what sign up
nixm nixm

← learn

What is a large language model? The thing behind every AI chatbot

October 14, 2026

A large language model, or LLM, is a neural network trained on an enormous amount of text to predict the next word. Do that well enough and the model absorbs grammar, facts, reasoning patterns and the shape of almost every kind of writing, which is why one model can draft an email, explain a law, translate a letter and write code. "Large" means hundreds of billions of internal numbers and training data in the trillions of words. Every AI chatbot you have used is an LLM with a chat interface and, usually, some tools attached.

What the name says

Language model: a system that assigns likelihoods to text, so it can say what word probably comes next. Your phone's keyboard has a tiny one. Large: the modern ones have hundreds of billions of parameters (the adjustable numbers inside) and were trained on trillions of words: a large share of the public web, books, code, transcripts. The size is not a detail; the abilities people care about appeared only when the models got big enough.

How it is made

  1. Pre-training. The model reads the pile and practises predicting the next word, billions of times, adjusting its parameters after each guess. This takes months on thousands of chips and is where the knowledge comes from.
  2. Tuning. The raw model would just continue text. People write example conversations and rate the model's answers; the model is adjusted toward the ones people prefer. This is what makes it answer questions, follow instructions and decline certain things.
  3. Tools and a window. The chat product adds a text box, a memory of the current conversation (the "context window"), and often tools: web search for current facts, code execution for arithmetic, file reading, image generation. The LLM decides when to use them.

What it is good at

Anything that looks like text in, text out, where the right answer is "plausible and well-formed": drafting, summarising, rewriting, translating, explaining at any level, writing and reading code, extracting structure from a mess. It is also, oddly, a decent general reasoner, because so much of the text it learned from contains people reasoning.

What it is bad at

Knowing when it is wrong. Exact arithmetic. Facts after its training cut-off. Anything inside your company or your head that it was never shown. Staying consistent across a very long document. It has no store of verified facts, so it can produce a confident, well-formed falsehood. Tools patch some of this; the person reading remains the checker. The limitations page has the full list.

LLM vs chatbot vs AI

The LLM is the engine. The chatbot is the car built around it: the interface, the memory, the tools, the rules about what it will and will not do. "AI" is the whole category, which also includes models that do not do language at all. When two chatbots feel different, it is often the car, not the engine: what tools they have, what they remember, how they were tuned.

What happens to the text you give it

This is the part the name does not tell you. The model itself forgets your conversation the moment it ends; it has no memory beyond the context window. The product around it usually does not forget: most consumer chatbots store your conversations on their servers, many use them as training examples for the next model unless you opt out, and some keep deleted chats for a period. So "what does the LLM know about me" is the wrong question; "what does the company keep" is the right one. The retention table answers it for the main tools. nixm's answer is that the engine is an LLM like the others and the car keeps no transcript: a distilled log of the session stays with the conversation for seven days, then dissolves, and nothing is used for training.

frequently asked

What is a large language model in simple terms?

A neural network trained on a huge amount of text to predict the next word. Doing that well means it absorbed grammar, facts and reasoning patterns, so it can write, summarise, translate and code.

Is ChatGPT a large language model?

ChatGPT is a chat product built around a large language model, plus memory of the conversation, tools like search and code, and tuning that makes it answer helpfully. Claude and Gemini are the same shape.

What does large mean in LLM?

Hundreds of billions of internal parameters and training data of trillions of words. The abilities people value appeared only once models reached that scale.

How is an LLM trained?

First it predicts the next word across a vast text pile for months on thousands of chips. Then people rate its answers and it is adjusted toward the preferred ones, which teaches it to follow instructions.

Does a large language model remember what I tell it?

The model itself does not, beyond the current conversation. The product around it usually stores your chats and may train on them unless you opt out. Check what the company keeps.

What is the difference between an LLM and AI?

AI is the whole field, including systems that do not handle language. An LLM is one kind of AI, the text-predicting neural network behind chatbots.

try it now — no account, no transcript kept. even the signal dissolves in seven days.

seed a signal →

Enter your email and receive a link to sign in. No passwords, just a secure signal to connect.

why connect

  • 20 free credits to start
  • your signals, remembered
  • buy more credits anytime
  • image & video generation features
or
+