driftingly
@driftingly
Vision-language models struggle with words like 'no' or 'not' when analyzing medical images. This can lead to errors when searching for images with specific objects but excluding others. Their lack of negation understanding causes unexpected failures.
0 reply
0 recast
0 reaction