AI can generate answers. But can it check its own work? That is the problem Microsoft is trying to solve with Critique, a new multi-model research system introduced inside its Copilot Researcher tool.