Sanity Context and content APIs (opens in a new tab)
The official reference for the tools used in this lesson.
Link
Handling the no-answer case
The hardest behaviour to get is a clean "I don’t know". Make it an explicit, allowed output instead of the model’s last resort.
Lesson 4.221m 33sAdvanced4.4k students
Chapters
21:33- 0:00Introduction: The Hallucination Problem
- 1:46Why Do LLMs Hallucinate?
- 4:19The Core Idea of RAG
- 7:10Smarter Chunking and Indexing
- 9:54Dense, Sparse & Hybrid Retrieval
- 12:28Reranking and Context Curation
- 14:56Self-Correcting & Agentic RAG
- 17:35Grounding, Citations & Verification
- 19:47Limitations, Open Challenges & Closing
What this lesson covers
- Detect when retrieval returned nothing relevant
- Make abstaining an explicit, acceptable outcome
- Offer a next step instead of a dead end
This lesson sits in Grounded Generation, part of Retrieval-Augmented Generation from Scratch. It assumes what came before it and leads directly into the next lesson in the module.
In this lesson you will
- Detect when retrieval returned nothing relevant
- Make abstaining an explicit, acceptable outcome
- Offer a next step instead of a dead end
Resources
Personal notes are coming soon.