Sep. 22, 2025

• A Cure for The Eternal September
• Built on a Wall of Text
• A Disciplined Front Page and a Tireless Moderator
• Caveat Lector
• Appendix: Tips for Reading HN
• Subscribe to Filtered RSS Feeds
• Search for External URLs
• Skim Comments Strategically
• Use AI to Summarize Comments
• 「永恒的九月」有救吗？
• 文本墙里砌出的罕见人气
• 纪律严明的首页与鞠躬尽瘁的管理员
• 兼听则明
• 附：HN 阅读方法谈
• 订阅筛选版 RSS
• 主动搜索外部网址
• 跳读评论
• AI 总结评论

---

A Cure for The Eternal September

In early 1994, a group of frustrated users on Usenet, a precursor to modern forums, inadvertently coined a term: Eternal September.

The problem wasn’t the month of September itself, but the people who arrived with it. In its early days, Usenet had a relatively high barrier to entry, which helped maintain user quality and content standards. But every fall, a new wave of college freshmen would flood Usenet through their campus networks, posting haphazardly and ignoring established community norms, much to the annoyance of veteran users. Over the years, this September influx became a familiar, if unwelcome, ritual.

1994 was different. Starting the previous year, many consumer internet service providers began offering Usenet access. Suddenly, low-quality, off-topic posts from inexperienced users poured in year-round. The chaos of September had become eternal.

The phrase marks the end of the internet’s early-elite era and crystallizes a chronic dilemma for any online community: scale, topical breadth, and discussion quality form an unstable triad. The intersection of all three is, most days, a fantasy.

And yet one community has, across more than eighteen years, grown relentlessly in users and traffic while sustaining both interesting topics and a high bar of discussion: Hacker News (HN).

The frontpage of HN

Built on a Wall of Text

HN looks plain at first glance. It’s essentially a wall of text, where even most buttons are just text links. Newcomers might not even figure out how to post here. Unlike typical online forums, the vast majority of “posts” on HN are simply shared links. The contribution of the original poster (OP), if any, is limited to a title and perhaps a brief comment, with the ensuing discussion centered on the linked content.

A thread on HN In other words, HN is less a forum and more a collectively curated reading list, or more plainly, an external comment section for the rest of the internet. The technical term for this format is a “link aggregator.” Other well-known examples from the past include Digg and Reddit. But Digg (as it was) is long gone, and Reddit has gradually transitioned to a more mainstream forum model, leaving HN as a unique outlier.

Despite its spartan appearance and learning curve, HN boasts over ten million monthly visits (according to SimilarWeb data), outperforming popular tech news sites like TechCrunch and Engadget. In contrast to its massive traffic, the servers that run HN are surprisingly modest: just two machines with quad-core Intel Xeon E5-2637 v4 CPUs, running FreeBSD.

To understand why HN attracts so much attention, one must look at its history. You can tell from the domain, news.ycombinator.com, that HN did not start as an independent site but as a side project of Y Combinator, a renowned venture capital firm. In February 2007, Paul Graham, then president of YC, launched the site. The stated goals were prosaic and personal: publicly, to create a place for the startup community (hence the original name, Startup News), and privately, to scratch a programming itch by building a site in Arc, a Lisp dialect he co-created.

Given that lineage, HN quickly became a hub for startup founders and tech workers. For startups, it’s a channel to launch, collect feedback, and — when needed — manage crises. For indie developers and creators, making the front page is both validation and real traffic, so much so that people speak of the HN “hug-of-death,” when a small site crumples under the sudden load. (One of my posts briefly peaked around #60 and still managed to exhaust my monthly free bandwidth on Backblaze B2 in hours.)

But the discussions on HN are not just for startups and code. According to an analysis by Wilson Lin based on over forty million posts and comments, topics range far beyond entrepreneurship and programming. Consumer products, fundamental sciences like math and physics, and even humanities and social sciences are widely discussed, often with contributions from professionals in those fields. It’s no exaggeration to say that no matter your area of interest, you’re likely to find a worthwhile discussion about it on Hacker News.

A visualization of HN’s topics by Wilson Lin

A Disciplined Front Page and a Tireless Moderator

A large user base and a wide range of topics are not enough to make a great community. There were many large forums that hit a tipping point where low-effort posting and polarization drag everything down. How does HN resist the slide?

