ChatGPT arrives, and anyone can now talk to artificial intelligence

Until that day, artificial intelligence was, for most people, either science fiction or an invisible part of the services they used: it picked their social media feeds and translated text, but you could not talk to it.

On 30 November 2022, the US company OpenAI opened ChatGPT, a website where you could simply type a question, as in a messaging app. You could ask it to explain a hard topic, draft a letter, write a poem or find a bug in a program. The bot answered coherently, remembered what the conversation was about, and could admit it had made a mistake.

OpenAI itself called the launch an experiment and warned openly that ChatGPT sometimes states falsehoods with confidence, can be wordy, and may answer differently if a question is phrased slightly differently.

That did not stop people. According to an estimate by the bank UBS, ChatGPT had around 100 million monthly users by January 2023. For comparison, by the same estimate TikTok took nine months to reach that audience, and Instagram two and a half years.

Why it matters

This is the moment AI left the labs and reached ordinary people. They began using it to write letters and do homework, to find bugs in code and come up with gift ideas. Teachers, employers and lawmakers suddenly had to decide what to make of it.

For technology companies, the launch was the starting gun: the biggest of them rushed to prepare answers of their own. This book counts its days from this one.

For the curious

ChatGPT is built on a model from the GPT-3.5 series (GPT stands for Generative Pre-trained Transformer). It was fine-tuned with human help: people rated its answers, and the model learned to respond the way people preferred. The method is called RLHF (reinforcement learning from human feedback); OpenAI had used it earlier for its InstructGPT model.