The Skill Explained
What Are Large Language Models (LLMs)?
A large language model (LLM) is a neural network trained on a huge amount of text to predict the next token, a piece of a word. Repeating that prediction lets it write answers, summaries and code. GPT, Claude, Gemini, Llama and Mistral models are all large language models.
Why AI Employers Ask for Them
Large Language Models have changed what software can do with text, so companies are adding them to products and internal tools. Employers want people who know how these models behave: their strengths, their mistakes, their cost and speed, and how to build reliable features on top of them. It is the most visible modern AI skill in job ads.
How It Works
An LLM reads text as tokens (pieces of words). Inside, a transformer network turns the tokens into numbers and computes, for each position, which earlier tokens matter. After training on vast text, the model is tuned further with human feedback and examples so it follows instructions and answers helpfully.
When you use one, you send a prompt and the model generates a reply token by token. Settings such as temperature control how varied the output is. The model only knows what was in its training data and what you put in the prompt (its context window), which is why techniques such as retrieval (RAG) add company documents at question time.
LLMs can sound confident and be wrong (hallucinations), so real applications add checks: structured outputs, citations, evaluation sets and human review where the stakes are high.
What You Do With It at Work
- Build a feature that calls an LLM API, for example to summarise support tickets.
- Write and test prompts, and ask for structured output such as JSON.
- Ground answers in company documents with retrieval.
- Measure quality, cost and speed, and choose the right model for the task.
- Add guardrails against wrong or unsafe answers.
Tools and Terms You Will Meet
- Token
- A piece of text the model reads and writes; costs and limits are counted in tokens.
- Context window
- How much text the model can consider at once.
- Hallucination
- A confident answer that is not true.
- Foundation model
- A large general model that many applications build on.
- Temperature
- A setting that makes answers more predictable (low) or more varied (high).
Industry Example: Pharma
A regulatory team uses an LLM to draft first summaries of long submission documents, with every statement linked to its source page so a reviewer can verify it before anything is sent.
A Common Mix-Up
An LLM does not look facts up the way a search engine does. Unless an application gives it documents or tools, it answers from patterns learned in training, which can be outdated or wrong.
How Deep Do You Need to Go?
- Know itExplain tokens, context windows, hallucinations and how an LLM generates text.
- Use itBuild features on LLM APIs with structured output, retrieval and evaluation.
- Master itChoose and adapt models, optimise cost and quality, and design safe LLM systems at scale.