September

2 events

Editors' summary

OpenAI gives away Whisper and drops the DALL·E 2 waitlist

On 21 September, OpenAI open-sourced Whisper, a speech recognition model trained on 680,000 hours of audio collected from the web, handling transcription in many languages and translation into English. A company built around a commercial API handed out the model and the inference code.

On the 28th it removed the DALL·E 2 waitlist, opening it to anyone who signed up.

Stable Diffusion, released in August, was spreading, and image generation was becoming something you ran on your own machine.

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

Record2 events

21
ModelOpenAI
Whisper speech recognition model open-sourced
OpenAI open-sourced Whisper, a speech recognition model. An encoder-decoder Transformer trained on 680,000 hours of multilingual and multitask supervised data collected from the web, it approaches human-level robustness and accuracy on English speech recognition, and supports transcription in multiple languages as well as translation into English. The release drew attention also because the company, then focused on commercial APIs, published the models and inference code.
28
ProductOpenAI
DALL·E 2 opens to everyone as waitlist is removed
OpenAI removed the waitlist for DALL·E 2, letting new users sign up and start creating immediately. By then, more than 1.5 million active users were generating over 2 million images a day, and over 100,000 users were sharing creations and feedback in its Discord community. The company's iterative deployment approach, expanding access while validating safety measures, took the system from a limited preview in April to full availability in about half a year.

↑ 2022

Whisper speech recognition model open-sourced

DALL·E 2 opens to everyone as waitlist is removed