The language model, Chinchilla, leverages the language knowledge acquired to provide the few-shot learning ability to OpenAI's image generating application DALL-E. The model can produce images from textual prompts and uses frozen data sets to connect the two modalities, visual and language.