T1#consumer
ChatGPT Launched
Metadata
- Date
- Decade
- 2020s
- Tier
- T1
- Sources
- 10
- Connections
- 04
- Tags
- #consumer
On 30 November 2022, OpenAI published a blog post titled "ChatGPT: Optimizing Language Models for Dialogue" (since retitled "Introducing ChatGPT"). The post called itself a "research preview" and noted that "during the research preview, usage of ChatGPT is free." A low-key release.
Within days it had become the fastest-ramping consumer application anyone could then remember.
Speed
- One million users in five days (stated by Sam Altman on 5 December 2022).
- An estimated one hundred million monthly active users within two months.
The second figure is not OpenAI's. It comes from a UBS study built on Similar Web traffic data, reported by Reuters on 1 February 2023. That report put ChatGPT's January traffic at about 13 million unique visitors a day, more than double December's. "In 20 years following the internet space, we cannot recall a faster ramp in a consumer internet app," the UBS analysts wrote.
For comparison, the same report cited Sensor Tower figures: TikTok took about nine months after its global launch to add 100 million users, Instagram two and a half years. ChatGPT took two months.
| about 2.5 years | ||
| TikTok, from global launch | about 9 months | |
| ChatGPT | about 2 months* |
The record did not hold. In July 2023 Meta's Threads passed 100 million sign-ups in five days. ChatGPT's claim is properly read as the fastest ramp as of early 2023, not as a standing record.
The Interface, Not the Technique
Technically, ChatGPT was an existing language model fine-tuned for dialogue. The launch post is explicit: "ChatGPT is fine-tuned from a model in the GPT-3.5 series, which finished training in early 2022." The method was Reinforcement Learning from Human Feedback (RLHF)—human trainers rank several candidate replies, a reward model is fitted to those rankings, and the policy is tuned with Proximal Policy Optimization—the same recipe as InstructGPT, with differences only in how the data was collected.
But the decisive factor was not the technique. The decisive factor was the interface.
GPT-3 (2020) had been distributed as an API. To use it, a developer had to join a waitlist, write code, tune parameters—a staircase. ChatGPT removed the staircase.
- Open a URL.
- Type a question.
- Read the answer.
Not quite unconditionally: at launch an OpenAI account was required, with email and phone verification. Using ChatGPT without signing up did not arrive until April 2024, nearly a year and a half later. Even so, against the cost of calling an API, the step was barely a step.
Strategic Pivots
Within a month, the New York Times reported (21 December 2022) that ChatGPT's release had led Google's management to declare a "code red." According to a memo and audio recording the paper obtained, Sundar Pichai joined a series of meetings to define Google's AI strategy and "upended the work of numerous groups inside the company" in response to the threat.
Microsoft had already invested on the order of a billion dollars in OpenAI (2019) and held an exclusive cloud arrangement. On 23 January 2023 the two announced the third phase of the partnership: a "multiyear, multibillion dollar investment." No figure was disclosed. The widely quoted ten billion comes from Semafor's reporting; neither company confirmed it.
On 7 February Microsoft launched the new Bing with chat built in, describing it as running on a next-generation OpenAI large language model "more powerful than ChatGPT and customized specifically for search." Microsoft confirmed that this had been an early version of GPT-4 only on 14 March 2023, the day GPT-4 was announced. Google announced Bard on 6 February, but the public could not use it until 21 March, in a staged rollout limited to the US and UK.
Anthropic—founded in 2021 by Dario and Daniela Amodei and colleagues who left OpenAI at the end of 2020—opened early API access to Claude and Claude Instant on 14 March 2023, the same day GPT-4 was announced. Meta announced LLaMA on 24 February, but distributed the weights through an application process, granted case by case to academic researchers.
In the six months from December 2022 to June 2023, the industry switched, completely, from a 'research then gradually productise' cadence to a continuous 'ship-improve-ship-improve' cadence.
Into Ordinary Life
In schools, debate over whether submitted essays were ChatGPT-generated began immediately. In January 2023 New York City Public Schools restricted ChatGPT on school-issued devices and networks; schools chancellor David Banks reversed the decision in May of the same year. Coding interviews, legal briefs, medical notes, copywriting—every form of "produce text" work felt the impact within weeks.
Perhaps this was always coming. But the fact that it began on 30 November 2022 will be recorded as a particular date.
Questions this page answers
- When was ChatGPT launched?
- On 30 November 2022, as a free research preview.
- Did ChatGPT really reach one million users in five days?
- That is Sam Altman's own figure, given five days after launch. It was the fastest ramp any consumer application had then managed; Meta's Threads passed 100 million sign-ups in five days in July 2023.
- Is the 100 million users in two months figure from OpenAI?
- No. It is a UBS estimate of monthly active users, built on Similar Web data and reported by Reuters on 1 February 2023. OpenAI never published it.
Sources
PrimaryChatGPT: Optimizing Language Models for Dialogue — OpenAI, Nov 30, 2022
Now titled "Introducing ChatGPT"
PrimaryAn important next step on our AI journey — Sundar Pichai, The Keyword (Google), Feb 6, 2023
The announcement of Bard
TertiaryChatGPT — Wikipedia
Last updated: