写在前面
韩晓光老师的《慢下来，你其实没有落下什么》，我认真读完了。
先说清楚：我不是来骂人的。韩老师在文章里很坦诚，他写自己读博时还没有顶会，就躲着朋友聚会；写每次顶会放榜，往往要花两三天消化；写“把我换到那个位置，我一样会焦虑”。一位老师愿意把这些摊开来讲，我是尊重的。
我也不是在为灌水辩护。这里说的快，指的是又快、质量又高：更早动手，更快验证，更快迭代，质量是底线。
倡议发出后，我把倡议帖和讨论帖底下的评论从头翻到了尾。倡议帖 562 个赞，一篇吐槽它的帖子 2071 个赞，吐槽帖下赞最多的一句话，有 1084 个赞：
“笑点解析：这些老师组里的本科生科研是最快的”
有意思的是，《慢下来》那篇长文本身有两千七百多个赞。可见“慢”这个字本身，大家并不反感。让人不舒服的，是后来那份带着签名名单的倡议。
一篇文章，是心里话。一份带着一百多个签名的联合声明，是表态。
表态，就要接受追问。这篇文章，我想把很多人想问、却没力气写成长文的追问，写完整。
1. 标准，学生早就读懂了
韩老师在文中给了一个诊断：学生们“并不了解老师们选拔学生的标准”，“这使得他们只能简单的把是否有论文和论文数量当作标准，因为这样更安全”。
我想说，学生恰恰太懂了。
同一段里，韩老师紧接着写：“坦白讲，如果有顶会论文确实会更加吸引我的眼球，但我会进一步考察他对于论文的理解和思考。”
谢谢您的坦白。不过，“进一步考察”有个前提：先得被吸引。一位老师一年只收一两个学生，收件箱里躺着几百封套磁信，“吸引我的眼球”这几个字的意思是：这封邮件，能不能被点开。这件事，每个经历过申请季的学生早就读懂了，用不着一份声明来澄清。
一位本科生的原话是：
“我当然知道自己不是真在做科研，但不弄篇文章出来后面就没书读了哒”
你看，学生什么都懂。她只是没得选。
文章还写到，如果一个低年级学生同时期有多篇投稿或者中稿，“我会非常顾虑”。把这几句话放在一起，学生读到的标准就变成了：论文要有，但不能多；要快，但不能显得快；要讲得透，但时间只有一个暑假。
这看上去是在降低要求，其实是在抬高要求。不看数量，看质量，这当然好。可一个还没有论文的学生，质量要从哪里看出来？
2. 慢，是一种奢侈品
《慢下来》里我最喜欢的一段，是登山。韩老师写：“选哪条路并不重要，重要的是往前走……只要这条路对我来说是新的，是能让我有收获的，那么就是值得走的路，哪怕最后真的是死胡同。”
这段话很美。但它有一个没写出来的前提：走进死胡同的人，还有时间退出来。
对一位已经拿到长聘的老师，死胡同是一次探索，回头还能写成一个“此路不通”的好故事。对一个大三学生，死胡同的意思是：9 月推免系统开了，你手里什么都没有；12 月申请截止了，你只能 gap 一年。而 gap 的那一年，房租、生活费、家里人欲言又止的眼神，都得自己扛。
容错率，从来都是分阶层的。
已经站稳的人，时间按职业生涯算，三十年，慢一年不算什么。还在路上的人，时间按申请季、按考核周期算：学生一年只有一个申请季，青年老师头上还有非升即走的时钟，错过就是错过。截止日期不会因为谁心态好，就往后推一天。
所以当一份倡议对学生说“慢下来”，很多人听到的其实是另一句话：“这一轮，你先别上场了。”
我相信很多人对这份倡议的感谢是真心的。但能让人睡个好觉的，除了一句“慢下来”，还得有一条走得通的路。
一开始就快的人，想慢的时候随时可以停下来，还能站出来呼吁大家慢一点。一开始就慢的人，等到想快的时候，可能已经被淘汰了，连桌都上不了，说话也没人听。
先上桌，才谈得上慢。
3. 论文是把烂尺子，可它是寒门的尺子
我承认，论文是一把很烂的尺子。它可以注水，可以挂名，可以靠 AI 堆出来。这些批评都对。
但请想一想：如果不看论文，看什么？
看面试？面试的前提，是你先被选进面试。看推荐信？推荐信得有人愿意写，写的人还得有分量。看“潜力”？潜力是见过你的人看出来的。那谁见过你？本校的、认识的、被熟人推荐来的。
不看论文，最后就只剩“举孝廉”。大家都慢下来，最先跑到前面的，往往是有关系的人。
举孝廉，是汉代的选官办法，靠推荐和名声。到了魏晋的九品中正制，结果是《晋书》里那句话：“上品无寒门，下品无势族。”后来才有了科举。科举很死板，八股文很可笑，可它给了一样任何推荐制度都给不了的东西：一个没有关系的人，只要卷子写得好，就能被看见。
