Lesson finished

Answers from your documents (RAG)

This lesson is marked as finished on this device. Nothing here is a test, and you can leave whenever you like.

Your progress needs JavaScript, which is off in this browser, so nothing here is kept.

Your badge is earned separately. A check-in is a few short questions near the end of a lesson, and answering them earns the badge. Go to the check-in

Your save code

Keep this code to restore your finished lessons and badges in another browser or on another device. Your progress stays in this browser until you do.

Or open this restore link on the other device:

Where next

The 2020 paper that used the name RAG. arXiv is a website where researchers share papers. The page has the summary and a link to the full paper. NLP in the title means natural language processing, the handling of human-language information by a computer. It is written for researchers. The page loads files from arxiv.org, static.arxiv.org, code.jquery.com and cdn.jsdelivr.net. A cookie is a small piece of data that a site sends to your web browser. It sets one cookie, named arxiv_labs, before any choice. We saw no other hosts. We checked this on 9 October 2026.

Lewis and others, Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks, abstract page on arXiv (opens in a new tab)

Tell us what you thought of this lesson

The lesson map

Hide my progress

How AI works

What AI is and how it answers

5 lessons

1.1
What AI is and is not
Can say what AI is, and what it is not
About 8 minutes, or 14 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
1.2
Rules or examples?
Can explain the difference between written rules and learning from examples
About 8 minutes, or 13 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
1.3
Tokens
Can show how AI splits text into tokens, and why it matters
About 7 minutes, or 12 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
1.4
The next word
Can explain how a chatbot writes one likely word at a time
About 9 minutes, or 15 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
1.5
Where your question goes
Can identify web search in this lesson's recorded answers
About 9 minutes, or 15 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.

How it learns, and where it goes wrong

5 lessons

1.6
Why they make things up
Can spot a made-up answer, and explain why chatbots make them
About 9 minutes, or 15 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
1.7
Training
Can explain how a model learns by reducing its mistakes on examples
About 9 minutes, or 16 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
1.8
Attention and the context window
Can explain how a chatbot uses earlier words, and why it loses track in a long chat
About 8 minutes, or 16 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
1.9
Images and sound
Can explain how AI recognises pictures and sounds, and why it fails outside what it was trained on
About 9 minutes, or 15 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
1.10
Feeds: how apps choose what you see
Can explain how a feed picks what you see next, and change it
About 9 minutes, or 14 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.

Prompting

Tell it what you need

4 lessons

2.1
Bad and good prompts
Can turn a vague request into a clear prompt
About 9 minutes, or 15 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
2.2
Context
Can give an AI the material it needs, and spot hidden text in what you paste
About 9 minutes, or 15 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
2.3
Roles and ground rules
Can tell an AI what job to do and what not to do
About 8 minutes, or 14 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
2.4
Examples
Can show an AI an example of the answer you want
About 7 minutes, or 12 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.

Shape it, check it, reuse it

3 lessons

2.5
Format
Can ask for an answer in the shape and reading level you need
About 7 minutes, or 13 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
2.6
Iterate and check
Can improve an answer step by step, and tell when asking again is not checking
About 9 minutes, or 15 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
2.7
A skill is a written procedure (SOP)
Can write a procedure card for a repeated task, and check what it changed
About 10 minutes, or 16 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.

Ethics and society

What goes into AI

4 lessons

3.1
Who shapes what AI shows
Can explain how people's choices shape what AI shows
About 8 minutes, or 13 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
3.2
Bias
Can name four ways unfair patterns get into AI, and one way to check
About 9 minutes, or 16 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
3.3
Privacy: where your words go
Can explain where what you type into AI can end up, and check it on this site
About 10 minutes, or 16 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
3.4
Environmental cost
Can compare AI's energy and water use with everyday things, using honest ranges
About 8 minutes, or 14 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.

Who answers for AI

4 lessons

3.5
Harm and misuse
Can use a real incident to show what went wrong and who was responsible
About 9 minutes, or 15 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
3.6
Work, creativity and credit
Can credit work made with AI, and say what a licence controls
About 9 minutes, or 15 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
3.7
The law: AI rules in Ireland and the EU
Can say which AI uses the EU bans, which are high-risk, and what Article 4 asks
About 10 minutes, or 15 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
3.8
What trustworthy AI needs
Can use the EU's seven requirements to judge an AI system
About 9 minutes, or 15 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.

Using AI well

Using AI for study

4 lessons

4.1
Check it before you use it
Can check an AI answer's claims and citations against real sources
About 10 minutes, or 16 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
4.2
Study honestly with AI
Can use AI for study in ways you can declare
About 9 minutes, or 14 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
4.3
AI detectors: what a flag proves
Can explain why a detector's flag is not proof
About 8 minutes, or 14 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
4.4
Privacy habits
Can decide what is safe to type into an AI tool, and read its terms for three answers
About 8 minutes, or 13 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.

