft at the forefront of the AI research race, showcasing the strategic value of multi-model orchestration.
Source link
