2021
OpenAI starts selling what it builds, and Anthropic is founded
This was the year OpenAI began pushing what it built out the door. DALL·E and CLIP — generating images and recognising them — came in January. Copilot, built with GitHub, entered technical preview in June, and in August OpenAI opened Codex, the model underneath it, through the API. In November Microsoft launched Azure OpenAI Service, letting enterprises use GPT-3 under cloud governance. In May, OpenAI also set up a $100 million Startup Fund.
In July it disbanded its robotics research team, choosing domains where data is plentiful, as with language, over the work of gathering it from the physical world.
On 28 May, Anthropic's founding was announced. Dario and Daniela Amodei, who had left OpenAI at the end of the previous year, raised $124 million. In December the company published its first major paper, setting out helpful, honest and harmless as its criteria.
Google announced LaMDA and MUM at I/O in May, showing fluent conversation without shipping it as a product.
In April, Microsoft acquired Nuance for $19.7 billion, reaching for speech recognition and a foothold in healthcare.
This block is written by the editors. It is kept separate from the sourced record below.
- January1DALL·E draws from text, shown but not handed over
- April1Microsoft buys Nuance, reaching for speech and healthcare
- May3Anthropic announces itself, Google shows LaMDA
- June1Copilot moves into the editor, and the training data argument starts
- July1OpenAI folds up robotics and moves resources to language
- August1Codex, the model under Copilot, opens as an API
- November1GPT-3 arrives on Azure and enters enterprise procurement
- December1Anthropic puts the criteria it aims at into a paper