Search papers, labs, and topics across Lattice.
1
0
3
GPT-4o's hate speech detection is hamstrung by a "Unimodal Bottleneck," where context-blind safety filters preempt its advanced multimodal reasoning, leading to predictable false positives.
