AI academic research

This paper systematically evaluates the ability of large language models (LLMs) to generate and recognize increasingly complex visual concepts, and shows how to use text models to train preliminary visual representation learning systems. Although language models cannot directly handle pixel-level visual information, images are studied using codes to represent them. Although the images generated by LLM do not look like natural images, the results in image generation and correction show that accurately modeling strings can teach language models many aspects of the visual world. Furthermore, experiments on self-supervised visual representation learning using images generated by text models highlight the potential of training visual models capable of semantic evaluation of natural images using only LLMs.