On November 30, 2022, OpenAI released ChatGPT. The product was an interface that wrapped a fine-tuned version of GPT-3.5 in a conversational form, with a clean web interface and a free tier that anyone could use without a waitlist. Within five days, the service had a million users. Within two months, it had reached a hundred million monthly active users, the fastest consumer product growth in the history of the internet to that point.
What ChatGPT did was take a capability that had been available to developers via the OpenAI API for two years and put it into a form where ordinary people could experience it directly. The conversational interface eliminated the friction of designing prompts. The free access eliminated the financial commitment. The browser-based delivery eliminated installation and account complexity beyond a basic sign-up.
The capability under the surface was not new in any radical sense. The GPT-3 base model had been released in 2020. The fine-tuning techniques that produced the conversational behaviour were refinements of methods that had been published in academic literature. What was new was the packaging, the UI, the moderation layer that made the service usable for a broad audience, and the timing.
The reaction across various professional categories was significant within the first few weeks. Educators discovered that students could produce passable essays in minutes. Developers found that ChatGPT could answer programming questions and produce small functional code snippets at a level that was hard to ignore. Marketers began using it to draft emails, social media posts, and other short-form content. Researchers found it useful for brainstorming and rough drafts.
What was harder to assess in those first few weeks was how reliable any of this was. ChatGPT confidently produced text that contained errors. It made up references that did not exist. It described historical events that had not happened. The output was fluent. The relationship between fluency and accuracy was different from anything users had previously interacted with.
The launch reset expectations across the technology industry within months. Microsoft, which had invested heavily in OpenAI, accelerated integration plans. Google declared a code red and reorganised significant resources around language model competition. Major companies that had paid little public attention to AI started announcing strategies. The infrastructure of how technology companies thought about AI had been transformed by a single product launch in a way nobody, including OpenAI itself, had fully anticipated.