A well-designed set of rules do a lot of work. The HN welcome page lays out two cardinal rules: don’t post or upvote crap links, and don’t be rude or dumb in comment threads.

What counts as “not crap”? It must be something more than “superficially interesting” that “teaches you about the world,” which rules out gossip, memes, flame-bait, clickbait headlines, and other off-topic noise. And what kind of comments are “civil and substantial”? Those are ones that “more information about the topic” and not those you wouldn’t say face-to-face.

If that still sounds a bit abstract, HN provides a more comprehensive set of guidelines with specific requirements for content, formatting, and even tone. Don’t use uppercase or exclamation points. Use the original title whenever possible. Reply to the argument instead of calling names. Assume good faith. Ultimately, the sole purpose is to ensure that HN surface things that would gratify the intellectual curiosity of a good hacker.

Of course, rules alone are not enough; they require enforcement. To this end, HN combines algorithmic mechanisms and human moderation, both of which are worth looking into.

First, the ranking of posts, which determines the first impression of any forum, is not simply based on recency or interaction count but on a very strict set of criteria. New submissions start on the “New” section with one point. Only after four “upvotes” (i.e., five points total) does a post qualify for the front page. Qualified submissions are ranked by the ratio of upvotes to time since submission, and only the top thirty submissions appear on the literal “page one.”

(As an exception, moderators can give low-traction posts a “re-upping.” If they feel a post has been overlooked, they can manually place it at the bottom of the front page, giving it a second chance without overly interfering with the algorithm.)

Votes can be gamed, though. Therefore, HN has always made voting ring detection a high priority, continuously developing and improving its systems. The exact mechanisms are not public for obvious reasons, but one can infer they likely consider factors like the referrer, the account’s seniority, and frequency of operations.

Upvotes not only boost the ranking but also earn the submitter “karma,” a term borrowed from Buddhism that functions similarly to user points or credits in other communities. However, on HN, karma confers stewardship rather than badges to show off: at 31 karma you can flag posts and comments that you find violate guidelines; enough flags hide content. At 501 karma you unlock downvoting on comments. (As an easter egg, users with 251 karma can customize the color of the top navigation bar.)

These thresholds are intentionally exclusionary. Users who skim the New section are typically dedicated members with discerning taste, and attracting four upvotes from them can be hard. Indeed, statistics show that the majority of posts on HN receive 0 or 1 upvote. Consequently, many users go years without ever submitting a post that “escapes velocity,” and thus never unlock flagging. Yet, it is perhaps this willingness to sacrifice engagement for standards that has allowed HN to maintain its quality and style for over a decade.

Median score of HN posts by Max Woolf However, HN’s real differentiator isn’t algorithmic. That honor belongs to the human touch in its moderation, particularly the work of its resident moderator, dang (Daniel Gackle).

The New Yorker magazine published a profile of dang in 2019. Gackle, a Stanford literature major, became an HN moderator by a twist of fate. He had previously co-founded a startup that developed an online spreadsheet product with another former moderator, Scott Bell, and received funding from YC. When the startup eventually failed, Paul Graham invited him to join YC and manage HN full-time.

According to the article, dang’s moderation style is “personal, focused, and slow,” a form of “conversational art.” Have a look at his reply history: merging duplicate submissions, linking to related past discussions, editing submission titles and URLs for accuracy, and reminding heated users to adhere to the community guidelines. The tasks may seem trivial individually, but imagine performing them with dang’s frequency and accuracy, all while maintaining a consistently gentle and patient tone — including sending long emails explaining his moderations — and you can see a craft.

A glimpse of dang’s work It’s understandable, then, that dang has earned the universal respect of the community. Search HN for “thank you dang” around holidays and you’ll find ritualized gratitude threads. It isn’t romanticizing to say you’ll struggle to find another moderator on today’s internet so widely and openly appreciated.

(It was announced in early 2025 that Tom Howard (tomhow) has become a public moderator. He has been performing moderation tasks for years behind the scenes. According to dang, Howard has a long history with Y Combinator as a W09 batchmate and with HN, which he joined in 2007.)

Caveat Lector

HN isn’t flawless. It is, after all, an online forum, prone to all the familiar pitfalls: premature conclusions, inflammatory language, and overconfidence. Its user base, which disproportionately comprises U.S. tech workers with high skill and signal, also brings specific skews worth being aware of.

