Module 0 · Lesson
AI in one lesson
The vocabulary
| Term | Plain-language meaning |
|---|---|
| Artificial intelligence (AI) | An umbrella term for software that performs tasks we associate with human intelligence, such as understanding language or recognising patterns. |
| Machine learning (ML) | The main way modern AI is built. Instead of following hand-written rules, the system learns patterns from examples (data). |
| Predictive AI | ML that outputs a prediction, score or category. Example: a model that scores how likely an employee is to resign. |
| Generative AI | ML that produces new content such as text, images or audio. ChatGPT, Claude, Gemini and Copilot are generative AI products. |
| Large language model (LLM) | The kind of model behind text-based generative AI. It has been trained on very large amounts of text. |
| Prompt | The instruction or question you give the model. |
| Token | The unit of text a model reads and writes. A token is roughly a short word or part of a word. |
| Context window | The maximum amount of text (measured in tokens) the model can take into account at one time, including your prompt, any documents you supply, and its own reply. |
| Hallucination | Output that sounds fluent and confident but is false or not supported by any source. |
| Agent | An AI system that can take actions (send an email, update a record) by using tools, often across several steps, rather than only producing text. |
How a large language model actually works An LLM is trained by showing it enormous quantities of text and having it repeatedly predict what comes next. Over many rounds it becomes very good at continuing text in ways that match the patterns it has seen. When you type a prompt, the model generates a reply one token at a time, each time choosing a likely next token given everything before it.
That one fact explains most of what you need to know about using these tools well:
| Property of LLMs | What it means for your HR work |
|---|---|
| They generate plausible text; they do not look facts up in a database. | Output can be wrong while sounding right. Anything factual (dates, legal requirements, policy details, numbers) must be checked against a source. |
| Property of LLMs | What it means for your HR work |
|---|---|
| They only know what was in their training data, plus what you give them. | The model does not know your company’s policies, people or systems unless you supply that information in the prompt or attach it as a file. |
| Training data stops at a cut-off date. | The model may not know recent legal changes unless it has a web-search tool, and even then you should check the sources it cites. |
| They learned from human-written text. | They can reproduce stereotypes and biases present in that text. |
| Output involves an element of chance. | The same prompt can produce different answers on different runs. Consistency has to be designed in (Module 2). |
| They respond to the instructions and context given. | Better instructions and better context produce better output. This skill is called prompting (Lesson 1.4). |
| By default a new chat starts fresh. | Unless a memory or project feature is switched on, the model does not remember earlier conversations. |
What these tools are good and bad at
| Usually strong | Usually weak or risky |
|---|---|
| Drafting and rewriting text in a requested tone Summarising long documents you supply Reformatting (notes into a table, a policy into a checklist) Brainstorming options and questions Sorting text into categories you define Translating and simplifying language Explaining concepts at a chosen level | Stating facts without a source to work from Arithmetic done “in its head” rather than with a calculation tool Knowing anything about your organisation unprompted Recent events and current law Consistent, fair judgements about individual people Keeping information confidential unless the tool and settings are set up for it Telling you when it does not know |
Three ways AI shows up in HR work
- Chat assistants. You type a prompt and get a response. You remain in control of every step. Modules 1, 2 and 4 use this mode.
- Custom assistants. A chat assistant that has been given standing instructions and a set of reference documents, so that other people can ask it questions. Module 3 builds one.
- Automations and agents. AI embedded in a workflow that runs without you typing each time, or that takes actions in other systems. Module 5 builds one, with human checkpoints.
There is also a fourth category you will meet as a buyer more than a builder: AI features embedded in HR software, such as an applicant tracking system that scores or ranks candidates. These are often predictive models, they sit close to decisions about individuals, and they are where most HR-specific regulation is aimed. Lesson 1.3 covers how to question a vendor about them.
Jot down key ideas in your own words. Saves automatically.