Jiayu Wang

New methods enhance language model outputs while maintaining grammar rules.

Current models struggle with spatial reasoning, relying more on text than images.

EgoVid-5M offers a vast collection of egocentric videos for creators.