Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
2
0
5
Subjective biases in text can significantly distort truth, with this study revealing how large language models can effectively identify and classify these biases across multiple dimensions.
HarmVideoBench reveals that existing benchmarks miss critical layers of harmful video understanding, while a new method boosts model accuracy by over 20%.