What is an LLM

What is it

An LLM (Large Language Model) is an AI model trained to understand and generate text. It does this by predicting what comes next: given a context, it calculates which token is most likely to follow.

That seemingly simple idea is what makes it possible for a model to answer questions, write code, translate languages, summarize documents, or hold a coherent conversation. It's all text prediction at scale.

"Large" isn't just a marketing term — it refers to the number of parameters (the numerical values the model learns during training). Modern models have anywhere from billions to trillions of parameters. That scale is what gives them generalization ability: they don't memorize answers, they learn patterns from human language.

Mental model

Think of a system that has read practically the entire internet, every digitized book, all the public code on GitHub. Not to memorize, but to learn the structure of language: which words go together, how arguments are built, what typically follows "the result was".

When you ask it a question, the model doesn't "look up" the answer in a database. It constructs the answer token by token, choosing at each step what makes the most sense given what it's already written.

A token isn't a word — it's a fragment of text. "developer" might be 2 tokens. A special character can be 1. On average, a token is about 4 characters in English. This matters because models have a limit on how many tokens they can process at once — the context window.

How it's used

The most direct way to use an LLM is through an API. You send a message, you get a response.

from anthropic import Anthropic
 
client = Anthropic()
 
response = client.messages.create(
    model="claude-opus-4-5",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Explain what tokenization is in one sentence."}
    ]
)
 
print(response.content[0].text)

What you send is called a prompt. How you construct it directly affects the quality of the response — that's what prompting is about.

LLMs can also serve as the foundation for more complex systems: agents that use tools, RAG pipelines that combine them with knowledge bases, or workflows where multiple models collaborate. But all of that builds on this foundation: a model that predicts text.

When to use it / when not to

Use it when:

• You need to process or generate text with flexibility: summarizing, translating, classifying, extracting information.
• The task requires reasoning over natural language.
• The input might be ambiguous or in unstructured formats.
• You want a conversational interface to expose functionality in your system.

Don't use it when:

• You need deterministic precision. LLMs are probabilistic — for exact calculation, use code.
• The task requires real-time, up-to-date information. The model has a knowledge cutoff.
• You're handling sensitive data that can't leave your infrastructure (unless using on-premise models like Llama).
• You need full auditability of why a specific decision was made.

Related concepts

• What is a token — the basic unit LLMs process
• What is a context window — how much information the model can "see" at once
• What is an embedding — how LLMs represent meaning numerically
• Prompting — how to instruct the model to get better results
• What is an AI agent — how LLMs are used as the engine of more complex systems

External resources

Resources are managed from the admin panel.

History and evolution