Home › Key concepts › Hallucination and grounding

Key concept · Crawl (beginner) · the same on every platform

Hallucination and grounding

Confident answers that are not true, and how to anchor answers to sources.

What it is

A model generates likely-sounding text, which is sometimes false: a made-up clause, a wrong number. Grounding means the answer must come from given sources, cite them, and say 'I don't know' when they don't cover it. You check grounding in code or with a judge.

Why it matters

It is the main reason AI answers need checks before they reach people.

An example

Asked about a clause not in the policy, a grounded system says it cannot find it instead of inventing one.

The common mistake

Assuming a bigger model does not hallucinate.

Where you meet it on each platform

The idea stays the same; only the product name changes.

JobAWSAzureGoogle CloudDatabricksOpen source
Evaluationprove it works on cases you wrote down, every releaseBedrock evaluationsFoundry evaluationsVertex AI Gen AI evaluationMLflow evaluationEval Harness Framework or promptfoo
Vector searchfind passages by meaning, not exact wordsOpenSearch Serverless / Aurora pgvectorAzure AI SearchVertex AI Vector Search / AlloyDBVector Searchpgvector or Qdrant

Related ideas

Where it shows up

Part of the Crawl level on these paths: AI Engineering (working with models) · LLM Engineering (quality and data).