💬 Chatbots and Language Models
5th Grade 🤖 Artificial Intelligence
A chatbot is a program you can type to and get an answer back, as if you were having a conversation. Today’s newer chatbots are built on a “large language model” — a neural network trained on an enormous amount of text that learned to predict which word will probably come next after the words before it.
What is a chatbot?
The word “chatbot” combines “chat” — a written conversation — and “bot,” short for robot. Simple chatbots have been around for many years: for example, a little window on a company’s website that asks “How can I help you?” and offers a few buttons. Chatbots like that follow a fixed script someone wrote ahead of time.
In late 2022, a chatbot that could write free-form answers was released to the public, and millions of people started trying chatbots that write poems, explanations, stories and even computer code. Since then, many companies have released chatbots of their own. They don’t follow a script — they produce brand-new text every time using a language model.
Predicting the next word
Remember the phone keyboard that suggests the next word? A language model does something similar, but on a gigantic scale. During training it “reads” huge amounts of text — books, websites, articles — and learns patterns: how sentences are built, which words show up together, what an answer to a question looks like.
When you ask a question, the model calculates which words (or pieces of words) fit as the next part, picks one of the best fits, adds it and then calculates the next one — and so on, word after word, until a whole answer is built. Because it learned from so much text, the answers can be long, organized and impressive.
After that first training, companies keep training the model with the help of people who rate its answers, so it will respond in a way that’s more helpful, polite and safe.
Sounds smart — but doesn’t understand like a person
A chatbot writes in natural, flowing language, so it’s easy to think someone on the other side is thinking and feeling. But a chatbot is a program. It has no feelings and no friends, and it doesn’t really get “hurt” or “happy,” even if it writes sentences like that. Scientists are still studying how much, and in what sense, these models “understand” what they write — but it’s clearly very different from the way a person understands.
Because the model builds an answer from what sounds right, it can write things that aren’t true with total confidence — making up a fact, a date or even a source that doesn’t exist. Mistakes like these are called “hallucinations.” It may also not know about things that happened after it was trained, or it may get math wrong. Some chatbots can search the internet too — but even then they can make mistakes or rely on an unreliable source.
How a chatbot can help
Used the right way, a chatbot can be a good helper: it can explain an idea you didn’t understand in different words, suggest ideas for a game or story, help you practice questions before a test, or check whether a sentence you wrote is clear. It’s like a friend who has read a lot — but who sometimes mixes things up very confidently.
The best way to use it is as a starting point, not the finish line: ask, read carefully, and then check the important information in a book, on a trustworthy website or with a teacher.
It’s also good to know that the way you ask changes the answer. A clear, detailed question — for example, “Explain to a sixth grader, in simple words, what photosynthesis is, with an example from home” — will usually get a more helpful answer than a one-word question.
Safety rules for chatting with a chatbot
Never type personal details to a chatbot: your full name, address, phone number, school name, passwords or photos of yourself. Conversations with chatbots are often saved by the company that runs them.
Double-check every important fact in another source. If a chatbot writes something that scares you, confuses you or seems inappropriate — stop and tell an adult. And remember that many chatbots have age limits, so use them only with your parents’ permission and according to the rules.
Key facts
- A language model writes text by predicting the next word, over and over.
- Language models are trained on enormous amounts of text.
- Chatbots that write free-form answers became widely available to the public in late 2022.
- A chatbot can write wrong things with confidence — these are called “hallucinations.”
- A chatbot has no feelings, and you should never share personal details with it.
Quick quiz
How does a language model write an answer?
What is a large language model trained on?
When did chatbots that write free-form answers become widely available to the public?
What is a chatbot “hallucination”?
A chatbot wrote “I’m happy to help you!” What’s true?
What should you never type to a chatbot?
What’s the right way to use a chatbot’s answer for a school project?
What’s the difference between a simple website chatbot and one built on a language model?
💭 Think and talk about it
- Why do you think it’s so easy to believe a chatbot “understands” us, even though it’s a program?
- Think of a question whose answer really matters to you. In which two other sources would you check it?
- How can a chatbot help you learn, and when does using it stop being fair?
Questions kids ask
Does a chatbot know everything?
No. It learned from text up to a certain point in time, so it may not know about recent events, may get facts wrong or may make information up. Even a chatbot that searches the internet can be wrong. So check anything important.
Does a chatbot remember me?
It depends on the program and its settings. Some chatbots save earlier conversations. Either way, it’s smart to assume that what you write may be saved — so don’t type personal details.
Is it okay to use a chatbot for homework?
That depends on your teacher’s and school’s rules. Usually it’s fine to use one to help you understand, but turning in work a chatbot wrote as if it were yours isn’t fair. If you’re not sure — ask your teacher.