ChatGPT dropped last week. Here's why this feels different.

I've seen a lot of AI demos. I've used GPT-3 since it launched in 2020. I watched DALL-E 2 arrive in April and change how people thought about image generation. I've been in the GitHub Copilot preview. None of those felt quite like this.

ChatGPT launched on November 30th, 2022. Within five days it had a million users. I know a few of you reading this are rolling your eyes. Another AI hype cycle, another tool that will be forgotten in six months. I thought that too, for about the first 20 minutes.

What's actually different

The model underneath ChatGPT is called GPT-3.5, a fine-tuned version of the GPT-3 family trained with a technique called RLHF (Reinforcement Learning from Human Feedback). Put simply: instead of just predicting the next token, it's been shaped by human raters to produce responses that feel helpful, accurate, and not unhinged.

The result is something that feels qualitatively different from previous chatbot experiences. It doesn't just pattern-match. It follows multi-step instructions. It maintains context across a conversation. It argues back when you're wrong and (mostly) admits its own mistakes.

I asked it to help me draft a problematic email. Then to rewrite it with a different tone. Then to suggest why the recipient might have reacted badly. Then to role-play as the recipient. It did all of this, in sequence, remembering the context throughout.

That's not a parlour trick. That's a working memory with reasoning applied to a real task.

What it's actually useful for (so far)

In the week I've been using it, I've found real utility in:

• Drafting: Initial passes at emails, proposals, documentation. Faster than blank-page starting, easier to edit than to write.
• Explaining things: I gave it a complex Conditional Access policy and asked it to explain it in plain English for a customer. Nearly perfect first attempt.
• Debugging: Pasted in a PowerShell error, got a plausible diagnosis and three possible fixes within seconds.
• Summarising: Threw a 4,000-word Microsoft blog post at it. Asked for the key points. Accurate, concise, useful.

It gets things wrong. It confidently hallucinates facts. I asked it about a specific Microsoft announcement and it gave me a plausible-sounding answer that was factually incorrect. You cannot trust it blindly. But the raw utility, even with that caveat, is significant.

The thing nobody's saying clearly enough

The reason this feels different isn't the technology itself; it's the accessibility. GPT-3 has existed since 2020, but most people never used it because it required API access, prompt engineering knowledge, and a certain level of technical confidence.

ChatGPT removes all of that. It's a text box. You write a thing, it writes back. That's the entire interface.

The audience for this tool is not developers. It's not AI researchers. It's everyone. Your colleagues who have never thought about language models are using this thing right now, and some of them are going to start doing their jobs differently because of it.

What I don't know yet

I don't know whether this holds up at scale. I don't know whether it gets substantially better (the rumours about GPT-4 are interesting). I don't know what the enterprise security and data privacy implications look like.

What I do know is that the week ChatGPT launched felt like a line being crossed. Not in a science-fiction way. In a "this is going into my workflow and I'm going to have to figure out what that means" way.

Start figuring it out now. Don't wait until someone else has to explain it to you.