Unverified50% confidenceOpinionExact time
The standard practice aggregates all heterogeneous harmful prompts into a single uniform mean vector, which may conflate distinct refusal-related phenomena.
1
Sources
50%
Confidence
Long-term
Relevance
10/22/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe weighted acceptance rate is calculated by assigning different weights to each alert according to its severity and then computing the acceptance proportion, rather than simply counting the ratio of confirmed alerts to total alerts.73% similarUnverifiedAverage accuracy masks distributional skew; a judge that gets all easy cases right and is overconfident on hard cases can post a high average score while failing at critical points.72% similarUnverifiedExperiments show that models processed with abliteration exhibit almost no degradation on standard capability evaluations, but the probability of refusing harmful requests drops significantly.71% similarUnverified语义等价的不同表述可能导致模型回答质量出现高达30%的波动，该现象被称为'提示敏感性'（prompt sensitivity）71% similarUnverified多重比较问题指同时测试多个模型变体时，即使每个单独检验显著性水平设为0.05，整体假阳性率也会急剧膨胀71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1084576API
curl https://kongchang.com/api/v1/knowledge/claims/1084576MCP
get_claim(id=1084576)