January

1 events

Editors' summary

DALL·E draws from text, shown but not handed over

On 5 January, OpenAI announced DALL·E and CLIP together. DALL·E generates images from text; CLIP puts images and text in a shared space so they can be matched. The "armchair in the shape of an avocado" was quoted everywhere.

DALL·E itself was not released — only selected examples were shown. As with GPT-3, OpenAI demonstrated what it could build without handing it over.

This block is written by the editors. It is kept separate from the sourced record below.

Record1 events

5
ResearchOpenAI / Aditya Ramesh / Alec Radford / Ilya Sutskever
OpenAI unveils DALL·E and CLIP
OpenAI announced two models on the same day: DALL·E, which generates images from text, and CLIP, which learns visual concepts from natural language supervision. DALL·E, a 12-billion-parameter version of GPT-3 trained on text-image pairs, could render prompts like "an armchair in the shape of an avocado" as images. CLIP performs zero-shot image classification given only the names of the target categories. Both works carried language-model techniques into vision, showing that text and images could be handled within a single framework.

↑ 2021

OpenAI unveils DALL·E and CLIP