On November 30, 2022, OpenAI released ChatGPT as a public demo. Within 5 days it had 1 million users. Within 2 months it had 100 million — the fastest consumer product in history to reach that milestone.
But ChatGPT wasn't built in 2022. The underlying technology — the GPT large language model — had been in development since 2018. What OpenAI released was a chat interface on top of an already-powerful model, plus a technique called RLHF (Reinforcement Learning from Human Feedback) that made the model far more useful and controllable.
The lesson: the technology had been maturing for years. A good product interface, at the right quality threshold, triggered mass adoption overnight.
Previous AI (voice assistants, recommendation engines) felt like a feature baked into something else. ChatGPT was the first AI that felt like talking to something. Users could: - Ask any question in plain English - Get a thoughtful, essay-quality response - Refine the response through follow-up conversation - Ask it to write code, poems, legal summaries, marketing copy — anything
This generality — one system doing many things previously requiring separate specialized tools — was the disruption.
AI has had multiple cycles of excitement and disappointment:
| Era | What happened |
|---|---|
| 1956 | Term 'Artificial Intelligence' coined; early optimism |
| 1970s–80s | 'AI Winter' — progress stalled; funding dried up |
| 1997 | IBM Deep Blue beats chess world champion |
| 2012 | Deep learning breakthrough — AlexNet wins image recognition contest |
| 2017 | Transformer architecture introduced (the foundation of all modern LLMs) |
| 2020 | GPT-3 released — first model impressive enough for practical use |
| 2022 | ChatGPT — mass consumer adoption begins |
| 2023–26 | Rapid model improvement; AI agents; multimodal models; vibe coding |
In 2017, Google researchers published a paper titled Attention Is All You Need, introducing the transformer architecture. This is the 'T' in GPT (Generative Pre-trained Transformer).
Without going into math: the transformer solved the problem of handling long, complex text — keeping track of how words relate to each other across many sentences. This enabled models to be trained at unprecedented scale on vast amounts of text, producing systems that could write, reason, and generate in ways no previous architecture could.
On November 30, 2022, OpenAI released ChatGPT as a public demo. Within 5 days it had 1 million users. Within 2 months it had 100 million — the fastest consumer product in history to reach that milestone.
But ChatGPT wasn't built in 2022. The underlying technology — the GPT large language model — had been in development since 2018. What OpenAI released was a chat interface on top of an already-powerful model, plus a technique called RLHF (Reinforcement Learning from Human Feedback) that made the model far more useful and controllable.
The lesson: the technology had been maturing for years. A good product interface, at the right quality threshold, triggered mass adoption overnight.
Previous AI (voice assistants, recommendation engines) felt like a feature baked into something else. ChatGPT was the first AI that felt like talking to something. Users could: - Ask any question in plain English - Get a thoughtful, essay-quality response - Refine the response through follow-up conversation - Ask it to write code, poems, legal summaries, marketing copy — anything
This generality — one system doing many things previously requiring separate specialized tools — was the disruption.
AI has had multiple cycles of excitement and disappointment:
| Era | What happened |
|---|---|
| 1956 | Term 'Artificial Intelligence' coined; early optimism |
| 1970s–80s | 'AI Winter' — progress stalled; funding dried up |
| 1997 | IBM Deep Blue beats chess world champion |
| 2012 | Deep learning breakthrough — AlexNet wins image recognition contest |
| 2017 | Transformer architecture introduced (the foundation of all modern LLMs) |
| 2020 | GPT-3 released — first model impressive enough for practical use |
| 2022 | ChatGPT — mass consumer adoption begins |
| 2023–26 | Rapid model improvement; AI agents; multimodal models; vibe coding |
In 2017, Google researchers published a paper titled Attention Is All You Need, introducing the transformer architecture. This is the 'T' in GPT (Generative Pre-trained Transformer).
Without going into math: the transformer solved the problem of handling long, complex text — keeping track of how words relate to each other across many sentences. This enabled models to be trained at unprecedented scale on vast amounts of text, producing systems that could write, reason, and generate in ways no previous architecture could.