AI, Safety and What Goes WrongLesson 21 of 73

Module 3 · Truth, sources and verification

Retrieval, and its three failure points

In this lesson

  • 9 min read
  • 6 reading sections
  • Knowledge check
  • Lesson 4 of 8 in this chapter

This chapter’s goalEstimate from the shape of a question how likely a fabrication is, verify a claim at a cost proportional to the consequence of being wrong, and explain why a cited source that exists is not yet a verified claim.

This chapter’s media

The film explains chapter ideas. The audiobook narrates the complete teaching text.

FormatEnglishहिंदी
Chapter filmNot availableWatch2:37
Full audiobookNot availableNot available
Retrieval-augmented generation, with its three failure pointsChunkDocuments cut intopieces of a fewhundred words; atable’s heading canland in one chunk andits numbers in anotherEmbedEach chunk becomes alist of numbers placednear chunks withsimilar meaningRetrieve the nearestchunksFailure one: the 2019policy scores assimilar as the 2024one. Failure two:nothing relevantexists and the nearestchunks come backanywayGenerate, ‘usingonly the materialprovided’Failure three: thesentence is writtenfluently whether ornot the retrievedpassage supports itThe instruction at the last step is a request, not a constraint. A system failing at retrieval and asystem failing at faithfulness need different fixes, and one accuracy number hides which you have.
The instruction at the last step is a request, not a constraint. A system failing at retrieval and a system failing at faithfulness need different fixes, and one accuracy number hides which you have.

Figure

A little at a time

Explore an idea, try it, then check what stays.

Step 1 of 8Free · no account needed
Lesson steps

What “it searched for that” actually means

Most serious AI products no longer answer from the model alone. They retrieve first. The pattern has a name (retrieval-augmented generation, usually shortened to RAG) and it is worth understanding at the level of the plumbing, because it is now behind most enterprise assistants, most “chat with your documents” tools, and the search summaries you meet daily.

The pipeline has four steps.

  1. Chunking. Your documents are cut into pieces, typically a few hundred words each.
  2. Embedding. Each chunk is converted into a list of numbers (an embedding) positioned so that pieces of text with similar meaning sit close together in that space.
  3. Retrieval. Your question is embedded the same way, and the chunks nearest to it are pulled out. Usually three to ten of them.
  4. Generation. Those chunks are pasted into the model’s context along with your question and an instruction like “answer using only the provided material”.

Your place and first choices stay on this browser. Classroom credit is recorded separately.

Stuck on something?Ask Rukmini, your AI tutor

How to cite: Addaly (2026) ‘Retrieval, and its three failure points’, in AI, Safety and What Goes Wrong. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-and-you/retrieval-does-not-mean-true (Accessed: 21 November 2026).