Citation

Reducing Hallucinations in AI-Generated Wiki Content - Footnote 35

partial90% confidence

1 evidence check

Last checked: 4/3/2026

The claim states that k=20 chunks was optimal for balancing retrieval quality and hallucination reduction, but the source only mentions that it was the 'magic number' for the query bot. The claim also mentions iterative testing, but the source only mentions iterative testing in the context of prompt engineering.

Evidence — 1 source, 1 check

www.tredence.com/blog/mitigating-hallucination-in-large-language-models(1 check)

partial90%Haiku 4.5 · 4/3/2026

Found: Research on production query bots found k=20 chunks optimal for balancing retrieval quality and hallucination reduction after iterative testing.

Note: The claim states that k=20 chunks was optimal for balancing retrieval quality and hallucination reduction, but the source only mentions that it was the 'magic number' for the query bot. The claim also mentions iterative testing, but the source only mentions iterative testing in the context of prompt engineering.

Debug info

Record type: citation

Record ID: page:reducing-hallucinations:fn35