A text box

For two years, all of this stayed among people who could write code.

GPT-3 was reachable from 2020, but only through the API, which meant you had to be a developer with an approved account to send it anything. The argument about what these systems could do was being had by researchers, in papers, about benchmarks.

Then in November 2022, OpenAI put a post-trained model behind a box you could type in, called it ChatGPT, and let anyone use it.

Nothing in that sentence is a technical breakthrough. The model was a tuned version of one that already existed. What changed was that the question stopped being theoretical. You no longer had to read about whether a machine could write a decent paragraph. You could ask it to, and see.

By one widely cited estimate it was being used by a hundred million people within two months, quicker than any consumer application before it. It did that by removing two barriers that had nothing to do with the model. You did not need permission, and you did not need to know how to program.

The machinery underneath had been built over decades. The public arrival took a text box.