AI 与写作2017techcrunch.com
谷歌资助开发自动撰写本地新闻的软件
Google is funding the creation of software that writes local news stories
导读
2017 年 TechCrunch 报道 Google 资助开发自动撰写本地新闻的软件，HN 评论围绕 AI 生成内容是否可读、是否有意义展开争论。对写作者而言，这是一次反向提醒：什么样的文字才算“有意义”——细节要服务于理解，叙事要基于经验，信息要能帮读者筛选而非增加噪音。
可以直接用的写作要点
- 01
精确数字和细节本身不构成意义；每个细节都要让读者读懂，否则只是“古怪的具体”。
- 02
把原始数据或信息转成有观点、有张力的叙事，才是写作的价值所在（如里根把电报纸条讲成现场解说）。
- 03
意义来自读者的切身经验，写作要调用具体情境与感受，而不是只拼接词句模式。
- 04
篇幅和详略可按读者兴趣伸缩：先想清楚读者需要什么，再决定展开还是压缩。
HN 评论精选
AI 文本的典型破绽是堆砌精确数字却无可读意义，读者需要标注与过滤。
我在研究一只股票时读到这篇文章，读着读着一直觉得：这不是人写的，是软件写的。我查了署名，上面有个人名。当然，谁都能写点东西发出来，这大概也只是又一个例子。但那种古怪的文风，既不像“按我的思路写”，也不像赶时间没编辑就发。比如第二段：“The corresponding low also paints a picture and suggests that the low is nothing but a 97.89% since 11/14/16.” 我完全无法从中读出任何意义，可它又带着古怪的具体细节。我不乐见这种趋势，也不乐见 Google 走上这条路。我想这不可避免，除非这套 AI 内置了专业能力，能替我从未数据中提炼意义，并呈现得比我更有洞见、更有趣，否则它只会成为又一种我需要过滤的谷壳。至少，能不能请把 AI 生成的文字标注出来？
11thEarlOfMar查看英文原文
I ran across this article while researching a stock and as I read, I kept thinking, "This was not written by a person. This was written by software." [0] I checked the attribution, and there is a person's name on it. Sure, any hack can write and publish and this is probably just another example. But the odd style doesn't even strike me as 'writing the way I think' or writing and publishing quickly without editing. For example, from the 2nd paragraph, "The corresponding low also paints a picture and suggests that the low is nothing but a 97.89% since 11/14/16." I can't gather any meaning from that statement, yet it has oddly specific details. I am not glad to see this trend and not glad that Google is embarking on this path. I suppose it is inevitable, but unless there is expertise built into this AI that can extract meaning from data on my behalf and present it in a way that is more insightful and interesting than I am, it will become yet another source of chaff I'll have to filter. Can we at least, please, flag AI generated prose as such? [0] https://www.nystocknews.com/2017/07/05/tesla-inc-tsla-showcases-what-it-has-to-offer-via-the-technical-chart/
语义来自具身经验，AI 只有字符串模式，写不出真正有意义的新闻。
我非常怀疑当前这一代 AI 能写出像样的新闻，哪怕是最“容易”的题材。AI 也许能产出那种自相矛盾、本质上无意义的词串，但连那种东西背后也缺少统一的意图。特朗普至少还能一贯地调动某种原始的 #MAGA 情绪，我怀疑 AI 连这都做不到。如果 AI 真能写出像样的新闻、且陈述始终有意义，那对语言学理论会有巨大影响。目前我们无法在计算机中表示抽象的语义；比如我说“正义”，你们都懂，因为你们在个人生活中有过“正义”（或其反面，如霸凌）的切身经验。AI 从未接触过这种具身的、经验性的输入，它只能接触人类产出的字符串模式。既然输入更少，凭什么指望它产出和我们一样的输出？有人说新闻不是小说，所需理解门槛更低——我不这么认为。我们往往低估了哪怕最基础的语言使用中的具身性。
igammarays查看英文原文
Color me extremely skeptical that current-generation AI can ever write decent quality news articles, even on the "easiest" subjects (i.e. non-emotional topics that may be most amenable to computerization). Sure, an AI might be able to produce the type of contradictory and fundamentally meaningless strings-of-words that characterize Trumpian speech, but even that would lack a unified agenda behind it. Even Trump seems to appeal to some raw masculine #MAGA emotion pretty consistently, but I doubt an AI would be able to do even that. If an AI could produce decent news articles with consistently meaningful statements, I think it would have huge implications for linguistic theory. Currently, we have no way of representing abstract semantic meaning in a computer, for example when I say the word "justice", you all know what I'm talking about because you all have embodied experiences of "justice" (or its opposite) in your personal lives (e.g. bullying). An AI simply never had access to this kind of embodied, experiential input. It only has access to patterns of strings that we humans produce. And so why should we expect an AI to be able to ever produce the same output that we produce when it has access to less input? Sure, one might argue that news articles are not novels, and so require a lower threshold of understanding to produce. I don't think so. We tend to underestimate the embodied nature of even the most basic use of language[0]. [0] See The Embodied Mind by Varela, Thompson, and Rosch.
