作者：Issie Lapowsky关于写作的痛苦，人们已经写过太多。乔治·奥威尔曾将其比作“一场漫长而痛苦的疾病”。已故体育记者雷德·史密斯也曾说，他的专栏文章是“一滴滴鲜血凝结而成”。尽管如此，还是有一些“受虐狂”义无反顾地选择了写作这条路。
By Issie Lapowsky Plenty has been written about the agony of writing. George Orwell once likened it to “a long bout of some painful illness.”The late sportswriter Red Smith similarly told that his column “comes out with little drops of blood.”Despite this pain, there are people sick enough to sign up for the writer’s life.
维沙尔·苏德（Vishal Sood）并不在其中。苏德是一位受过专业训练的计算机科学家，他在印度出生长大，曾在微软公司从事了近20年的产品开发工作。后来，他担任了Typeface公司的研发主管，该公司专门为营销人员开发人工智能工具。在近30年的职业生涯中，苏德积累了丰富的专业知识，并产生了分享这些知识的冲动。
Vishal Sood isn’t one of them. A computer scientist by training, Sood was born and raised in India and spent almost 20 years building products at Microsoft Corp. He went on to become head of research and development for Typeface Inc., a company that makes artificial intelligence tools for marketers. Over the course of his almost 30-year career, Sood developed plenty of expertise and an impulse to share it.
然而，写作——尤其是用他的第二语言英语写作——对他来说始终是一道巨大的障碍。况且，他说：“那又不是我的本职工作。”但人工智能写作工具的出现改变了一切。现在，苏德可以利用他与生俱来的技能来弥补自己的短板。他利用Claude Code构建了一个他称之为“AI大脑”的系统，并向其中输入了数百份他自己呕心沥血撰写的文档和文章。
But the act of writing — in his second language of English, no less — always felt like the biggest barrier. Besides, he says, “that is not my day job.”But the arrival of AI writing tools changed things. Now, Sood could use the skills that came naturally to him to bolster the skills that didn’t. He built an “AI brain,” as he calls it, with Claude Code and fed it hundreds of documents and posts he’d painstakingly written himself.
他还制定了一套100分的评分标准，根据系统生成的内容与他个人观点和语气的契合度来对其进行评估。
He created a 100-point rubric to grade the system’s writing on the basis of how well it tracks with his views and voice.
现在，苏德会口述自己的想法大纲，由AI笔记工具记录下来。他将转录稿输入到这个“大脑”中，然后根据自己的喜好修改输出内容，用于内部沟通和他在Substack上发布的公开通讯。整个过程可能需要几个小时，但手头有了初稿，就减轻了面对空白页面时的焦虑。
Sood now dictates outlines of his thoughts out loud to an AI notetaker. He feeds the transcript into the “brain,” then edits the output to his liking for both internal communications and his public Substack newsletter. The whole process can take hours, but having a draft in hand helps relieve the anxiety of facing an empty page.
这些文章读起来像AI写的吗？当然像。但这算粗制滥造吗？在苏德看来并非如此。“只要我能传达出我的想法——而且你无法仅仅通过在Claude中输入一个提示词就得到同样的想法——它就有价值，”他说。他想要的只是表达观点，而不是赢得普利策奖。
Does most of this writing read like AI? You bet. Does that make it slop? Not as far as Sood is concerned. “As long as I’m getting an idea across — and you can’t just type in a prompt in Claude and get the same idea — there’s value to it,” he says. All he wants is to convey a point, not win a Pulitzer.
关于人工智能写作的争论由来已久，如今正愈演愈烈。小说家们因涉嫌使用人工智能而痛失出版合同；Anthropic公司正在为其人工智能Claude生成的文字添加水印；Pangram Labs公司则通过揭露人工智能使用者建立起了一项蒸蒸日上的业务。最近，亿万富翁投资者斯坦利·德鲁肯米勒（Stanley Druckenmiller）发表了一篇专栏文章（事后证明是由其人工智能助手代笔），这引发了舆论对编辑们的如潮批评，而这些编辑在刊发文章后仍公然坚持自己的决定。
The long-simmering debate over AI writing is reaching a boiling point. Novelists are losing book deals for suspected AI use. Anthropic PBC is watermarking the words Claude generates. Pangram Labs Inc. has built a thriving business exposing AI users. And a recent op-ed written by billionaire investor Stanley Druckenmiller (and, it turns out, his AI assistant) resulted in an avalanche of criticism directed toward the editors who let it run and then defiantly stood by that decision.
