Search papers, labs, and topics across Lattice.
3
0
8
LLMs can reason more effectively by directly tracking their own belief in the correct answer throughout the reasoning process, enabling more targeted policy updates.
Forget retraining: SHARPEN surgically repairs defects in neural networks by pinpointing faulty neurons with explainable AI and then fixing them with gradient-free optimization.
Even the most advanced MLLMs like GPT-5 and Gemini struggle to spot the "odd one out" in simple visual grids, revealing a surprising weakness in fine-grained visual perception.
