How AI language models work, and where they fail

Ages 13 to 17 Read alone · about 55 minutes

Read together

A large language model is a computer program trained on huge amounts of text. It learns patterns in language and, given some words, predicts likely next words. Chatbots are built on this idea. The key design behind many of them, called the transformer, was described by researchers at Google in 2017.

These systems can be useful: they can explain, summarise, translate and help brainstorm. But they can also state false things with confidence, make up sources, repeat unfair stereotypes from their training data, and fail at tasks that seem easy. They do not check facts unless built to.

So use them as a tool, not as an authority. Check claims against reliable sources. Be honest about when you use AI for school or work. Do not share private information with a chatbot. This site uses AI in its writing, and says so openly, because honesty about tools matters.

Do it

1. Play “predict the next word”: one person reads half of a familiar sentence, the others guess the rest. Notice how patterns help, and where they fail with new sentences.
2. With a grown-up, ask an AI tool five questions you already know the answers to, including one tricky one. Check each answer against a reliable source. How many were right?
3. Ask it for a source on something, then try to find that source. Does it exist?
4. Write a short personal policy: when would you use AI, when not, and how would you say so honestly?

Talk about it

• Why can an AI sound sure and still be wrong?
• How would you tell someone you used AI to help?

For the grown-up

AI tools are now part of school and work, so teens benefit from learning how they work and where they fail. The 2017 paper “Attention Is All You Need” introduced the transformer design. The US National Institute of Standards and Technology published an AI Risk Management Framework in 2023, about trustworthy and responsible use. Talk with your teen about honesty, schoolwork rules, privacy and critical thinking.

Mark as done

Next lesson →

Where this comes from

• Ashish Vaswani et al., “Attention Is All You Need”, Advances in Neural Information Processing Systems 30 (2017)
• Emily M. Bender, Timnit Gebru, Angelina McMillan-Major and Shmargaret Shmitchell, “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?”, Proceedings of FAccT (2021)
• National Institute of Standards and Technology (NIST), Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1 (2023)

Written with AI assistance and checked against the sources listed. Please read it before sharing it with a child.