Search papers, labs, and topics across Lattice.
Affiliation:
3
0
5
Role-linked visual biases in text-to-image models persist and even intensify when placed in unrelated contexts, challenging assumptions about context-free evaluations.
A mere 15 visual attributes drive 80% of social bias in MLLMs, highlighting the concentrated nature of bias in visual cues.
LLMs can flip their answers up to 97.3% of the time when faced with counterarguments, revealing a critical instability that accuracy metrics overlook.
