Unverified60% confidenceFactExact time
Research 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.
2
Sources
60%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
Unverified丰富的上下文信息能收窄模型输出概率分布并减少幻觉，这一现象被称为「知识锚定效应」83% similarVerifiedResearch 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.82% similarUnverified模型规模越大，幻觉的表述往往越流畅、越具迷惑性，识别难度反而上升79% similarUnverifiedReasoning capabilities can be compressed through knowledge distillation and reinforcement learning fine-tuning, but the storage of factual knowledge depends on parameter scale, and compression leads to reduced knowledge coverage and higher hallucination rates.78% similarUnverifiedHallucination and randomness are not in a simple causal relationship—even with fully deterministic decoding, a model may still fabricate content due to missing knowledge.78% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/563193API
curl https://kongchang.com/api/v1/knowledge/claims/563193MCP
get_claim(id=563193)