今天的论文，就是一个双非学生、一个跨专业的学生、一个没有任何师承的学生，能靠自己挣到、陌生人也认的那张卷子。读博如此，找工作也一样：大厂的研究岗、算法岗，论文依然是简历上最硬的那一行。
只认“高质量”，还会带来另一个问题。韩老师在第五节写道：“一小撮人把人类知识的边界突破到了一个高度，而没有算力和数据的普通人却连这个已知的边界都够不到”。这句话放到招生上，一样成立。如果只有“高质量”的工作才算数，最后被认可的，只会是资源最多的几个组、几所学校。
而 AI 正在做相反的事：它把写代码、跑实验、写论文的门槛拉低了。门槛一低，最先受益的，就是原本够不着门槛的人。
论文当然很烂。但在更好的尺子被造出来、并且公开之前，请不要先折断我们手里唯一的这一把。
4. 灌水的药方，叫可验证
倡议里有一句话，我读了很多遍：“两三个月完成一篇‘顶会论文’，在绝大多数情况下意味着：问题不是你提出的，方法不是你设计的，论文甚至不是你写的。”
我只想补一个问题：那这篇论文，是怎么被投出去的？
韩老师在文中回忆：“回想自己读博的那个年代，如果对于某一项研究还存在一些疑惑，导师一般是不让投稿的。”
对。投不投稿，很少是一个本科生自己说了算。一个本科生的第一篇论文，问题多半是老师给的，方法多半是师兄带着定的，这在任何一个正常的课题组里，都叫带教。每一篇被称作“灌水”的论文，背后都有学生、导师、合作者和审稿人各自的一次放行。
所以如果真要给灌水“减分”，这一分，不该只扣在本科生的简历上。
有人按 DBLP 统计过签名学者 2025 年的发文量，中位数是 20 篇。这个数字本身没什么好指摘的：带一个大组，一年二十篇很正常，合作论文还会被重复计算。它说明的是另一件事：这套规则奖励产出，对学生如此，对老师们也一样。
当然，倡议里说对的地方，我也认。AI 生成的论文越来越多，审稿越来越难，有的同学确实讲不清自己的论文，基础课也被挤得越来越薄。周围人都在卷论文、卷比赛，有了 AI 以后，那种不踏实的感觉越来越重。这种不踏实，很多人都有。
可灌水的病根，是只数数、不检验。治它的药叫“可验证”，光靠慢，治不好。
我也想替“快”说句公道话：快，本来就是科研训练的一部分。韩老师自己也写过，科研就是一次次试错、分析原因、再尝试的循环。一个本科生用一个暑假，走完从想法、实验、写作到被拒的一整圈，学到的东西，很可能比在课堂上坐一年还多。与其等着跟不上趟的课程，不如早点进实验室，学点真东西。
很多学生巴不得导师天天带着做。没人带，全靠自己摸索，能不慢吗？很多时候，所谓的“慢”，只是没人带。
AI 这个领域能有今天，靠的恰恰是快：论文当天挂上预印本，代码当天开源，一个想法几周就有人复现、有人改进。韩老师在第五节里问：“我们攀登的那座山，边界还在不在原处？”
这正是学生不敢慢的原因：山在动。一个题目慢慢做两年，做完的时候，可能已经被一个大模型顺手解决了。这一点，做三维的韩老师，应该比我们体会更深。
5. 劝最弱的人先停下，最没用
这是整件事最核心的地方。现在的局面，是一个标准的囚徒困境：每个人都知道灌水不好，可别人都在灌，你不灌就被淘汰。
在囚徒困境里，道德呼吁会发生什么？它没法让所有人一起停下来。它只会筛出最老实、最听劝的那批人，然后让他们吃亏。
有人说，他们组今年说好抵制 NeurIPS，一篇都不让投，结果别的组照投照中。这就是单方面“慢下来”的真实结局。
就算你真慢下来了，审稿也不会陪你慢。你花两年磨出来的东西，照样可能被一个用 AI 写审稿意见的人，几分钟就拒掉。
要走出囚徒困境，只有两个办法：改规则，或者让所有人同时停下。前一个，签名的老师们有能力推动；后一个，靠学生自觉，根本做不到。
所以真正该被呼吁的，是基金委、科技部和学校：考核制度怎么改，年轻人怎么被善待。签名的老师们里，很多人有这个地位和话语权。人是环境的产物，要改变环境，不该从最弱势的学生开始。
我也想替老师们说句话。没有帽子的助理教授，工资不高，学校每年还有经费考核，非升即走的时钟不会因为一份倡议停下来。教授们也一样：基金要申，帽子要评，组里几十个学生的出路要操心。老师和学生，被同一套评价体系推着跑。正因为如此，这份倡议才更该往上递。
6. 我们发起“快科研”倡议
所以，我们几个做科研的人，想发起一份倡议，叫“快科研”。它不分学生和老师：本科生、研究生、博士后、青年教师，也欢迎认同它的教授们一起加入。
