大型語言模型的工作原理是將你的話語分割成稱為「詞元」（tokens）的小塊，然後利用統計方法分析這些詞元，從而得到適當的回應。這代表你說的每一個字詞，甚至是一個額外的逗號，都可能影響AI的回答。問題在於，這種影響幾乎無法預測。雖然已經有許多研究試圖從AI提示的細微變化中尋找規律，但大部分證據相互矛盾，結論也不明確。
Крупнейшая нефтяная компания мира задумалась об альтернативе для морских перевозок нефти14:56
。体育直播对此有专业解读
Студенты нашли останки викингов в яме для наказаний14:52
Open-source projects that rely on volunteer contributors
We see a divide in society between people who want AI to do impressive things with their photos and videos, and those who don't want AI to do anything with photos and videos because it's eroding our ability to believe that what we have seen is real, destroying the concept of photographic evidence.