Unverified50% confidenceSolutionExact time
The compression method involves only removing the multilingual portions of the model while retaining full English capabilities, rather than conventional brute-force quantization.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
Unverified通用大模型在多层嵌套从句、修辞性倒装等复杂阿拉伯语场景中准确率显著下降64% 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.62% similarUnverified通用大模型在特定语言的方言处理、形态变化丰富的语言以及文化特定表达上仍存在明显不足61% similarUnverified大型语言模型在无约束空间中会倾向于创造性发散，而非工程所需的收敛执行61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/909846API
curl https://kongchang.com/api/v1/knowledge/claims/909846MCP
get_claim(id=909846)