Citation

Reducing Hallucinations in AI-Generated Wiki Content - Footnote 26

partial85% confidence

1 evidence check

Last checked: 4/3/2026

The AI Index Report is mentioned, but the year is not specified in the source. The source does not mention RLAIF (Reinforcement Learning from AI Feedback) or DPO (Direct Preference Optimization).

Evidence — 1 source, 1 check

www.voiceflow.com/blog/prevent-llm-hallucinations(1 check)

partial85%Haiku 4.5 · 4/3/2026

Found: **Reinforcement Learning from Human Feedback (RLHF)** trains models to prefer outputs that human reviewers label as correct. Research shows RLHF can reduce factual errors by 40% (GPT-4) and harmful ha…

Note: The AI Index Report is mentioned, but the year is not specified in the source. The source does not mention RLAIF (Reinforcement Learning from AI Feedback) or DPO (Direct Preference Optimization).

Debug info

Record type: citation

Record ID: page:reducing-hallucinations:fn26