这场争议大多集中在媒体和出版行业，因为在这些行业中，写作本身就是产品。这在情理之中。对于那些靠“折磨”自己来写作并以此为生的人来说，将工作外包给机器不仅是对技艺的放弃，更是对这一职业本身的威胁。越来越多的证据（由研究人员记录，也由我们亲眼所见）表明，人工智能使写作变得平庸，使其更加同质化且索然无味。
Much of this controversy has centered on media and publishing, industries where writing is the product. And that’s to be expected. For people who get paid to torture themselves by writing, offloading that work to a machine feels not only like an abdication of the craft but also like a threat to the profession itself. And there’s mounting evidence (documented by researchers and, you know, one’s own two eyes) that AI flattens writing, making it more homogeneous and bland.
但对于更广泛的企业界人士而言，是否使用人工智能进行写作的问题并没有那么深奥。在许多工作场所，写作只是一种营销产品、安抚不满客户、向投资者通报情况或在团队间分享进度更新的手段。许多人厌恶写作，而且与奥威尔或史密斯不同，他们并不觉得写作有什么光荣之处。他们不禁会想，如果有一个工具能帮他们完成这些任务，而无需经历那种“呕心沥血”的痛苦，那他们为什么要拒绝呢？
But for people in the wider corporate world, the question of whether to use AI to write is simply not that deep. In many workplaces, writing is a way to market a product, placate a disgruntled customer, update investors or share status updates among teams. Many people abhor it and, unlike Orwell or Smith, find no glory within. If a tool can help check that item off the list without all the bloodletting and rending of garments, they wonder, why exactly should they resist?
诺拉·罗杰斯（Nora Rodgers）半开玩笑地说，她创业的原因之一是她讨厌写求职信。“我讨厌写求职信。整个过程，把它发出去，然后让别人看到我的错误和弱点，”她说，“我患有严重的阅读障碍，这个问题一直伴随着我进入成年。”2019年，罗杰斯创办了一家公司，让她尽可能远离写作：她在费城开了一家名为Wag Days LLC的狗狗日托中心。
Nora Rodgers says, only half-joking, that one reason she started her business was because she disliked writing cover letters. “I hated it. The whole process, sending it out and someone else seeing my mistakes and weaknesses,” she says. “I’m extremely dyslexic, and it followed me into adulthood.”In 2019, Rodgers started a company to take her about as far from writing as possible: a dog daycare in Philadelphia called Wag Days LLC.
现在，她每天都要同时照看几十只狗，并管理员工的各种需求和排班，公司目前已有两家分店。她需要逐项列出开支、面试新员工，而且不得不提的是，她还得清理狗屎。
Now she spends her days wrangling as many as several dozen dogs at a time and managing the various needs and shifts of her employees, now in two locations. There are expenses to itemize, new hires to interview and, it must be said, there’s poop to pick up.
当她确实有时间写作时——比如写营销邮件或给客户的信——通常是在下班后。罗杰斯说，拥有一种能阅读她的初稿、纠正语法、修正拼写并使她的信息更简洁的工具，简直是“天赐之物”。她对人工智能可能改变她的语气持谨慎态度，但除了她母亲之外，这项技术是她预算范围内唯一的编辑。“在小企业里，我还雇不起专门做这件事的人，”她说。
When she does find time to write — marketing emails, say, or letters to clients — it’s usually after hours. Having a tool that can read her first drafts, correct her grammar, fix her spelling and make her messages a bit cleaner has been “a godsend,” Rodgers says. She’s wary about AI voice creeping into her tone, but, aside from her mother, the technology is the only editor in her price range. “In a small business, I can’t hire people to do this specific thing yet,” she says.
马特·布隆伯格（Matt Blumberg）使用人工智能写作并要求团队也这样做，有着更具动力的原因。他的公司Markup AI旨在帮助其他企业评估其内容并进行润色，以确保其符合品牌形象。Markup的每位员工都被要求将他们发布的任何内容通过公司自己的技术进行处理，部分原因是为了测试该技术的运行效果。
Matt Blumberg has more motivated reasons for using AI to write and requiring his team to do it too. His company, Markup AI, exists to help other businesses assess their content and spruce it up to ensure it stays on brand. Every employee at Markup is expected to run whatever they publish through the company’s own technology, in part as a way to test how it’s working.
但布隆伯格坚持认为，将写作外包还有其他优势。他也建立了一个类似人工智能大脑的东西，他称之为“马特机器人”（MattBot）。正如他在最近的一篇博客文章中所记录的那样，他曾用它为那些试图在大学创业课上采访他的好奇学生起草回答，并将公共关系的谈话要点转化为领英（LinkedIn）帖子。
But Blumberg insists there are other advantages to outsourcing your prose. He, too, has built a sort of AI brain, which he calls MattBot. As he chronicled in a recent blog post, he’s used it to draft answers for inquisitive kids trying to interview him for their college entrepreneurship classes and to turn public-relations talking points into LinkedIn posts.