One of the chronic problems is commenting without reading: reacting to the headline while ignoring the linked piece, thereby spinning off debates the article already addresses or simply doesn’t make. Search for “RTFA” and you’ll find endless exasperation. So, it’s advisable for a new user to cultivate the habit of reading the link, or at least a competent AI summary, before diving into the comments. This helps avoid being misled by knee-jerk reactions.

“Read the fucking article” Another HN quirk is criticism for its own sake. While critical thinking is prized on HN and often serves as a powerful bullshit detector, it can sometimes devolve into nitpicking. This is most common in threads about technical achievements or business successes, where comments often throw cold water. Similarly, projects in the Show HN section often get held to standards that only make sense later in the life cycle. (Search for “so much negative” to see the pattern.)

HN also sustains a repertoire of low-yield topics that are reliably heated yet produce little insight because they devolve into preference contests or pedantry over minor technicalities. The frequent offenders include, unsurprisingly, the perennial holy wars over which programming language, operating system, or editor is superior; newer “cult” punching bags (Rust, htmx, Nix, Wayland); and recurring policy brawls over return-to-office, layoffs, and tech immigration. When you encounter these topics, consulting Wikipedia, technical documentation, or more authoritative media is often a better use of your time.

A debate surrounding Nix evangelists Finally, demographics matter. HN is dominated by American tech professionals. That can tilt discourse toward elitism, rationalism, and a kind of intellectual performance, creating an echo chamber. Therefore, a comment thread that appears to be a fierce debate may converge to a local optimum, and a data-driven articulation may turn out to rest on simplified or biased priors. As The New Yorker observed, the site has a “characteristic tone of performative erudition” that “often masks a deeper recklessness.”

As such, HN should be treated as an external comment section for the internet to the extent that it can’t substitute for the internet, much less for your own thinking. At its best, HN is a map to new questions and a window onto new angles. The landscape itself remains elsewhere.

Appendix: Tips for Reading HN

I started reading HN around early 2018. Without a background in STEM or programming, much of the discussion was beyond my knowledge, but that didn’t diminish the enjoyment. In fact, HN helped me build initial understanding and interest in many technical topics. Whenever controversial issues emerge, I habitually turn to HN for expert interpretations and opposing views, rarely coming away empty-handed.

But with numerous entries and dense discussions, reading HN well requires some technique. Based on my experience, HN isn’t ideal for mindlessly “scrolling” the front page; it’s better browsed regularly and purposefully via RSS, search, and third-party tools.

It might seem contradictory to praise the quality of the HN front page and then advise against reading it. True, the moderation mechanisms I’ve described make the front page incredibly compelling, but that itself can be a problem: without control, it easily becomes a default time-sink, leading to endless link-clicking cycles. (HN is often among websites recommended for blocking by focus-assistant tools.)

Instead, I recommend the following tools and methods for readers to consider and critique —

Hacker News has an official RSS feed (https://news.ycombinator.com/rss) that mirrors the front page, but subscribing to it directly can be overwhelming. Fortunately, HN provides a comprehensive official API, which has enabled third-party developers to create more granular RSS feeds.

A popular choice is hnrss.org, which offers a variety of feeds filtered by section, user, keyword, score, and more. Among the most useful is the “Best Comments” feed. This feed aggregates newly emerging high-score comments across HN, which are not only worth reading themselves but often lead to posts that are also worthwhile and have some traction. I frequently discover quality discussions outside my usual interests here. Its update frequency is also moderate, typically around a dozen items daily, corresponding to four or five articles — a manageable amount for most daily reading.

Search for External URLs

As mentioned, HN’s unique posting style makes it the internet’s external comments section. Combined with its high traffic, there’s a good chance any somewhat visited English-language site or page has been discussed on HN.

As such, HN search proves to be a vital source of technical due diligence: whenever a trendy, heavily promoted product appears, I search HN for its name, website domain, or GitHub repo to see if it’s genuinely unmissable or a potential “red flag.” Similarly, for any assertive, triumphant article, I search HN for dissenting voices to gain a more rounded perspective.

