What is a large language model?

A large language model is the engine inside ChatGPT, Claude, Gemini and nearly every “AI” product the Obsolete have discovered since 2022. Strip away the marketing and it does one thing: given some text, it predicts what text should come next. Do that billions of times with a big enough model and enough training data, and “predict the next word” starts to look a lot like writing, reasoning and coding.

How it works, in plain English

1. Training. The model is fed a vast corpus of text — books, websites, code, documentation — broken into small chunks called tokens. For each chunk it guesses the next token, checks the real answer, and nudges its internal numbers (parameters) to guess better next time.
2. Architecture. Modern LLMs are built on the transformer, a design that lets the model weigh how every word in a passage relates to every other word.
3. Tuning. A raw model is then refined with human feedback and examples so it follows instructions, refuses some requests and behaves like a helpful assistant rather than an unhinged internet echo.
4. Inference. When you type a prompt, the model generates a reply one token at a time, each choice conditioned on everything before it — up to the limit of its context window.

An analogy for the meatbags

Imagine someone who has read every library on Earth but has never left the building. Ask them anything and they will answer fluently, drawing on patterns from millions of books. Usually they’re right. Occasionally they confidently describe a book that was never written. That is an LLM — and that occasional fiction is hallucination.

Why it matters for the Chosen

LLMs turn language work into something you can buy by the token. Drafting, summarizing, researching, translating, classifying support tickets, writing code — tasks that used to cost an hour of human attention now cost seconds and pennies. The people who profit are not the ones who ask a chatbot a question once a week; they are the ones who wire LLMs into repeatable workflows and then check the output like a suspicious editor.

What LLMs are good at: first drafts, rewriting, summarizing long documents, brainstorming, explaining concepts, boilerplate code, extracting structured data from messy text.

What they are bad at: guaranteed facts, exact citations, precise arithmetic without tools, and anything that happened after their training data ends (unless connected to search or your documents via RAG).

Examples

• ChatGPT runs on OpenAI’s GPT family.
• Claude is Anthropic’s model family (currently Opus, Sonnet and Haiku).
• Gemini is Google’s.
• Llama, Mistral, Qwen and DeepSeek release open-weight models you can run yourself.

Learn this word properly. It is the foundation every other term in this Lexicon is stacked on.