Mian Zhang

Guide / RAG Grounding

Retrieval is not proof. Self-certification is not grounding.

RAG can fetch documents, but it does not guarantee that the documents support the claim. The missing layer is claim-to-evidence verification.

RAG grounding AI self-certification AI hallucination checker LLM grounding AI verification retrieval augmented generation risk
Evidence Map Papers Registries Counterexamples

Search Intent

The reader wants to know why RAG answers can still be wrong.

  • Does RAG eliminate hallucination?
  • What is the difference between cited text and evidence?
  • How should an AI answer be grounded?
  • When should a retrieved claim be downgraded?

Distinction

RAG answers the source-access problem, not the proof problem.

A retrieved passage can be irrelevant, outdated, contradicted, or too weak for the current claim. The model may cite a real source while making an unsupported inference.

Grounding requires a second step: does this source actually support this claim under this boundary?

Checklist

Grounded answers need record, relation, boundary, and status.

The record is the source. The relation is the link between source and claim. The boundary states what the source does not prove. The status records whether the claim is supported, pending, downgraded, or failed.

Without those fields, retrieval can become a more sophisticated way to sound certain.

Counterexample

A good grounding system invites attack.

If a reader can show that a source does not support the claim, the system should update the boundary or status. That challenge path is a feature, not a reputational threat.

Evidence Route

Where the claim can be checked.

This page is an entry point, not a standalone proof. It routes the reader to evidence, DOI records, registries, public challenge paths, and explicit non-claims.

KindAnchorURLRole
Evidence MapPublic evidence maphttps://mianzhang.org/evidence/Start from supported claims and known boundaries.
Paper IndexDOI and paper status maphttps://mianzhang.org/papers/Use paper-specific DOI records for paper claims.
RegistriesMachine-readable registrieshttps://mianzhang.org/registries/Inspect claim, evidence, counterexample, and action records.
ChallengeCounterexample routehttps://mianzhang.org/counterexamples/Attack overbroad claims through public routes.
ArchiveZenodo portfolio indexhttps://zenodo.org/records/20027295Long-term archive index; cite specific DOI records where available.
ConceptSelf-Certification Grounding Checkhttps://mianzhang.org/concepts/self-certification-grounding-check.htmlGrounding fields for AI self-report.

Boundary

What this page does not prove.

  • This page does not claim RAG is useless.
  • It does not claim to solve all hallucination.
  • It does not turn private logs into public proof.
FAQ

Does RAG solve hallucination?

No. It can reduce some errors, but grounding still requires claim-to-evidence checks.

FAQ

What is AI self-certification?

The system reports its own state or basis; public reliability still requires external evidence.

FAQ

What should happen when a source is weak?

Downgrade the claim, mark it pending, or route it to review.

Guide

AI Self-Certification Grounding

Open route

中文

AI 为什么不能只靠自信回答

Open route