Unverified50% confidenceFactExact time
By family, the four prompt families with grammatical conditions had the highest hallucination share, reaching 38% to 42%; the family using only IPIP personas was lowest, at 28%.
1
Sources
50%
Confidence
Long-term
Relevance
10/27/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedOn the cluster panel, replacing the model's structured decision-making with code (based on the same extracted facts) improved family accuracy by an average of 20.5 percentage points, with Llama 3.1 8B rising from 1% under self-judgment to 56% under code adjudication.62% similarUnverifiedQwen3-32B with thinking mode disabled approved all eight applications in the opening technical writing job family (including four that did not meet the experience requirement), with only 39% of families having all eight correct and 41% of unqualified applicants incorrectly approved.60% similarUnverifiedThe gap between selection and generation exceeds 50 percentage points, revealing that reliability is far lower than benchmarks suggest when users need the model to directly write out kinship terms.59% similarUnverifiedThe class collapse phenomenon of scorers also appeared with a 7B scorer from a fourth family, new harmful requests, and target models from a third family.59% similarUnverifiedThe robust policy fidelity metric is measured by family-level pass rate, so a score of 50% means half of the families passed all checks, not that half of the individual answers were correct.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1102118API
curl https://kongchang.com/api/v1/knowledge/claims/1102118MCP
get_claim(id=1102118)