Unverified50% confidenceFactExact time
Controlled deletion experiments show that removing the relevant analysis from a report lowers the score of the corresponding dimension.
1
Sources
50%
Confidence
Long-term
Relevance
10/28/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn one task continuation experiment, despite the reduced context, the output quality as assessed by human evaluation was actually lower.75% similarUnverifiedThe system has a Critic review step that screens and removes low-value content before comments are published, and has discarded approximately one-quarter of drafted comments to date.72% similarUnverifiedAfter decomposing the taxonomy into a series of yes/no questions, the accuracy of all models decreased.71% similarUnverifiedExperiments show that models processed with abliteration exhibit almost no degradation on standard capability evaluations, but the probability of refusing harmful requests drops significantly.70% similarUnverifiedThe difference in effect between simple deletion and information replacement is enormous: the drop from 0.73 to 0.52 is roughly equivalent to the model's capability falling by a tier, enough to degrade a usable enterprise-grade application into an unusable one.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1106111API
curl https://kongchang.com/api/v1/knowledge/claims/1106111MCP
get_claim(id=1106111)