Unverified50% confidenceFactExact time
Large language models that have undergone safety alignment training often exhibit over-refusal behavior, erroneously refusing instructions that are themselves harmless but superficially touch on safety-related topics.
1
Sources
50%
Confidence
Long-term
Relevance
9/23/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedNegative prompting is often ineffective for language models, because negated concepts actually activate the model's attention.75% similarUnverifiedCurrent mainstream large language models are generally overconfident, partly because the training objective optimizes for the fluency of predicting the next word rather than the actual reliability of the output.73% similarUnverifiedRecursive language models fundamentally eliminate the failure mode of cheating by using information outside the subtask through context isolation, and can only learn truly recursive rules.73% similarUnverified大语言模型在生成摘要时天然倾向于反映评论中的多数意见，可能将低频的安全警告边缘化72% similarUnverifiedCross-lingual jailbreaking exploits the asymmetry between a model's concept space and its language space, as the learning of safety-refusal behavior is highly language-specific.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/942930API
curl https://kongchang.com/api/v1/knowledge/claims/942930MCP
get_claim(id=942930)