Microsoft's Research Assistant Now Pits AI Models Against Each Other

Microsoft is testing a new method for its Copilot research tool: making competing AI models work together. A feature called Critique, now in preview, chains responses from OpenAI's GPT with refinements from Anthropic's Claude to handle multi-step research questions. Microsoft describes the process as a feedback loop designed to improve factual accuracy and analytical depth, mimicking a structured review.

In internal benchmarks, Microsoft says this combined approach outperformed recent models from competitor Perplexity on measures of accuracy and objectivity. For users who want more direct comparison, a separate Model Council feature provides side-by-side responses from GPT and Claude, accompanied by a report highlighting where the models' answers align and diverge.

These features are currently available within the Microsoft 365 Copilot Frontier program, an early-access environment. The move signals a shift from relying on a single model for complex tasks toward a more orchestrated, multi-model strategy. It also reflects a broader industry trend where the choice isn't necessarily about picking one 'best' model, but about strategically deploying different models for their respective strengths.

Source: Engadget