Vision AI struggles to read its own instructions
AI vision models drop sharply in accuracy when questions appear as text inside images rather than typed separately.
AI vision models drop sharply in accuracy when questions appear as text inside images rather than typed separately.