What actually happens when I ask ChatGPT a question?

It feels like magic, but what's happening under the hood is a super-fast prediction game. The AI isn't searching the web or 'thinking' like we do. It's generating one word at a time, each one based on the entire conversation so far. Think of it like a really aggressive autocomplete on your phone, but one that's read a huge chunk of the internet. When you type 'What's the capital of France?', the model doesn't pull up a fact sheet. It calculates that, based on its training, the most likely sequence of words to follow your question is 'The capital of France is Paris.' I've found this is the single most helpful thing to understand when you're starting out. It explains why these tools sometimes sound so confident but are completely wrong. They're not fact-checking; they're pattern-matching. For a concrete example, imagine you ask it to write a haiku about a soggy taco. It's never read about a soggy taco before. But it has seen millions of haikus and millions of descriptions of food. It blends the 5-7-5 syllable pattern with words associated with 'soggy' and 'taco' to create something new. The real tip here is to think of your prompt as setting a trajectory, not giving a command. The first few words you get back lock in a certain path, and the model just keeps following that path, one token at a time. That's the whole show. A very fast, very sophisticated word-guessing engine.