Search papers, labs, and topics across Lattice.
1
0
2
5
Transcript-based shortcuts in dialogue models lead to a staggering drop in accuracy, revealing a critical flaw in current evaluation methods.
