If your feeds look anything like mine this week, you have seen the screenshots: poems about SQL, rap lyrics in Bahasa Indonesia, working Python code generated from a one-line request. So, what is ChatGPT, and does it matter for your business, or is this another tech demo that trends for two weeks and disappears?
ChatGPT is a chatbot released by OpenAI on November 30 as a free "research preview." You type a request in plain language, it answers in plain language, and it remembers the conversation so you can refine what you asked. Under the hood it is a large language model, the same family of technology behind GitHub Copilot, tuned specifically for dialogue. Reportedly over a million people signed up within the first five days, which explains why it keeps showing the "at capacity" page.
I have spent several evenings this week testing it against real work: code, business writing, explanations, Indonesian-language tasks. Here is my honest first read, written in week one, before the hype and the backlash have fully formed.
What it does genuinely well
I went in skeptical and came out surprised in specific places.
Drafting is its strongest suit. I asked it for a polite payment reminder email, a job description for a junior developer, and product descriptions in three tones. Every draft was 70 to 80 percent usable, produced in seconds. Not final copy, but a first draft that beats a blank page by a wide margin, especially if writing is not your strength.
It explains things patiently. Ask it to explain VAT, API integrations, or accrual accounting "like I'm a business owner, not an accountant," and the explanations are clear, structured, and adjustable. Ask a follow-up and it builds on what it already said. That conversational memory is what makes it feel different from anything before it.
Code assistance is real. I fed it error messages and got plausible diagnoses. I asked for a regex, a SQL query, and a small JavaScript function, and got working results faster than searching Stack Overflow. As someone who writes code daily, I can verify its output, which turns out to be the crucial skill.
Indonesian works better than expected. It is noticeably stronger in English, but it drafted a reasonable customer-service reply and translated marketing copy with decent fluency. Formal Indonesian is fine; natural casual register is hit and miss.
Where it confidently lies
Now the serious caveat, and I want to be blunt because the screenshots going around do not show this part.
ChatGPT will state false things with the exact same fluent confidence as true things. I asked it about Indonesian tax regulations and it produced specific-sounding rates and rules that were partly wrong. I asked for the biography of a moderately known Indonesian founder and it invented plausible details. I asked it to cite sources and it fabricated convincing references to articles that do not exist.
This is not an occasional glitch, it is how the technology works. The model generates text that is statistically likely to follow your question. Truth is not part of the mechanism. It also knows nothing recent: its training data ends in 2021, so it is unaware of anything from this year, and it cannot look anything up.
The practical rule that falls out of this: use it for tasks where you can verify the output, never for tasks where you would need to trust it. Drafting an email you will read before sending is safe. Asking it for tax rules, legal requirements, medical guidance, or facts you cannot check yourself is genuinely risky. It fails exactly when you can least afford it: when you do not know the answer either.
What this means for your business right now
My advice for this week is deliberately modest.
Try it personally, free, this month. The best way to calibrate your expectations is 30 minutes of hands-on use on your own real tasks. Ask it to draft something you actually need. Then ask it about a topic you know deeply, and watch how convincingly it gets details wrong. Both experiences are the education.
Do not connect it to customers. There is no official business product here, only a preview website, and the confident-nonsense problem makes it unsuitable for unsupervised customer contact. An AI that politely invents your return policy is worse than no AI. What businesses can realistically deploy today remains closer to the structured chatbots I described in AI chatbots for customer support: realistic expectations, which are boring but do not improvise.
Where I would experiment: internal drafting. First drafts of emails, announcements, job posts, product descriptions, and explanations of technical topics, always with a human reviewing before anything ships. Treat it like a fast, well-read intern who never says "I don't know." Useful, supervised.
Neither dismissal nor hype survives contact with it
Two camps are already forming, and I think both are wrong.
The dismissal camp says it is a toy that produces generic text and confident errors. True as far as it goes, but I have watched it compress 30-minute writing tasks into 3 minutes, this week, for free. Tools that save real time on real work are not toys.
The hype camp says this replaces writers, programmers, and search engines imminently. That ignores the fabrication problem, the 2021 knowledge cutoff, and the fact that a free research preview is not a business platform. It also ignores that quality still requires a human who knows what good looks like.
My actual takeaway is about direction. This year alone brought image generators like DALL-E 2, Midjourney, and Stable Diffusion, and now this. The pace is the story. Whatever ChatGPT is today, the interesting question is what this class of tools looks like after a few more turns of improvement, and whether OpenAI or others wrap it in products businesses can actually build on.
The takeaway: watch this space, hands on
What is ChatGPT? A genuinely impressive drafting and explanation tool wearing the costume of an all-knowing oracle. Use the tool, distrust the costume.
Concretely: spend 30 minutes with it this month, use it only where you can verify the output, keep it away from customers, and add it to the short list of technologies worth tracking through 2023. I make no prediction about mass adoption. I will say I have not been this curious about a new tool in years, and I do not say that often.