What Is an LLM? Large Language Models Explained for HR
Large language models are the engines behind ChatGPT, Claude, Gemini, Copilot and most new HR AI features. You do not need to know the mathematics, but you do need to know what they are good at, and what they are not.
Short answer
A large language model (LLM) is an AI system trained on enormous amounts of text to predict the most likely next piece of text, which lets it understand instructions and generate fluent language. LLMs power AI chatbots and writing assistants. They are excellent at drafting, summarising, rewriting and answering questions, but they can produce confident errors called hallucinations, so for HR use they should be grounded in trusted documents and their outputs checked.
Key takeaways
- LLMs generate language by predicting likely next words, not by looking up facts.
- They are strong at drafting, summarising, rewriting, translating and extracting information.
- Hallucinations are a built-in risk; grounding in your own documents reduces them.
- Data protection depends on the version you use, so HR should use approved enterprise tools.
How an LLM works, simply
Imagine the autocomplete on your phone, trained not on your messages but on a vast library of books, websites and documents, and made powerful enough to continue not just a word but an entire essay. That is the core idea of a large language model.
- Training: the model reads huge amounts of text and learns statistical patterns in how language fits together.
- Tuning: it is further trained, often with human feedback, to follow instructions and respond helpfully and safely.
- Use: when you write a prompt, the model generates a response one small piece at a time, each time choosing a likely continuation given everything before it.
Key concepts in plain language
| Concept | What it means | Why HR should care |
|---|---|---|
| Token | A chunk of text, roughly a word or part of one | Limits and pricing are measured in tokens |
| Context window | How much text the model can consider at once | Determines whether it can read a whole policy handbook in one go |
| Prompt | Your instruction | Better prompts give better outputs. See HR prompts |
| Hallucination | A fluent but false output | Risky for policy, pay and legal answers |
| Grounding / RAG | Answering from retrieved trusted documents | The main way to make HR chatbots accurate |
| Fine-tuning | Extra training on specialist data | Rarely needed for HR; grounding usually suffices |
| Knowledge cutoff | The point after which the model's training data ends | It may not know recent law changes unless given sources |
What LLMs are good at
- Drafting and rewriting documents in a chosen tone.
- Summarising long texts such as policies, reports or transcripts.
- Extracting structured information, such as skills from a CV.
- Answering questions from documents you provide.
- Translating and adapting content for different audiences.
- Theming large volumes of open-text feedback.
What LLMs are bad at
- Guaranteeing facts: without sources, they can invent details.
- Knowing your organisation: they do not know your policies unless you provide them.
- Current events and law: knowledge ends at the training cutoff.
- Precise calculations: they may make arithmetic errors unless connected to calculation tools.
- Consistent judgement about people: outputs can vary and reflect biases in training data.
Why LLMs hallucinate
An LLM is designed to produce plausible language. When it lacks the information needed, it may still produce a plausible-sounding answer rather than saying it does not know. Grounding the model in your documents, instructing it to answer only from them, asking it to cite sources and routing uncertain questions to people all reduce this risk. See how to build an HR chatbot.
Using LLMs safely in HR
- Use enterprise versions with contractual commitments on data use and retention.
- Avoid entering identifiable personal data unless the tool is approved for it.
- Ground answers in approved sources for anything factual.
- Keep a person accountable for outputs that affect people.
- Cover all of this in an AI policy for employees.
For related concepts, see machine learning in HR and the full AI glossary.
Related guides
- Generative AI for HR: Uses, Risks and How to Get Started
What generative AI is, where it helps HR, what can go wrong, and how to start safely.
- What Is Machine Learning in HR? How It Works, Uses and Limits
How machine learning learns from people data, where HR uses it, and its limitations.
- How to Build an HR Chatbot: An Eight-Step Guide
Eight steps from scope and content through grounding, escalation, testing and pilot.
- Common AI Terms Explained Simply: A Guide for HR Teams
Fifteen core AI terms with everyday analogies, HR examples and commonly confused pairs.
Frequently asked questions
What is an LLM in simple terms?
A large language model is AI trained on huge amounts of text to predict likely next words, which lets it understand instructions and write fluent language. It powers tools such as ChatGPT, Claude, Gemini and Copilot.
Is ChatGPT an LLM?
ChatGPT is an application built on large language models developed by OpenAI. The LLM is the underlying engine; the chat interface and extra features are built around it.
Why do LLMs make things up?
They are designed to produce plausible text, so when information is missing they may generate a convincing but false answer. Grounding in trusted documents and requiring citations reduces this.
Can an LLM read our HR policies?
Yes, if you provide them, either in the prompt or through a retrieval system that connects the model to your policy library. This is how accurate HR chatbots are built.