評論
打印
Books & arts
來源于《圖書與藝術》版塊
Johnson
約翰遜語言專欄
Don’t fear the Writernator
不要害怕機器寫手
Computer-generated writing will never replace the human kind
計算機生成的文字永遠不會取代人類
Many people will be familiar with automated writing through two features of Gmail. Smart Reply proffers brief answers to routine emails. If someone asks “Do you want to meet at 3pm?”, Gmail offers one-click responses such as “Sure!” More strikingly, Smart Compose kicks in as you write, suggesting endings to your sentences. Both are not only rendered in flawless English; they often eerily seem to have guessed what you want to say. If someone sends bad news, Smart Reply might offer “Ugh.”
許多人將通過Gmail的兩個特性來熟悉自動寫作。Smart Reply 會給常規郵件提供簡短的回復。如果有人問“你想下午三點見面嗎?”，Gmail會提供一鍵式的回答，如“當然!”更引人注目的是，當你寫作時，Smart Compose會發揮作用，為句子提供結尾。這兩個功能不僅用完美的英語呈現；還常常奇怪地似乎猜到了你想說什么。如果某人發來不好的消息，Smart Reply可能會說“呃······”
The New Yorker’s John Seabrook recently described a more powerful version of this technology, called GPT-2, which can ably mimic his magazine’s style. Such systems use a digital network of billions of artificial “neurons” with virtual “synapses”—the connections between neurons—that strengthen as the network “learns”, in this case from 40 gigabytes-worth of online writing. The version Mr Seabrook tested was refined with back-issues of the New Yorker.
《紐約客》的（特約撰稿人）約翰•西布魯克最近描述了這種技術的一個更強大的版本——GPT-2，可以巧妙地模仿撰稿風格。這些系統使用了一個由數十億人造“神經元”和虛擬“突觸”（神經元之間的聯系）組成的數字網絡，突觸隨網絡的“學習”而增強，這可以在相當于40千兆的網絡寫作中找到例子。西布魯克先生測試的版本在《紐約客》的后刊中進行了修改。
The metaphor of the brain is tempting, but “neurons” and “synapses” deserve those scare-quotes. The system is merely making some—admittedly very sophisticated—statistical guesses about which words follow which in a New Yorker-style sentence. At a simple level, imagine beginning an email with “Happy…” Having looked at millions of other emails, Gmail can plausibly guess that the next word will be “birthday”. GPT-2 makes predictions of the same sort.
大腦這個比喻聽起來很誘人，但是“神經元”和“突觸”值得我們引用一下。該系統只是對該雜志風格的句子中哪個詞跟哪個詞做了一些統計上的猜測，誠然，這些猜測非常復雜。簡單地說，想象一下以“Happy…”開頭的郵件。查看了數以百萬計的其他電子郵件后，Gmail似乎可以合理地猜測下一個單詞將是“birthday”。GPT-2做出了類似的預測。
譯文由可可原創，僅供學習交流使用，未經許可請勿轉載。
