Junyang from Alibaba discusses the upgraded Quenvl multimodal model with enhanced OCR capabilities and emergent properties. The larger the language model, the better it performs on visual tasks. The model can even solve math problems from screenshots.