这并不是什么新奇或高尚的事情——世界真的需要另一篇领英（LinkedIn）帖子吗？——但这正是现代首席执行官日常工作中琐碎事务的一种，而现在他完成这些工作所需的时间更少了。“我不会用自动化来处理与员工的谈话，不会用自动化来处理销售电话，也不会用自动化来处理投资者会议，”布隆伯格（Blumberg）说，“但它为我节省的每一分钟都是宝贵的。”
It isn’t particularly novel or noble — does the world really need another LinkedIn post? — but it’s the sort of minutia that occupies a modern chief executive officer’s days, and now it takes him less time to do it. “I don’t automate talking to an employee. I don’t automate a sales call. I don’t automate an investor meeting,” Blumberg says. “But any minute that it creates for me is a good minute.”
然而，围绕人工智能写作产生的文化冲突，使得商业领袖和普通员工都难以遵循这些快速演变的规则。在一家公司能让你获得晋升和赞誉的行为，在另一家公司却可能让你被解雇或受到惩罚。通常，员工只能自己去摸索他们应该站在哪一边。
Yet the culture clash over AI writing is making the fast-evolving rules hard to follow for business leaders and rank-and-file workers alike. What could get you promoted and praised at one company can just as easily get you fired or punished at another. Often it’s up to workers to suss out which side of the line they’re supposed to be on.
当凯蒂·戴维斯（Katie Davies）在上一份工作中开始使用微软Copilot撰写自我评估时，她承认自己并不确定这是否被允许。“我感觉自己像是在作弊，”她说。直到人力资源部门专门推出了一款用于撰写评估的定制人工智能代理，她才意识到使用人工智能不仅是被允许的，甚至是受到鼓励的。
When Katie Davies began using Microsoft Copilot to write self-evaluations at her last job, she admits she wasn’t sure whether it was allowed. “I felt like I was cheating,” she says. It wasn’t until the human resources department rolled out a custom AI agent specifically for writing evaluations that she realized using AI wasn’t just allowed; it was encouraged.
现在，戴维斯正在寻找新工作（为了遵守离职协议的条款，她要求对前雇主的名称保密），她并没有向潜在雇主隐瞒自己使用人工智能的情况。“他们想知道员工是否懂得如何使用人工智能，”她说，“这是我被期望具备的一项技能。”尽管如此，仍有许多雇主——甚至包括那些在人工智能行业工作的雇主——正因为其他人拥抱人工智能的同样原因而对其敬而远之：为了提高速度。
Now, Davies is on the job market (she asked to keep her former employer’s name private to satisfy the terms of a severance agreement), and she’s not hiding her AI use from potential employers. “They want to know that people know how to use AI,” she says. “It’s a skill set I’m expected to have.”Still, there are plenty of employers — even those working in the AI industry — who are turning away from AI writing for precisely the same reason others are embracing it: to speed things up.
在帮助律师进行合同审查的人工智能公司Spellbook，首席执行官斯科特·史蒂文森（Scott Stevenson）厌倦了筛选那些充斥着人工智能特征、却始终无法切中要点的多页备忘录。对他的员工来说，用这种方式写作可能更快，但他阅读起来却要花费大量时间。“这只会让我们积压工作，”史蒂文森说。
At Spellbook, an AI company that helps lawyers with contract reviews, CEO Scott Stevenson was sick of sifting through multipage memos, riddled with AI tics, that never seemed to actually land on a point. It may have been faster for his employees to write this way, but it was taking him forever to read. “It just made us get backlogged,” Stevenson says.
更糟糕的是，他担心有些人通过提交长篇大论、辞藻华丽的战略文档来掩盖他们尚未成熟的想法。“许多初级员工认为好的写作就是‘如何写出润色精良、结构严谨且篇幅冗长的内容？’”他说，“而我认为好的写作是‘如何用最少的字词传达最丰富的含义？’”本月初，史蒂文森向他旗下的200多名员工发送了一份备忘录，制定了关于人工智能写作的新规则，其中包括禁止使用人工智能生成客户沟通内容和项目建议书。
Worse, he worried some people were shipping long, glossy strategy documents as a way of masking their underbaked ideas. “A lot of junior people think good writing is, ‘How do I make something that’s polished and structured and long?’” he says. “I think good writing is, ‘How can you pack the most meaning into the fewest amount of words?’”Earlier this month, Stevenson sent a note to his more than 200 employees setting some new rules around AI writing, including a prohibition on AI-generated client communications and project proposals.
相反，他写道，他想要的是员工想法的“D+版本”。换句话说，就是他们自己独立思考出来的东西。“即使有点混乱也没关系，”他说，“这反而能帮助公司更快地推进工作。”
Instead, he wrote, he wanted the “D+ version” of people’s ideas. In other words, the ones they generated on their own. “Even if it’s a bit messy,” he says. “That just helps the company move much faster.”