Unverified50% confidenceFactExact time
The refusal slice borrows from the practice of slice analysis in machine learning evaluation, specifically referring to the refusal accuracy and invalid completion rate on the subgroup of negative control tasks.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedModel abstention can avoid confident errors caused by compression distortion.70% similarUnverifiedA challengeable scoring mechanism allows agents or team supervisors to dispute specific scores to trigger manual review, which both prevents AI misjudgments from affecting performance evaluations and accumulates labeled data to reversely optimize the scoring model.69% similarUnverifiedThe mechanism by which an AI system refuses to answer is academically referred to as selective prediction or abstention.69% similarUnverifiedUncensoring primarily affects a model's tendency to refuse, rather than its fundamental reasoning and generation capabilities, so capability loss is not significant for alignment-insensitive tasks such as coding.69% similarUnverifiedModels labeled 'obliterated' are uncensored versions with refusal mechanisms removed; in testing they more easily miss certain requirements in prompts, and uncensored does not mean better.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1013948API
curl https://kongchang.com/api/v1/knowledge/claims/1013948MCP
get_claim(id=1013948)