The chatbots before it were scripts and decision trees: "Press 1 for billing, say 'support' for support." Step a single inch off-script and they fell over.
ChatGPT — built by a company called OpenAI — launched in November 2022 on a model called GPT-3.5 (that's just the version they shipped with; you can forget the number). Four things made it feel different the moment you used it, and you could feel all four without anyone explaining them:
- It held a conversation. Follow-ups worked. "Make it shorter." "Now friendlier." "What about the second point?" — same thread, and it kept up.
- It wrote things worth keeping. Not one-line FAQ answers. Actual paragraphs — emails, essays, marketing copy, explanations. Good enough to edit, not just laugh at.
- There was no friction. A browser. Free at the start. No developer kit, no "integrate our API" (an API is just how apps talk to each other behind the scenes — you needed none of that; you just typed).
- It could turn its hand to anything. One box for dozens of jobs — translating, code-shaped text, recipes, a stand-in debate partner, lesson plans. Rubbish at some, scary-good at others.
It's the same core trick you met in Lesson 1 — next-word prediction — just scaled and tuned until it crossed the line from toy to genuine tool. And because the replies were so fluent, your brain did the very human thing and assumed someone was home. Nobody was. But that illusion was brand new, and it hooked people within minutes.
Continue — the numbers behind the hysteria.