Unverified50% confidenceFactExact time
The study finds that on certain models, prompt perturbation can actually mitigate bias and hallucination, which contradicts the assumptions of some prior research.
1
Sources
50%
Confidence
Long-term
Relevance
9/30/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedResearch indicates that hallucination rates do not have a linear negative correlation with model scale, and larger models sometimes produce harder-to-detect misinformation in a more fluent manner.76% similarUnverifiedResearch shows that hallucination rate is negatively correlated with the knowledge density of the prompt; the richer and more precise the context provided in the prompt, the lower the probability that the model will generate fabricated content.75% similarUnverifiedDenser relevance judgments reduce the probability that correct results retrieved by a model are mistakenly judged as irrelevant, thereby improving evaluation reliability.74% similarUnverifiedPrevious research has mostly attributed over-refusal to static representation overlap, where benign prompts and harmful prompts are too close to each other in the model's internal representation space.74% similarUnverifiedBy injecting large amounts of domain expertise into the model, fine-tuning can significantly reduce the hallucination rate in specific scenarios73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/961152API
curl https://kongchang.com/api/v1/knowledge/claims/961152MCP
get_claim(id=961152)