AI chatbots can write, summarize and answer questions in natural language. Behind the friendly chat window is a large language model: a program trained on enormous amounts of text to predict what comes next.
Step 1: Text becomes tokens
The model does not read words the way we do. Text is split into small pieces called tokens, such as words, parts of words and punctuation, and each token is converted into numbers.
Step 2: Training
During training, the model adjusts billions of internal numbers, called parameters, so that its predictions come closer to real text. In the process it picks up patterns of grammar, style and the facts as they appear in its training data.
Step 3: Answering
To reply, the model generates one token at a time, each time choosing among likely continuations. A little randomness is built in, which is why the same question can produce different answers.
Step 4: Tuning
Developers then train the model further, often with feedback from people, so that it follows instructions and avoids harmful output.
What this means for you
- It can sound confident and still be wrong. Check names, dates, quotes and figures before you rely on them.
- It does not look things up unless it is connected to search or other tools.
- Do not paste passwords, ID numbers or private documents into tools you do not control.
- Use it for drafts, summaries and brainstorming, and keep a person responsible for final decisions.
Image: Map of data center infrastructure in the United States, 2025, NREL (public domain), via Wikimedia Commons.
Sources: Giles Thomas: What AI chatbots are actually doing under the hood · Miguel Grinberg: How LLMs Work, Explained Without Math

