Search papers, labs, and topics across Lattice.
1
0
3
7
High initial confidence in LLMs can lead to catastrophic failures in complex reasoning tasks, but a new framework shows how to harness confidence trajectories for better outcomes.
