Unverified50% confidenceCautionExact time
The training corpora of mainstream large models are predominantly English-language and heavily weighted toward U.S. legal content, making models prone to applying U.S. legal frameworks and giving incorrect answers when responding to questions about Australia-specific employment regulations.
1
Sources
50%
Confidence
Long-term
Relevance
9/1/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe training data of current mainstream large models contains a relatively high proportion of English corpus, which may cause the model to favor English prosodic patterns in syllable boundary processing and pause rhythm, producing a foreign accent67% similarUnverifiedEnglish content in mainstream large language model training corpora typically exceeds 50%, leading to significant performance gaps for non-English languages.63% similarUnverifiedThe rise of large language models has exposed traditionally high-skill white-collar tasks such as legal document drafting, financial analysis, and code generation to automation pressure.61% similarUnverifiedThe safety alignment of many models degrades significantly in non-English languages, and low-resource languages are especially easy to exploit because the human annotation data for RLHF alignment training is highly concentrated in English.61% similarUnverifiedPerez et al.的《Discovering Language Model Behaviors with Model-Written Evaluations》证实经过RLHF训练的模型普遍存在谄媚倾向61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/837369API
curl https://kongchang.com/api/v1/knowledge/claims/837369MCP
get_claim(id=837369)