One million users in five days.

I keep coming back to that number. Instagram took two and a half months to hit a million. Facebook needed ten. Netflix took three and a half years. ChatGPT did it in five days. And everyone I’ve talked to this week — engineers, product managers, even my dentist — has an opinion about it.

I’ve been working in and around conversational AI at Google for over a year now. I’ve seen impressive demos. I’ve built prototypes with language models. None of that prepared me for the visceral reaction people are having to ChatGPT. This isn’t excitement about a new product. It’s something closer to a rupture.

What actually happened #

On November 30, OpenAI released ChatGPT as a free research preview. It’s built on GPT-3.5, fine-tuned using RLHF — Reinforcement Learning from Human Feedback, if you’re not steeped in the jargon. The underlying model isn’t dramatically different from what was already available through the OpenAI API. InstructGPT, published back in January, laid the theoretical groundwork. The capabilities have been improving steadily.

So why does ChatGPT feel different from everything that came before?

The interface. That’s it. A chat window. No API keys, no prompt engineering, no Python notebooks. You type a question in plain English and get a coherent, useful answer back. OpenAI took a state-of-the-art language model and put it behind the simplest possible interaction pattern.

The RLHF fine-tuning matters too. GPT-3.5 with RLHF produces noticeably better outputs than the base model — more aligned with what you actually asked for, less prone to the weird tangents and fabrications that made earlier models feel unreliable. The conversation feels natural in a way that previous chatbots never quite managed.

But the interface is the thing. The interface is always the thing.

The code red #

I’ll be careful here about what I say regarding internal reactions at Google. But the publicly reported “code red” is real in spirit, if not in those exact words. When a competitor launches a product that reaches a million users in five days — a product that sits directly in the path of your core business — you pay attention.

Search has been Google’s foundation for over two decades. The assumption has always been that search is a solved problem; the moat is infrastructure, index quality, and the feedback loop of billions of daily queries. ChatGPT challenges that assumption by demonstrating a fundamentally different interaction model. Instead of typing keywords and scanning ten blue links, you ask a question and get an answer.

Is it a better answer than Google would give you? Often, no. ChatGPT hallucinates. It presents fabricated information with the same confidence as accurate information. It has no concept of recency — its training data has a cutoff. For factual queries, Google Search is still more reliable.

But reliability isn’t the only thing users care about. The experience of having a conversation with an AI that understands your intent — even imperfectly — is compelling in a way that keyword search isn’t. People are willing to tolerate some inaccuracy for a dramatically better interaction.

Why “rupture” and not “evolution” #

I chose that word deliberately. Evolution implies continuity: things getting gradually better along an established trajectory. Chatbots have existed for years. Siri, Alexa, Google Assistant — all conversational AI products. They’ve improved steadily. That’s evolution.

ChatGPT doesn’t feel like the next step in that progression. It feels like a discontinuity. The difference between Google Assistant answering “What’s the weather?” and ChatGPT writing a college essay on Kantian ethics isn’t a matter of degree. It’s a different category of capability.

The rupture isn’t in the technology — transformers, attention mechanisms, RLHF, all of that is published research. The rupture is in the public consciousness. Before November 30, most people had never interacted with a large language model. Now millions have. The concept of “talking to an AI” shifted from science fiction to Tuesday afternoon.

What I’m thinking about #

Three things keep me up at night. Well, it’s 1 AM as I write this, so literally.

The information integrity problem. ChatGPT generates plausible-sounding text on any topic. It can’t distinguish between facts it learned from training data and patterns it’s interpolating. If millions of people start treating ChatGPT outputs as authoritative, we have a new vector for misinformation that’s qualitatively different from social media. At least with a Google search, you can evaluate the source. ChatGPT has no sources; it has weights.

The education disruption. Every professor I know is panicking. Student essays are about to become undetectable AI outputs. The entire assessment model of “write an essay to demonstrate understanding” is in trouble. Some schools will ban it. Others will try to adapt. The interesting ones will rethink what they’re actually trying to measure.

The developer tooling implications. I wrote about Copilot’s productivity numbers last month. ChatGPT takes that further — you can paste in code, describe a bug, and get a reasonably accurate diagnosis. You can ask it to write a function and get working code back (usually). The boundary between “AI code assistant” and “AI that understands software” is blurring faster than I expected.

Where this goes #

I genuinely don’t know. That’s not a hedge; it’s an honest admission that the pace of change has outrun my ability to predict. Six months ago, I would have said mainstream adoption of conversational AI was 3-5 years away. It happened in a week.

What I do know is that the companies and individuals who treat this as a curiosity rather than a signal will find themselves behind. The technology will get better. The interfaces will get more polished. The use cases will multiply.

This is not another chatbot. This is a new interaction paradigm for computing, and it arrived faster than anyone — including, clearly, the incumbents — was ready for.