Search papers, labs, and topics across Lattice.
Rochester Institute of Technology
2
0
4
5
VLMs are prone to critical failures that vary significantly across cultures, exposing the inadequacy of Western-centric safety benchmarks.
LLMs can't capture the nuances of human disagreement, but DiADEM, a new architecture that explicitly models annotator demographics, achieves strong disagreement tracking by learning which demographic axes matter most.
