A help desk open around the clock - you type a question, you get a ready answer. A product of OpenAI that, since late 2022, made talking to AI something ordinary.
In practice ChatGPT is a help desk open 24 hours. You type "explain my lease agreement in plain language" or "write an email to a supplier who's running late" and you get text right away, instead of waiting in line or searching through ten sites. The difference from ordinary search is that it doesn't give you a list of links, it gives you a ready answer, put together specifically for your question.
Underneath sits a model trained to predict which word logically follows the previous one, over a huge amount of text from the internet. There's no database it "quotes" from - it generates the answer on the spot, word by word. That's why it sometimes sounds confident and wrong at the same time - the fact that the sentence is smooth doesn't mean it's true.
The company behind it, OpenAI, started as a nonprofit research lab in 2015, but today conversations with ChatGPT actually train and feed the next versions of the product. Students use it for homework, employees for email drafts, programmers for code, and companies build it into their own apps through paid access.
It affects you because since November 2022, every other chatbot you run into - in your banking app, in Word, on your phone - largely imitates exactly this form of conversation. ChatGPT set the template.
Don't Fall for the Headline
ChatGPT is the first interface that made people forget they're talking to a machine - and that's exactly where the trap is. A confident tone plus a mistake is a more dangerous combination than an obvious mistake, because it doesn't raise suspicion.
I use it every day, but I never take its word for anything without checking it against something outside. That's not some finicky personal caution - it's the only reasonable way to work with a tool whose job is to sound convincing, not to be right.