Announcement
NormViz benchmark measures AI understanding of visual norms across 16 countries
A paper author says Gemini 3 Flash was the best-performing model, scoring 25.3%.
TLDR
An author of a COLM 2026 paper introduces NormViz as a benchmark and framework for measuring AI understanding of visual norms across 16 countries. They say Gemini 3 Flash performed best at 25.3%, and cite gifting a plate with four koi in Japan and placing chopsticks upright in rice in China as examples of cultural norms AI models miss.
Combined views
1.4K
1 Source, first seen 4h ago