新闻写作的核心是把原始信息转成有张力的叙事，且篇幅可随读者兴趣伸缩。
讽刺的是，这正是广播新闻的起点。里根当“体育解说员”时，实际是读比赛进程的电报纸条（“史密斯上场。第一球好球未挥。第二球安打”），再编出一个激动人心的故事配上：“史密斯走向本垒。他绷紧肌肉，踢了踢土，摆好姿势。第一球他没挥……好球！下一球来了……他挥棒……球扎实地飞向三垒。是界外吗？不！！他安全上一垒！！！”其实大多数“新闻”文章本来也就几段长，完全可以按读者的兴趣当场扩写或压缩。
gumby查看英文原文
The irony is that the I how radio news got its start. Ronald Reagan became an actor after being a "sports announcer" -- what he really did was read the ticker tape of a game in progress ("smith at plate. first ball strike no swing. Second ball base hit") and create an exciting story to go with it "And smith steps up to the plate. He flexes his muscles, kicks the ground and takes his stance. He passes on the first ball....strike! Here's the next pitch...he swings...solid towards third base. Is it a foul? NO!! AND HE'S SAFE ON FIRST BASE!!!" Really most "news" articles are only a couple of paragraphs long anyway and could be expanded or contracted on the spot to match the interest of the reader.
低质自动内容会稀释信息环境，写作者更应提供真正的洞见。
我强烈希望机器人写的新闻不存在，不出现在我的搜索结果里，也不出现在我读的任何信息流里。它是污染，让真实的信息和洞见更难被找到。真有人喜欢这种东西吗？
jimrandomh查看英文原文
I would strongly prefer that robo-written news not exist, not appear in the results of any searches I make, and not appear in any feed that I read. It is pollution that makes real information and insight harder to find. Does anyone actually like this stuff?
同类还值得读
我是肯尼亚人：不是我把 ChatGPT 写成了我，而是它写起来像我
I'm Kenyan. I don't write like ChatGPT, ChatGPT writes like me
一位肯尼亚作者反驳“我写得太像 AI”的指责：不是他写得像 ChatGPT，而是 ChatGPT 写得像他——那种平滑、无地域色彩、貌似中立的英语，本来就是殖民教育与精英阶层塑造…
AI 与写作2025 · 799 分 · 504 评4 条写作要点编辑精选使用 AI 助手完成论文写作任务会累积认知债务
Accumulation of cognitive debt when using an AI assistant for essay writing task
这篇论文研究的是：在完成议论文写作任务时使用 AI 助手，是否会让写作者积累"认知债"——也就是把本应由自己承担的思考外包出去，短期省力、长期削弱写作与思考能力。
AI 与写作2025 · 391 分 · 426 评4 条写作要点编辑精选学生使用 ChatGPT 写作指南
A Student's Guide to Writing with ChatGPT
这是 OpenAI 面向学生的一份写作指南，讲的是如何把 ChatGPT 用作写作过程中的辅助工具，而不是替你写完作业的"枪手"。
AI 与写作2024 · 323 分 · 309 评4 条写作要点编辑精选Grammarly，你也背叛了我？
Et Tu, Grammarly?
本次输入里只给了标题和链接，正文和评论都是空的，所以没法据实写出导读。
AI 与写作2025 · 279 分 · 74 评编辑精选用 GPT-3 进行创意小说写作
GPT-3 Creative Fiction
Gwern 的长文系统考察 GPT-3 在创意小说写作上的能力与边界：它能维持人设、写出讽刺腔调，也会整段复述训练数据里的歌词。
AI 与写作2020 · 234 分 · 97 评4 条写作要点编辑精选