Microsoft Integrates GPT and Claude for Enhanced AI Research
- Microsoft has introduced Critique and Council modes in its Copilot Researcher tool, combining OpenAI’s GPT and Anthropic’s Claude for improved research quality.
- Critique mode involves collaboration between the models, with GPT generating a draft and Claude refining it, while Council mode runs them in parallel with a third model comparing results.
- On the DRACO benchmark, Microsoft’s combined system scored significantly higher than individual models, with Critique scoring a notable improvement of nearly 14% over the next best result.
- Both features are available to users enrolled in Microsoft’s Frontier program, requiring a Microsoft 365 Copilot license costing $30 per user per month.
Microsoft’s new multi-model approach aims to address common issues like hallucinations and weak citations found in single-model AI research tools by leveraging both collaboration and competition between models for optimal results.
The introduction of Critique and Council modes highlights Microsoft’s strategy to utilize multiple AI models for more accurate and comprehensive research outputs, as evidenced by their high performance on standardized benchmarks. Source
