2023

The excitement around generative AI that began with ChatGPT in November 2022 spread to other industries over the year, and drew in national governments.

In January, Microsoft committed another outsized investment to OpenAI, the company behind ChatGPT. Google shipped its own AI twice that year: Bard in February, Gemini in December. NVIDIA, which holds the GPUs and the CUDA software the field runs on, saw its stock surge.

Releasing model weights openly became more common. Newer names such as Anthropic and Mistral AI began drawing attention.

Among users, the craft of writing prompts spread. Beyond appending "let's think step by step" — the subject of a paper the year before — there was assigning the model a role, specifying the output format up front, showing it what a bad answer looked like. Fragments like these circulated in bulk through social media and paid courses, and "prompt engineer" turned up as a job title. For anyone with no access to the model itself it was one of the few levers available, and much of what got passed around was folklore that did not reliably reproduce. From that spring, Japanese social media also settled on a word — "odorokiya", roughly "astonishment-monger" — for accounts hyping each new model or tool as mind-blowing, and the label stuck.

Warnings about the dangers of AI grew louder. The major AI companies responded with a joint set of voluntary commitments, and agreement at the political level advanced too. The camps got names of their own — "doomers" for those warning of danger, "e/acc," effective accelerationism, for those urging speed — labels that had settled into English by the time of November's upheaval at OpenAI.