LLM (Large Language Model)
An LLM (Large Language Model) is a type of AI trained on vast amounts of text so it can understand, summarise, translate, and write human language. Almost every conversational AI you use today — including ChatGPT, Claude, and Gemini — is powered by an LLM under the hood.
Example: Ask an LLM to write a TikTok caption and it gives you lines that sound natural, because it has learned from millions of example sentences.
How does an LLM sound so human?
It is easy to assume an LLM has a giant encyclopedia database inside it where it looks up verified facts. In reality, an LLM is an extraordinarily capable next-word (or next-token) predictor.
During training, the model reads billions of sentences from books, articles, websites, and software code. Through that process, it learns how words, ideas, and sentence structures relate to one another across different situations. When you ask a question, the LLM calculates the most likely next piece of text, one token at a time, building a fluent reply in Malay, English, or code.
Why is it called 'Large'?
The word "Large" in Large Language Model points to two things that separate modern AI from older grammar-checking software:
- Massive training text: Rather than learning from a single subject, the model is trained on a huge breadth of public writing across dozens of languages, tones, and professions.
- Billions of parameters: Parameters are the adjustable mathematical weights inside the model's neural network where learned patterns live. Having billions of parameters allows the model to pick up subtle tone shifts, formal letter conventions in Malaysia, translation nuances, and programming logic.
What LLMs are great at (and where they struggle)
Once you see an LLM as a language and pattern engine rather than an all-knowing oracle, it becomes much easier to use well in everyday work or study:
- Great at: Drafting emails and formal letters, turning messy notes into clean meeting minutes, translating between English and Bahasa Melayu, summarising PDF documents you upload directly, and writing or explaining software code.
- Weak at (without outside tools): Doing complex arithmetic in its head without a calculator tool, quoting live news without web search (grounding), or citing specific local regulations from memory without making things up.
Three closely related terms worth knowing
As you use LLMs in daily work, three companion concepts explain almost all of their strengths and quirks:
- Token: The small chunks of words an LLM reads and writes under the hood when processing your prompt.
- Context window: The working memory limit of a single chat session — once a conversation or uploaded document gets too long, earlier details start dropping out.
- Hallucination: When the model predicts a smooth, confident-sounding sentence that happens to be factually wrong because it prioritises fluent language over verified truth.
Keeping these three ideas in mind helps you know when to start a fresh chat and when to verify facts against your original source.
Common questions
Are ChatGPT, Claude, and Gemini all LLMs?
Yes. All three are powered by Large Language Models (and many modern versions are also multimodal, meaning they can read images and files alongside text). Each has its own strengths in writing style, reasoning depth, and conversation memory.
Does an LLM actually understand what it is saying?
No. An LLM has no awareness, personal beliefs, or lived experience. It generates text by matching patterns across language and logic, which is why it can write impressively clear drafts while still needing a human to verify critical facts.
How do I stop an LLM from guessing or making things up?
Give it the source text directly inside your prompt (such as pasting the document you want summarised), tell it explicitly to say 'I am not sure' when facts are missing, and use web search grounding for current events.
Want to know which model fits your daily tasks best? Read the practical guides at /belajar/ or start step by step at /mula/.