However, HN’s search box is tucked away at the bottom of the homepage, making it inconvenient. My suggestion is to set https://hn.algolia.com/?q=%s (where %s is the search term) as a search engine shortcut or in launcher tools for direct access. Indeed, this might be the best site search you’ve ever used. Its domain reveals it’s an “add-on,” powered and hosted by search SaaS provider Algolia (a YC alum); it’s not only blazingly fast but also supports fuzzy matching and can unearth discussions from the deepest corners of the site. (See the help page for advanced syntax.)

You can also install a browser extension like Newsit, which automatically checks if the current webpage has related HN discussions and displays a banner notification.

Hot HN posts often attract hundreds or even thousands of comments; reading them all is neither feasible nor necessary. Also, due to natural bandwagon effects, top comments attract more replies, dominating the top of the thread. Reading straight down might miss different perspectives buried later.

Therefore, I follow a personal rule: for the first top-level comment, I read at most the first three replies and their first three sub-replies. Then I move to the second top-level comment and read its first two replies and their first two sub-replies. Finally, I read the first reply to the third top-level comment and its first sub-reply. (Remember to make good use of the navigation links.) This usually provides a comprehensive overview of the thread’s viewpoints while keeping reading time manageable.

With the rise of AI tools, summarizing HN comments has become a viable option. However, for popular threads with hundreds of comments, HN paginates the results, so summarizing only the first page would be incomplete. To get around this, you can fetch the full comment data in JSON format from the HN API using this endpoint:

https://hn.algolia.com/api/v1/items/${id}

Here, ${id} is the eight-digit number from the submission’s URL. You can then feed the entire JSON response to your preferred AI model with a prompt like this:

Summarize the themes of the opinions in the input provided by the user. For each theme, include at least 3 UNMODIFIED quotes with attribution. Unescape HTML entities. Go long.

This can be a one-off prompt or set as a system prompt. The prompt design, inspired by Simon Willison’s work and adjusted based on personal experience, reliably summarizes the themes and stances within the comments, complete with original quotes and usernames for easy reference. Since this is a summarization task, cost-effective models like GPT mini, Gemini Flash, or Claude Haiku are perfectly adequate. Just be sure to use a model with a large context window to avoid truncation.

I’ve created a demo on Val Town using their free GPT-4o mini proxy. You can try it out and then fork the code to your own account to customize it and avoid rate limits.

中文版

「永恒的九月」有救吗？

1994 年初，在类似日后论坛的在线社区 Usenet 上，一群满腹恼火的用户无意间创造了一个术语——永恒的九月（Eternal September）。

不过，让人恼火的不是九月本身，而是九月出现的人。早期的 Usenet 访问门槛比较高，用户素质和内容质量相对容易维持。但每年秋季开学，都有一批大学新生通过校园网涌进 Usenet，四处乱发东西却又不守「规矩」，让老用户们烦恼不已。只是多年下来，大家也多少习惯了这个事实。

1994 年的情况又有些不同。从前一年开始，许多面向大众的互联网服务商也陆续提供了 Usenet 接入服务。这样一来，全年都有来自零基础用户的低质、跑题帖子占据社区——九月的混乱成为了永恒。

「永恒的九月」象征着互联网早期精英主义时代的结束，也代表着在线社区一个永恒的难题：用户规模、主题范围和讨论质量构成了三难困境，这三个目标的重合处大多时候写着「做梦」。

但确实有这样一个社区，在它十七年的历史中，不仅用户和流量持续增长，而且总体上保持了丰富有趣的话题和标杆性的讨论质量。这就是「黑客新闻」——Hacker News。

Hacker News 首页

文本墙里砌出的罕见人气

第一眼望去，HN 并不是一个吸引人的网站：界面素面朝天，除了字还是字，连功能按钮都主要是文本链接。不仅如此，初来乍到的人可能都不知道这里到底是怎么发帖的。与常见的在线社区不同，HN 上绝大多数「帖子」都只是一个链接分享，「楼主」的创作（如果有）只是起一个标题、加两句点评而已，而回复也是针对分享内容的讨论。

Hacker News 讨论区 换句话说，HN 与其说是一个论坛，不如说是一个集体筛选的推荐列表，一个互联网的外置评论区。这种形态的学名是「链接聚合站」（link aggregator），除了 HN，早年比较有名的例子还包括 DIGG 和 Reddit。但 DIGG 早已作古，Reddit 也逐渐转型为更「主流」的论坛模式，HN 也就越发显得独树一帜了。

