What a language model is, and why it sometimes makes things up
ChatGPT, Gemini and Claude all work the same way under the hood. Here is how they write, why they get things wrong so confidently, and how to use them safely.
When you ask ChatGPT, Gemini or Claude a question, the answer sounds like it came from someone who knows the subject. What is actually on the other side is a language model: a program trained to guess, very accurately, what the next word in a piece of text should be.
That sounds modest. Yet this one simple idea, repeated billions of times, is what lets a chatbot write an email, summarize a contract or explain quantum physics to a ten-year-old.
How it learns
The model is trained on an enormous amount of text: books, websites, articles, code. During training it sees incomplete sentences and tries to predict what comes next. Every time it misses, it nudges its “weights”, billions of internal numbers that store the patterns of language.
After months of training on thousands of chips, you get a system that has picked up grammar, style, widely repeated facts and even some reasoning. Nobody wrote those rules by hand. They emerged from the patterns.
An extra stage, guided by human feedback, teaches the model to be helpful and polite. That is what turns a “text autocomplete” into a conversational assistant.
Why it makes things up
This is the part that trips people up. The model does not look facts up in a database. It generates the text that seems most likely. Most of the time, the most likely text is also the correct one. Sometimes it isn’t.
When information is missing, the model rarely says “I don’t know”. It fills the gap with something plausible: a book that was never written, a wrong date, a law with the wrong number. Researchers call this a hallucination, and it shows up most often in three situations:
- Very recent events that happened after the model was trained.
- Obscure details, like the name of a small town’s council member.
- Exact citations, with page numbers, links or word-for-word quotes.
Newer assistants reduce the problem by searching the web and showing where each claim came from. That helps a lot, but it does not remove the risk.
Getting the most out of it
A few simple habits make a big difference:
- Give context. “Write an email” gets you little. “Write a short, polite email to my landlord asking them to fix the garage light” gets you much more.
- Ask for sources when the topic matters, then open the links and check they actually say what the model claims.
- Use it for what it does well: drafting, rewriting, summarizing, translating, brainstorming and explaining concepts.
- Be suspicious of very round or very specific numbers that show up without an explanation.
Think of a language model as a brilliant, fast, extremely well-read intern who never admits they don’t know something. With supervision, it saves you hours. Without it, it can land you in an awkward spot.
