Unverified50% confidenceFactExact time
In LLMs, the influence of human-like factors such as commonness, distinctiveness and characteristicness is transmitted to model ratings indirectly through 'factual truth' judgments.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/21/2026
First Seen
Valid until: 1/19/2027
Sources
Related Entities
Related Claims
UnverifiedThe study suggests that models rely on distributional familiarity rather than genuine fact-verification mechanisms to judge the truth or falsity of statements.77% similarUnverifiedUnlike humans, LLMs' acceptance judgments of generic statements are not influenced by 'strikingness'.76% similarUnverifiedFor LLMs, whether a statement is factually true is the most important factor influencing their acceptance of generic statements.75% similarUnverifiedIn the NHANES data application, empirical conclusions change when population heterogeneity and error heteroscedasticity are properly accounted for75% similarUnverifiedThe alignment between the LLM reviewer and human expert evaluations exhibits a characteristic of being 'valuable but strongly metric-dependent,' with agreement on some dimensions but noticeably weaker alignment on others.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1079537API
curl https://kongchang.com/api/v1/knowledge/claims/1079537MCP
get_claim(id=1079537)