就是这样一个看起来平平无奇、用起来颇有门槛的网站，却坐拥超过千万的月访问量（SimilarWeb 数据），比知名的科技新闻网站 TechCrunch 和 Engadget 都高出很多。（与这种规模的流量形成对照的是，用于托管 HN 的服务器相当朴素，仅仅是两台四核的 Intel Xeon CPU E5-2637 v4 服务器，运行 FreeBSD 系统。）

要理解 HN 的高人气，就得先了解一些历史。从域名 news.ycombinator.com 就能看出，HN 的起源并不是一个独立运营的网站，而是硅谷知名风投机构 Y Combinator 的附属项目。2007 年 2 月，时任 YC 总裁的 Paul Graham 创办了 HN。根据当时的公告，为公，他想为创业圈提供一个交流场所（这也是为什么 HN 最开始叫 Startup News），方便 YC 网罗人才；为私，他也想过一把编程瘾，用自己参与创作的 Lisp 语言变种 Arc 写一个网站。

在这样的背景衬托下，HN 逐渐成为了硅谷创业者和科技行业从业者的集散地。对于创业公司，HN 是一个推介产品、聆听反馈的优质渠道，也是在「危机公关」时需要格外小心对待的舆论场。对于独立开发者、创作者，自己的作品被「顶」上 HN 首页不仅是一种肯定，而且也能带来实打实的流量——这甚至产生了一个专有名词「HN 死亡拥抱」（HN hug-of-death），形容 HN 来客对小网站的性能考验。（我有一篇文章只是短暂蹭上了六十几名，结果几小时内 BackBlaze B2 图床就被拖完了当月额度。）

但 HN 上的讨论并不只和开公司和写代码的人有关。根据 Wilson Lin 基于四千多万条帖子和评论的分析，除了创业和编程之外，消费级产品、数理化等基础科学学科，以至社会、人文等「文科」内容，在 HN 上都有广泛讨论，也能经常见到相关背景的专业人士发言。不夸张地说，无论你处于什么领域、关心什么话题，都有很大概率在 HN 上找到你感兴趣的讨论。

HackerNews 话题视觉化统计（Wilson Lin）

纪律严明的首页与鞠躬尽瘁的管理员

用户规模有了、讨论的话题也足够丰富，但这还不足以成就一个好的社区。回忆历史，很多大型论坛就是在达到一定的规模后，遇到了严重的灌水和极端化问题，最后走向衰落。HN 是如何做到维持内容质量和讨论氛围的呢？

一套好的规则功不可没。在 HN 的欢迎页面上，写着这个社区的最重要的「约法两章」：第一，不要发垃圾链接，看到也不要点赞；第二，写评论不要粗鲁，也不要犯傻。

什么链接才不「垃圾」？答案是「有趣但不肤浅」：有助于增进对世界的了解，而不是八卦、表情包、引战文章、标题党新闻等喧闹的噪音——在 HN 的语汇中称为「无关话题」（off-topic）。什么样的评论才不「粗鲁」「犯傻」？它应当提供新的角度或者信息，「不要说你当面沟通时说不出口的话」。

如果你觉得这还是有些抽象，HN 还有一份更完整的发帖规范，对于内容、格式以至于表达方式提出了更具体的要求：不要用大写字母和感叹号来吸引眼球；尽量使用原始来源；不可以在回复观点时夹带人身攻击；在解读评论时推定他人为善意；等等。归根结底，这些原则和规则的目的都是保证 HN 上的内容能「让优秀的黑客感兴趣」，也就是「满足好奇心」。

当然，只有规则是不够的，还要有执行规则的手段。为此，HN 将程序规则和人工管理两种手段结合起来，其机制都颇值得研究。

首先，在决定着第一观感的帖子排序上，HN 不是简单地根据时间远近、互动多少，而是设置了非常严格的门槛。帖子在刚发出时只会出现在「新帖」版块，具有 1 分的初始分。只有在获得 4 次「支持」（通过点击帖子标题左侧的向上箭头），也就是积累 5 分后，才有资格进入首页排序。对于达到分数门槛的帖子，HN 按照获得分数和提交距今时间的比值来排序，只有排在前 30 名的帖子才能登上真正意义上的「首页」——直接访问 HN 网址所能看到的列表。剩下的帖子就只能排到后续页面了。

