This page contains companion resources for Chapter 4.
The full chapter is in the book.
Hallucination detection pipeline from the book — RAG + self-consistency + LLM-as-judge. Zero setup, runs in your browser.
Open in Google ColabHallucination rates change as models update. The book explains why. These links show current numbers.
The paper that admits guessing outperforms honesty on benchmarks.
Read paper →Open the Colab notebook above. Remove the RAG context and let the model answer from memory. Ask: "What was the population of Liechtenstein in 1937?" — watch the consistency layer catch divergence.