Clicked Gallery
What is a large language model (LLM)?
Highlighted from a real engineering doc. Explained by Clicked.
Used in a sentence
Engineering Notes · AI Systems
The system is built on a large language model trained on several trillion words of text.
The reader highlighted one word in the docs. Clicked explained the technical term “large language model” in simple terms:
Explained in three depths
●○○
Overview
A large language model (LLM) is a program that predicts what text comes next. It is built in two stages. Pretraining reads an enormous amount of writing until those guesses get good. That alone produces something that continues your sentence rather than answering it. A second stage teaches it to answer questions instead. Both stages are complete long before you type anything into it.
●●○
Detail
Everything an LLM does rests on one operation repeated over and over: look at the text so far, produce the most likely next piece. Type "the capital of France is" and it answers Paris, because that pattern appeared constantly in what it read rather than because it checked a source. On its own that only gives you a machine that keeps writing from where you stopped. The second stage, reinforcement learning from human feedback, has people rank competing answers until it learns to reply rather than ramble. Why not just build a database of answers? Because a database returns only what someone put in it, and prediction handles the questions nobody thought to write down. The cost is that sounding right and being right are separate skills, so it can describe events that never happened. It also appears to remember you, through two separate mechanisms. Within a chat, the whole conversation is sent back every time you press enter, filling the context window. Across chats, products like ChatGPT keep notes about you in ordinary storage and paste them in later. Either way it is handed a reminder rather than recalling anything.
●●●
Analogy
Predictive text on your phone does a tiny version of this. Type "see you" and it offers "tomorrow", because that is what usually followed in the messages it has seen. A large language model is that idea with vastly more reading behind it and a much longer view of what came before. The jump in scale is what turns a next-word guess into something that can hold a conversation.
AI explanations may contain errors · Not professional advice
Same facts, different vibe — Slang mode 😎
●○○
Overview
Strip away the mystique and an LLM is a next-word guesser with an enormous reading habit. Left like that it would happily carry on writing from where you stopped and never once answer you. A second training stage, where humans sat there picking the better of two replies, is what turned the rambler into something that responds. 😎
●●○
Detail
An LLM has read a preposterous amount of text and learned what tends to follow what. The raw version of one is genuinely unhelpful, though. Ask it a question and it might reply with three more questions, because that is a perfectly normal thing to find after a question in the wild. Somebody had to sit through thousands of answer pairs marking which one was better, and that is where the polite assistant came from. So it is not a search engine and it is not looking anything up, which is why it can hand you a citation that never existed and sound thrilled about it. The memory thing catches people too. Inside a chat it seems to remember because the entire conversation gets posted back to it every single message. Across chats, the app keeps notes about you in a file somewhere and slips them in at the top of the next one. Neither is the model remembering. It is being handed a reminder and reading it like anything else you paste in. Which is why deleting those saved notes wipes the relationship instantly. 😎
●●●
Analogy
Someone who has read every cookbook ever written and never once cooked. Ask what goes with lamb and the answer is excellent, because that sentence has been written ten thousand times. Ask whether your specific lamb is done and you get an equally confident answer, delivered from a chair, in another building. 😎
AI explanations may contain errors · Not professional advice
Formal definition — The same term, explained the usual way
A large language model is a neural network trained on a large corpus to predict subsequent tokens from preceding ones, with parameter counts typically in the billions. Pretraining on this objective yields a text continuation system; fine-tuning on instruction examples and reinforcement learning from human feedback are applied afterwards to produce assistant behaviour. Parameters remain fixed at inference, and any apparent recall within a session derives from the conversation being resupplied in the context window.
Want Clicked to explain terms like “large language model” directly in your browser — including on PDFs?
Add to Chrome — Free50 free Explanations · No credit card required
More from the gallery
What is a Neural Network?
Math that learns the rules from examples instead of having them written, and why that wins.
What is RLHF (Reinforcement Learning from Human Feedback)?
The step that turns a text predictor into an assistant, and what it quietly teaches the model.
What is a Context Window?
How much text an AI can hold in mind at once — and why long chats forget their start.
What is Tokenization?
How AI chops text into chunks before reading — and why some languages cost more.
What is Inference (in AI)?
The model doing its actual job, and why the cheap half ends up costing the most.
What is an AI Hallucination?
Why AI states made-up facts with total confidence, and why you can't hear it happening.