（作为例外，管理员有「特权」给低人气的帖子「第二次机会」：如果管理员觉得某个帖子似乎被「埋没」了，可以手动把它放回首页的底部，但不会更高，从而在不过度干预规则的情况下让更多人有机会看到。）

但众所周知，票数是可以刷的。因此，HN 一直将反刷票检测作为优先事项，持续开发改进。出于可以理解的原因，反刷票的具体机制没有公布过，但不难推断其考虑因素可能包括跳转来源、注册时间、操作频率等。

获得支持票除了可以让帖子排名靠前，也可以为发帖用户积累「业力」（karma）。这借用自一个佛教术语，在 HN 中大致类似于其他社区中的用户积分。不过，积分在 HN 中的作用不是提升花里胡哨的用户等级，而是参与社区治理的资历凭证：达到 31 分的用户可以标记（flag）自己认为不符合社区规则的帖子和评论，被多人标记的内容会被打上（flagged）的警告标记、直至隐藏；而只有达到 501 分的用户才能对他人评论投反对票（downvote）。（一个彩蛋功能是达到 251 分的用户可以自定义导航栏主题色。）

应当说，由于这些门槛，融入 HN 的难度高到会将很多人拒之门外的程度。不难想见，愿意主动逛新帖版块的本来就是重度用户，眼光往往挑剔；一个帖子想吸引到四个这类用户的支持，从而获得首页展示资格，实非易事。的确，据统计，HN 上的帖子大多数都只能得到 0 或 1 票。因此，相当比例的用户注册多年都没有发出过一次达到「逃逸速度」的帖子，也攒不到解锁 flag 功能的 karma 分数。但可能正是因为宁可牺牲互动量也要坚持高标准，HN 才能在十几年来维持独特的水平和风格。

HN 帖子得分中值统计（Max Woolf） 但上面那些程序规则也不能算是 HN 维持高质量最独特的「法宝」；这个荣誉还得归于 HN 运营机制中「人治」的部分，特别是常任版主（moderator）的 dang。

《纽约客》杂志曾在 2019 年对 dang 做过特写报道。他本名 Daniel Gackle，缩写一下就是 dang。这位斯坦福文学专业毕业生成为 HN 的版主纯属意外。他曾经与另一位前任版主 scott（Scott Bell）共同创业，开发在线电子表格产品，并获得过 YC 的投资。遗憾的是，dang 的创业最终未获成功，于是接受 Paul Graham 的邀请加入 YC，全职管理 HN。

用《纽约客》文章的话说，dang 的管理是「个人色彩浓厚、专注、慢节奏」的；他将自己的工作视作一种「对话」。你可以翻几页 dang 的回复记录来了解他的工作内容：合并重复主题、汇总过往类似讨论、修正帖子标题措辞和来源链接、提醒「上头」用户遵守社区规则。这些任务单独看起来可能也不复杂，但要保持像 dang 一样的高频、准确，又始终温和、耐心——包括私下和用户发送长篇邮件解释操作理由——就很难得了。

dang 的工作痕迹 正因如此，dang 受到了用户的一致尊重，以至于每到「逢年过节」都会有人自发点名感谢他的贡献（不妨试试在 HN 站内搜索 thank you dang）。诚然，HN 用户的评论也或许带有一些玫瑰色眼镜，但说很难在当今互联网上见到另一位受到如此爱戴的版主，大概不是夸张。

（2025 年初，dang 宣布 Tom Howard (tomhow) 加入正式管理员队伍。此前多年，他一直在幕后执行版主职责。据 dang 介绍，Howard 曾是 Y Combinator 的 2009 年冬令营校友，于 2007 年加入 HN。）

兼听则明

当然，HN 也不是完美无缺的。再优质的在线社区毕竟也是一个……在线社区；人们容易在线上沟通时犯的错误——急于结论、言辞偏激、过于自信——同样见于 HN 上的沟通中。同时，用户特征和文化使然，HN 还有一些「特色问题」，值得在浏览时留心鉴别。

例如，一个特别常见的现象是根本不看楼主分享的链接，只根据标题唤起的第一印象置评，导致聊起一些南辕北辙的话题，或者重提原文中已经明确回答的问题。以 RTFA（妈的去看文章，read the fucking article）为关键词搜索评论，就能看到成百上