Search papers, labs, and topics across Lattice.
1
0
2
3
LLM-as-a-Judge accuracy hinges on temperature settings, revealing a task-dependent sweet spot that defies the common practice of fixed values like 0.1 or 1.0.
