Skip to content
Stela
ENFR

Halls

We built IWe discovered IIWe imagined IIIWe changed the world IVWhat we lost VWhat we leave behind VI

Explore

TimelineMapJourneysSearch the archive

Language

Français

Settings

RH-0093Artificial intelligenceLanguagesAmericasTurning point

ChatGPT, the machine that answered in words

Before 30 November 2022 holding a long conversation in ordinary language with a machine was largely the business of researchers and specialists; after it, a free public service put it within reach of anyone online, and within about two months it had, by one analyst estimate, 100 million monthly users.

Dated
2022
Place
Mission District, San Francisco, California, United States · OpenAI, an American artificial intelligence company based in San Francisco, and the human trainers and data workers who shaped its models
Coordinates
37° 46′ N · 122° 25′ W (approximate)
Remains
Digital

Transcription · Post on Twitter (now X)

today we launched ChatGPT. try talking with it here:

chat.openai.com

Sam Altman @sama11:38 AM Pacific Standard Time (local time)See the original Transcribed by the archive from the original. The profile picture, logos and interface are not reproduced.

From the archivist

There is nothing here to hold. The fragment is a box on a screen, a blinking line, and a machine that answers in whole sentences.

On the last day of November 2022, a company in San Francisco put it online as a test and let anyone use it for free. Its chief executive announced it in a single line, kept here exactly as it was written. People typed questions, requests, jokes, homework. The machine answered at once, politely, in the same language, and often it was right. Sometimes it was wrong with the same calm.

Within months, teachers were changing how they set essays, programmers were pasting its code into their work, and lawyers and writers were asking whose words it had learned from. Somewhere in Nairobi, people had been paid little to read the worst of what this civilization wrote so that the machine would learn not to repeat it.

An earlier fragment in this archive recalls a question asked by Alan Turing in 1950: can machines think? Here, millions of people put it to the test themselves, each at their own table, and did not agree on the answer.

Established

  • On 30 November 2022 OpenAI announced ChatGPT, a model that "interacts in a conversational way", and opened it to the public with the words "During the research preview, usage of ChatGPT is free" (OpenAI).
  • The same day, at 19:38 UTC (11:38 in San Francisco), OpenAI's chief executive Sam Altman announced it in a public post on Twitter: "today we launched ChatGPT. try talking with it here: chat.openai.com" (Altman, post of 30 November 2022).
  • ChatGPT was fine-tuned from a model in OpenAI's GPT-3.5 series by reinforcement learning from human feedback: human trainers wrote example conversations and ranked the model's answers (OpenAI).
  • The announcement itself warned that ChatGPT "sometimes writes plausible-sounding but incorrect or nonsensical answers", and that it would sometimes respond to harmful instructions or show biased behaviour (OpenAI).
  • According to a UBS analyst note based on Similarweb data and reported by Reuters, ChatGPT reached an estimated 100 million monthly active users in January 2023, about two months after its launch; this was an outside estimate, not a figure published by OpenAI (Reuters, February 2023).
  • A study by economists at OpenAI, Harvard and Duke, drawing on OpenAI's own data, reports that by July 2025 about 700 million users were sending 18 billion messages a week, around 10% of the world's adult population (Chatterji et al. 2025).

Interpretation

What changed is not in doubt: conversation with a machine in ordinary language, once a laboratory matter, became part of daily life for hundreds of millions of people, for writing, advice and looking things up. Almost everything else is argued. The answers can be fluent and wrong, the errors that came to be called hallucinations. Authors and publishers, among them The New York Times in a lawsuit filed in December 2023, contest the use of their work as training data. An investigation by TIME in January 2023 reported that workers in Kenya, employed through an outsourcing firm, took home between about 1.32 and 2 dollars an hour to label violent and abusive texts so that the system could learn to avoid them. The International Energy Agency estimated that data centres of all kinds used about 1.5% of the world's electricity in 2024 and expected their demand to more than double by 2030, with artificial intelligence as a major driver. Some researchers fear risks from far more capable systems; others think those fears distract from present harms. The archive is written too soon after the event to know its full weight.

In 10,000 years

In 10,000 years, will this be remembered as the day machines began to speak with everyone, or as one early version among many that nobody could tell apart?

Sources (8)
  1. Altman, S. (@sama). today we launched ChatGPT. try talking with it here: chat.openai.com. Post on Twitter (now X), 30 November 2022, 19:38 UTC. · x.com
  2. OpenAI. ChatGPT: Optimizing Language Models for Dialogue, 30 November 2022. · openai.com
  3. Chatterji, A., Cunningham, T., Deming, D. J., Hitzig, Z., Ong, C., Shan, C. Y., Wadman, K. How People Use ChatGPT. National Bureau of Economic Research, Working Paper 34255, September 2025. · www.nber.org
  4. Hu, K. ChatGPT sets record for fastest-growing user base, analyst note. Reuters, February 2023. · www.reuters.com
  5. Perrigo, B. OpenAI used Kenyan workers on less than $2 per hour to make ChatGPT less toxic. TIME, 18 January 2023. · time.com
  6. NPR. Report on The New York Times's copyright lawsuit against OpenAI and Microsoft, 27 December 2023. · www.npr.org
  7. International Energy Agency. Energy and AI, April 2025. · www.iea.org
  8. Zeff, M. Elon Musk's xAI moves into OpenAI's old HQ. TechCrunch, 2 October 2024. · techcrunch.com