Generative AI · Fast-moving · Beginner

Large Language Model

Also known as: LLM

A very large neural network trained on text to predict the next token, which yields broad language, reasoning and coding ability.

What Large Language Model is

An LLM is a transformer with billions of parameters trained on trillions of tokens. Its single mechanical skill — predicting what comes next — turns out to support translation, summarisation, question answering and code once the scale is large enough.

How it works

Pretraining learns general language structure; supervised fine-tuning teaches instruction following; preference tuning shapes behaviour and refusals. At inference, text is tokenised, processed through attention layers and decoded one token at a time under sampling settings such as temperature and top-p.

Why it matters

LLMs are the interface layer of modern AI. Understanding tokens, context windows and their tendency to hallucinate is what separates productive use from disappointment.

Common uses

→Assistants for writing, analysis and research
→Customer support deflection and triage
→Code generation, review and migration
→Structured extraction from unstructured documents
→Classification and routing without labelled training data

Strengths

✓General-purpose across many tasks
✓No task-specific training needed for many jobs
✓Natural language interface

Watch for

✓Hallucinates plausible falsehoods
✓Knowledge cutoff and stale facts
✓Inference cost at scale
✓Sensitive to prompt wording

Continue exploring

Retrieval-Augmented Generation AI Hallucination Small Language Model Reasoning Model Multimodal AI