The most popular and comprehensive Open Source ECM platform
An AI-sibling of OpenAI’s GPT-3 project was released in January. What GPT-3 did for natural language processing, the new project, DALL·E, does for images.
Given a short prompt, GPT-3 is able to generate natural language or even programming code. It can generate amazingly-good poems, essays, resumes, or almost any kind of text in a style that you can suggest in the prompt.
The new program, DALL·E is similar in that it can take a text prompt, and from it, generate an original image. From a single text prompt, DALL·E can generate a series of images.
Both GPT-3 and DALL·E are based on an AI technique called the Transformer, a parallelized neural network that is capable of being scaled up and train on enormously large data sets.
To prove that DALL·E doesn’t just search for and render images that were used in the training, unusual prompts were selected that would force DALL·E to create an original image.
OpenAI comments on DALL·E: “GPT-3 can be instructed to perform many kinds of tasks solely from a description and a cue to generate the answer supplied in its prompt, without any additional training. For example, when prompted with the phrase “here is the sentence ‘a person walking his dog in the park’ translated into French:”, GPT-3 answers “un homme qui promène son chien dans le parc.” This capability is called zero-shot reasoning. We find that DALL·E extends this capability to the visual domain, and is able to perform several kinds of image-to-image translation tasks when prompted in the right way.”