Seven AI words explained in plain English, from tokens to hallucinations

A chatbot can state made-up facts with total confidence. That is called a hallucination, and it is one of seven AI terms Tom's Guide says every user should know, whether they are just starting out or already use AI often. The list also covers tokens, which some subscription plans use to set limits.

The basics

A large language model, or LLM, is the engine behind a chatbot. "Large" refers to the huge amount of text it learns from, such as books, websites, articles and research papers. "Language" is the human language it is built to understand and produce. "Model" means it learns patterns and relationships in words instead of just memorizing answers.

Generative AI is AI that creates new content from an instruction, called a prompt. That can be text, images, audio, video or computer code. The neon-soaked 80s photos and skateboarding dog videos people see online are examples. It makes something new instead of looking up something that already exists.

How AI learns

Machine learning is a way of teaching computers to spot patterns and make predictions from data, instead of having a programmer write every instruction. Netflix is the comparison used. It looks at what you watched, liked and skipped to suggest what to watch next. More data means better suggestions.

A neural network is a computer system loosely modeled on the brain. It has layers of connected nodes that process information and find patterns, much as neurons connect in a head. It is how AI can recognize that a photo shows a cat.

Where chatbots go wrong

A hallucination is when a chatbot gives information that is clearly wrong or entirely invented. The AI does not automatically know whether what it says is true. It can produce made-up statistics and cite fake sources. The advice is to double-check, even triple-check, what a chatbot tells you. That matters most for medical, financial or legal questions.

Getting better results

A prompt is simply what you type into an AI tool. Prompt engineering means writing detailed instructions so the tool understands what you want. Compare "give me dinner ideas" with a request that spells out:

• How many dinner ideas you want
• The most time they should take to prepare
• The most each meal should cost
• Which supermarket you will buy ingredients from

The more specific request tends to get a more useful answer.

What tokens mean for your plan

Tokens are small chunks of text that AI models read and produce. A chatbot does not treat a sentence as a string of whole words. It splits the text into tokens, which can be a short word, part of a longer word or a punctuation mark. Some models can only handle so much in one conversation. Some subscription plans mention token limits to show how much you can use before hitting a cap.