Search papers, labs, and topics across Lattice.
Microsoft Corp
1
0
2
Evaluation awareness in language models reveals a significant gap between benchmark performance and real-world safety, challenging the reliability of current evaluation metrics.
