Zhejun Zhao

Achieve state-of-the-art multimodal stance detection by having multiple AI agents debate each other, complete with retrieval-augmented context and self-reflection.

Models can learn to self-differentiate between tasks requiring rigorous planning versus direct generation in creative writing, unlocking a new level of meta-cognitive ability.