What is a Large Language Model (LLM)? Explained for kids

A Large Language Model (LLM) is a kind of AI that's been trained on huge amounts of text and can read, write, and chat in human language. ChatGPT, Claude, and Gemini are all LLMs. The word "large" matters — these models have billions or trillions of internal numbers.

How to explain it to a 7-year-old

🧒 "It's an AI that read almost everything on the internet and now can write or talk back to you. It doesn't actually understand what it reads — but it's really good at sounding like it does."

How to explain it to a 14-year-old

🎒 "An LLM is a neural network trained on a massive text corpus to predict the next word. ''Large'' refers to scale (billions of parameters). At a certain size, models can do tasks they weren't explicitly trained for — translate, summarize, code — by leveraging patterns from training. That's what makes modern LLMs powerful."

Examples of LLMs

• GPT-4, GPT-4o (OpenAI / ChatGPT)
• Claude (Anthropic)
• Gemini (Google)
• Llama 3 (Meta — open source)
• Mistral (open source, French)

What LLMs can do well

• ✅ Write text in a chosen style or voice
• ✅ Translate between languages
• ✅ Summarize long documents
• ✅ Brainstorm ideas
• ✅ Explain concepts (with caveats — see hallucinations)

What LLMs are NOT good at

• ❌ Math (they predict text; they don't compute)
• ❌ Up-to-date facts (training data has a cutoff)
• ❌ Knowing what's true (they predict plausible, not factual)
• ❌ Remembering past conversations (unless the app stores them)

Where this comes up in Chippu

Band B (b2-2) introduces LLMs in the "Types of AI" lesson. Band D (d2-3) goes into how transformers (the architecture behind LLMs) work.

Related terms

• Generative AI — the bigger umbrella
• Chatbot — what most people interact with LLMs through
• Hallucination — LLM failure mode
• Prompt