What are LLMs and AI agents?

What's an LLM?

LLM stands for large language model. "Large" because it's trained on an enormous amount of text, "language" because words are what it works with, and "model" because it's a mathematical model of how language tends to flow.

When you ask it something, it writes the answer a little at a time, each time predicting what would most likely come next, based on the patterns it learned.

Why it can sound sure and still be wrong

An LLM is built to produce answers that sound right, and most of the time sounding right and being right line up. But it has no built-in fact-checker. When it doesn't really "know," it can still produce a smooth, confident answer with a made-up date, number, quote, book title or web link. This is called a hallucination.

Many assistants can now search the web and show sources, which helps. Still, for anything that matters (health, money, legal, travel times), check an official source. Practice spotting mistakes in AI or Not?

Three words you'll hear a lot

• Training: the long, expensive process where the model learns patterns from huge amounts of text (and sometimes images and audio). The material it learns from is called training data. After training, a model doesn't automatically know about events that happened later, unless the app looks things up on the web.
• Prompt: whatever you type (or say) to the AI. Clearer prompts get better answers. Try the Prompt Builder game or our guide to asking your first questions.
• Tokens and context: AI reads text in small chunks called tokens (often a short word or part of a word). Each model can only "hold in mind" a certain amount at once, called its context window. In a very long chat, it may lose track of things you said much earlier. Starting a new chat for a new topic helps.

What's an AI agent?

A regular chatbot answers. An AI agent does. You give it a goal, and it works out the steps and uses tools to carry them out, such as:

• browsing websites and comparing what it finds,
• filling in forms or adding things to an online cart,
• drafting and (if you allow it) sending emails or calendar invites,
• working with files you've shared with it.

Good agents pause and ask your permission before important steps, like paying for something or sending a message.

Everyday examples

The upside

• Saves time on boring, multi-step chores: comparing prices, filling in the same details again and again, sorting through information.
• Helps people who find screens hard, for example by doing the clicking while you describe what you want.
• Works while you do something else. Some agents can keep going in the background and report back.

Staying safe with agents

• Give only the access it needs. If it only needs your calendar, don't connect your email and bank too. You can usually disconnect apps later in settings.
• Review before it sends or buys. Read the email, check the cart, confirm the date. Keep "ask me first" turned on for payments and messages.
• Use official apps only. Get agents from the company's own website or the official app store, not from ads or links in messages.
• Watch for scams wearing an "AI" costume. A message saying "Your AI assistant needs you to verify your card" or a "smart AI investing agent" promising guaranteed returns is a scam. Real companies don't ask for passwords or codes in a chat.
• Agents can be tricked, too. A sneaky website can hide instructions meant for the AI. Another reason to keep permissions small and check before anything important happens.

Try this today

• Ask an AI a question you know the answer to, and notice how confident it sounds either way.
• Start a new chat when you change topics, and see if the answers get sharper.
• Play Agent or Chatbot? to lock in the new words.