Give 'em HALI'm writing this part on February 22. Two days prior, a project called "Pi for Excel: AI sidebar add-in for Excel" released and got good traction on Hacker News. As I noted in the XPER column, our current "AI" boom is the biggest, but not the first. English language interactions, first by keyboard and fingers-crossed-one-day-by-voice-if-AI-technology-continues-along-our-projected-path-of-wishes-and-dreams, were available as add-ins to various programs.
撰文｜不慌实验室，主笔｜钱晶，策划｜陈肖冉
，更多细节参见有道翻译
乘缆车观云海 诗仙笔下瀑布再现仙境
«Мы умрем как мученики»Жители Ирана и Израиля — о ежедневных бомбежках, разрушенных городах и страхе перед большой войной17 июня 2025
The raw plot already tells a story. All six pairs start at moderate similarity (0.4–0.6 at the embedding layer), then rapidly converge to near-1.0 by about layer 5. Through the mid-stack, they stay high but begin to separate. The blue line (EN poem ↔ ZH poem: different language, same content) stays above the red line (EN fact ↔ EN poem: same language, different content) from roughly layer 15 onward. The model’s internal representation cares more about what you’re saying than what language you’re saying it in.