Is Your AI Safer in English Than in Spanish? I Ran the Experiment
中文摘要
探讨AI在英语和西班牙语中的安全防御一致性
English Summary
Research shows large language models often provide inconsistent safety responses across different languages, suggesting AI may be less secure or more susceptible to exploitation in non-English languages.
Original Excerpt
Most people assume that when a large language model refuses to help with something harmful, it does so consistently, regardless of the… Continue reading on Medium »