Search papers, labs, and topics across Lattice.
Westlake University, Zhipu AI
1
0
2
16
Current judge models for instruction-following are surprisingly unreliable, but a new benchmark exposes their flaws and offers a path to better alignment.