Using AI day to day and at work

5 lessons

4.5
Real or fake?
Can name the checks to make before trusting or sharing a photo, voice message or video
About 8 minutes, or 14 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
4.6
When text tricks the AI
Can spot a hidden instruction aimed at an AI tool
About 8 minutes, or 14 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
4.7
Delegating at work
Can split a task into parts and decide which parts AI should do
About 8 minutes, or 14 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
4.8
AI for access
Can use AI to make hard material easier to reach, and check what was lost
About 8 minutes, or 14 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
4.9
When not to use AI
Can decide when a person, a book or your own thinking is the better tool
About 8 minutes, or 13 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.

Going deeper

How a model is made and tested

2 lessons

5.1
How a chatbot is made
Can describe the stages that turn a text predictor into a chatbot, and why answers change over time
About 10 minutes, or 15 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
5.2
Evaluation: keep, fix or switch off?
Can read an AI test report and decide to keep, fix or switch off a system
About 10 minutes, or 16 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.

Building with a model

5 lessons

5.3
Agents and tools
Can follow what an AI agent does step by step, and say who answers for it
About 10 minutes, or 16 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
5.4
Anatomy of a skill file
Can read a skill file and spot what makes it reliable or vague
About 9 minutes, or 16 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
5.5
Answers from your documents (RAG)
Can explain how a system finds passages before it answers, and why it can still misquote them
About 10 minutes, or 16 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
5.6
Prompt, RAG or fine-tune?
Can choose between prompting, RAG and fine-tuning for a task
About 9 minutes, or 16 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.
5.7
Open models on your own machine
Can decide when an open model on your own computer suits a task
About 9 minutes, or 16 with the Deep layers
Not yet
No badge yet. Answer the check-in questions in the lesson to earn it.

What you can do

How AI works
- What AI is and is notCan say what AI is, and what it is not
- Rules or examples?Can explain the difference between written rules and learning from examples
- TokensCan show how AI splits text into tokens, and why it matters
- The next wordCan explain how a chatbot writes one likely word at a time
- Where your question goesCan identify web search in this lesson's recorded answers
- Why they make things upCan spot a made-up answer, and explain why chatbots make them
- TrainingCan explain how a model learns by reducing its mistakes on examples
- Attention and the context windowCan explain how a chatbot uses earlier words, and why it loses track in a long chat
- Images and soundCan explain how AI recognises pictures and sounds, and why it fails outside what it was trained on
- Feeds: how apps choose what you seeCan explain how a feed picks what you see next, and change it
Prompting
- Bad and good promptsCan turn a vague request into a clear prompt
- ContextCan give an AI the material it needs, and spot hidden text in what you paste
- Roles and ground rulesCan tell an AI what job to do and what not to do
- ExamplesCan show an AI an example of the answer you want
- FormatCan ask for an answer in the shape and reading level you need
- Iterate and checkCan improve an answer step by step, and tell when asking again is not checking
- A skill is a written procedure (SOP)Can write a procedure card for a repeated task, and check what it changed
Ethics and society
- Who shapes what AI showsCan explain how people's choices shape what AI shows
- BiasCan name four ways unfair patterns get into AI, and one way to check
- Privacy: where your words goCan explain where what you type into AI can end up, and check it on this site
- Environmental costCan compare AI's energy and water use with everyday things, using honest ranges
- Harm and misuseCan use a real incident to show what went wrong and who was responsible
- Work, creativity and creditCan credit work made with AI, and say what a licence controls
- The law: AI rules in Ireland and the EUCan say which AI uses the EU bans, which are high-risk, and what Article 4 asks
- What trustworthy AI needsCan use the EU's seven requirements to judge an AI system
Using AI well
- Check it before you use itCan check an AI answer's claims and citations against real sources
- Study honestly with AICan use AI for study in ways you can declare
- AI detectors: what a flag provesCan explain why a detector's flag is not proof
- Privacy habitsCan decide what is safe to type into an AI tool, and read its terms for three answers
- Real or fake?Can name the checks to make before trusting or sharing a photo, voice message or video
- When text tricks the AICan spot a hidden instruction aimed at an AI tool
- Delegating at workCan split a task into parts and decide which parts AI should do
- AI for accessCan use AI to make hard material easier to reach, and check what was lost
- When not to use AICan decide when a person, a book or your own thinking is the better tool
Going deeper
- How a chatbot is madeCan describe the stages that turn a text predictor into a chatbot, and why answers change over time
- Evaluation: keep, fix or switch off?Can read an AI test report and decide to keep, fix or switch off a system
- Agents and toolsCan follow what an AI agent does step by step, and say who answers for it
- Anatomy of a skill fileCan read a skill file and spot what makes it reliable or vague
- Answers from your documents (RAG)Can explain how a system finds passages before it answers, and why it can still misquote them
- Prompt, RAG or fine-tune?Can choose between prompting, RAG and fine-tuning for a task
- Open models on your own machineCan decide when an open model on your own computer suits a task