A county auditor asks whether a vendor chatbot with no retrieval layer can still hallucinate ordinance numbers. Which fundamentals insight is most accurate?
Select an answer to reveal the explanation.
Short Explanation
With no retrieval layer, the chatbot can still invent ordinance numbers that sound official. You need grounding to keep it from making up citations.
Full Explanation
A county auditor asking whether a vendor chatbot with no retrieval layer can still hallucinate ordinance numbers should conclude that ungrounded generative models can confidently invent ordinance citations that look plausible but are wrong. Without retrieval over a maintained code corpus, the model relies on parametric patterns and may fabricate section numbers, effective dates, or requirements that sound official. That fundamentals insight drives why auditors push for grounding controls on civic legal-tech chatbots.
The claim that without retrieval generative models cannot produce any text about ordinances is false; they can produce fluent, incorrect text freely. The claim that hallucinations only occur in image GANs and never in text chatbots ignores well-known text hallucination risk. The claim that chatbots without retrieval are immune to factual errors on local municipal codes is the exact false assurance auditors must reject.
Exam caveat: no-retrieval civic chatbots remain hallucination-capable; claims that they cannot produce text or are immune to local-code errors are distractors.
Operational check: require retrieval over current ordinance corpora for any resident- or staff-facing legal FAQ, display citations, add human legal review for high-impact answers, and treat invented ordinance numbers in UAT as blocking defects before go-live.