1 paper
Bilal Jawaid, Hamza Hanif, Muhammad Maaz Uddin +8
Vision-language models (VLMs) can answer many geometric and physical reasoning questions correctly, yet still behave inconsistently in interaction, for example, abandoning a correc…