What is DALL-E?
This week on the Fireside AI podcast, our guest was Ming Cheuk, CTO of ElementX. Together with Daniel and Morgan, he discussed the DALL-E image generation model, which is a new AI system that can create realistic images and art from a written description. It uses a 12-billion parameter version of GPT-3 (a natural language model) to generate images from prompts like the ones our hosts explored in the episode.
You can see the results of those prompts below.
Ming: We've seen AI creating images for a while now, and in the early days the images that these type of models would create would be very abstract. It doesn't reflect any realistic photo or, you know, image that you'd see in the world. But over the years, especially the last one year or so, they've really made a big breakthrough in creating images that look very realistic, whether it's photorealistic or just realistic in general, even if it's of a cartoon character.
Listen to Fireside AI Episode 3: DALL-E for more on the following:
If you’ve made it this far, you’re most likely here for the stunning works of art they generated during the podcast. Be warned: once viewed, you’ll never unsee them.
Listen to the full episode here.
Play with DALL-E yourself here.