Click here to .The Microsoft Research paper introduces CLIO (Cognitive Loop via In-Situ Optimization), an innovative approach designed to enhance large language models (LLMs) for scientific discovery.Unlike existing AI development paradigms that abstract reasoning control or rely on post-training, CLIO empowers scientists with deep and precise steerability over the AI's thought processes in real-time. By enabling LLMs to self-formulate problem-solving strategies, adapt behavior when uncertain, and provide transparent belief states through graph structures, CLIO significantly improves accuracy and explainability. This method was shown to surpass the performance of base models and other reasoning models like o3 on the Humanity's Last Exam (HLE) biology and medicine questions, demonstrating its effectiveness in fostering a more collaborative human-machine teaming for complex scientific challenges. The research also highlights how monitoring internal uncertainty oscillations within CLIO can serve as a critical signal for human intervention, ensuring more trustworthy and controllable AI applications in high-stakes domains.For the source article, click here.

Ver todos los episodios