先把话说清楚：快科研不是灌水。我们的目标是又快、质量又高。快，是更早动手、更快验证、更快迭代；质量，是底线。
对自己，我们承诺三件事：
快，但可验证。代码开源，实验可复现，写清楚哪部分是自己做的，哪部分用了 AI。
快，但讲得清。挂着自己名字的论文，每一个设计决策都答得上来。
快，但不造假。不挂名、不代写、不编数据。
签署“慢科研”倡议的老师们，是最有分量推动规则改变的一群人。我们对各位有六个很具体的请求：
一，公开招生标准。写出来，越具体越好。能写出来的，才叫标准；写不出来的，只能叫印象。
二，只看代表作。请申请者列出自己最有代表性的三篇工作，老师只看这三篇：看贡献，看质量，看能不能讲清楚。只看三篇，刷数量就没有意义；能拿出三篇好工作的人，快一点也不吃亏。
三，招生季结束后，公开录取学生的论文数分布。脱敏即可。君子论迹不论心，我们想看见这个“迹”。
四，不成熟的论文，作为通讯作者不签字。管住署名栏的最后一格，比劝学生别焦虑有效得多。
五，向上倡议。推免加分、奖学金评定、基金和职称考核，这些规则管着学生，也管着老师们自己。在这件事上，你们说话比我们管用。
六，建一个可验证的成果平台。论文附上代码、权重和实验细节，接受同行检验，招生、评审和考核都看这个，而不是数 CCF-A。这比任何声明都更像真正的慢科研。
这些请求，其实也是为了老师们自己。评价如果看的是代表作和可验证的贡献，青年老师就不用为考核凑篇数，教授们也不用在几百封套磁信、几万篇投稿里，靠数字做判断。
我们并不需要老师们慢下来。我们需要的，是改变来得快一点。
请把“慢”写进规则，别只写进鸡汤。
规则不快，我们就只能快。
7. 快步人生路
韩老师的登山比喻里，还有一位老者。他在半山腰拦住你，问：“在你经过的第三个亭子的旁边有一束罕见的玫瑰花，它是什么颜色的？”
我很喜欢这个问题。可现实里的那座山，你还没走到半山腰，就会在山门口遇见一位保安。他不问玫瑰，只查票：“有一作吗？”
我们不是不想看玫瑰。我们只是怕看完玫瑰，山门关了。
文章最后，韩老师写到路边两个三四岁的小孩：一个一直在哭，一个一直在说对不起。哭的孩子想要的，是妈妈的怀抱；说对不起的孩子发现道歉没有用，最后自己也哭了，“他想要的对方也给不了”。
读到这里，我觉得那个哭的孩子很像我们。我们想要的，是一条写清楚的规则，一个能被看见的机会。“慢下来”也好，“你其实没有落下什么”也好，话再温柔，也替代不了它们。
那个说对不起的孩子，很像签名的老师们。我愿意相信你们是真心的。你们也被困在同一套系统里，也想给，却给不了。
那就别在原地互相道歉了。一起去找给得了的人吧：写考核办法的，定推免规则的，评基金的。
韩老师写《慢下来》的时候，正值 ICLR 截稿。他说自己没有像以前一样盯着学生的进度，而是选择相信他们，感受到了平静。
我相信这份平静是真的。但也请老师们看见：这份平静的另一端，学生的进度条，还在跑。
现在是 27fall 申请季，很多人今晚还在跑实验、改论文、写套磁信。我想对你们说：
快一点，你其实没有做错什么。你只是比谁都更早读懂了规则。
只要做得扎实、经得起检验，快就不是错。又快，又好，才是我们要的。
真正该慢下来想一想的，是写规则的人。
附：这份“快科研”倡议欢迎所有科研工作者联署、留言，不要求实名。去联署 ›
Translated from the Chinese original. Comments quoted in this essay are translated from Chinese.
Before we begin
I read Professor Han Xiaoguang’s essay “Slow Down: You Haven’t Really Fallen Behind” from beginning to end.
Let me be clear first: I am not here to attack anyone. Professor Han is candid in that essay. He writes that during his PhD, before he had a top-conference paper, he avoided gatherings with friends; that each time conference decisions come out, it often takes him two or three days to digest them; and that “if you put me in their position, I would be just as anxious.” I respect a teacher who is willing to lay all this out.
Nor am I defending paper padding. The “fast” in this essay means fast and high in quality: start earlier, verify sooner, iterate faster, with quality as the floor.
After the statement came out, I read every comment under the statement post and under a post discussing it. The statement post had 562 likes; a post mocking it had 2,071. The most-liked line under that post, with 1,084 likes, was:
“Punchline: the undergraduates in these professors’ own labs do the fastest research of all”
Interestingly, the long essay “Slow Down” itself has more than 2,700 likes. So people have no problem with the word “slow”. What made them uneasy was the statement that came later, with its list of signatories.
An essay is a personal reflection. A joint statement with more than a hundred signatures is a public position.
A public position invites questions. In this essay, I want to finish asking the questions many people wanted to ask but didn’t have the energy to write out.
1. Students understood the criteria long ago
Professor Han offers a diagnosis: students “do not understand the criteria teachers use to select students”, which “leaves them simply treating whether they have papers, and how many, as the criterion, because that feels safer.”
I would say students understand it all too well.
In the same paragraph, Professor Han continues: “To be frank, a top-conference paper does catch my eye more, but I will then look further into how the student understands and thinks about the paper.”
Thank you for your frankness. But “looking further” has a precondition: something has to catch your eye first. A professor who takes one or two students a year has hundreds of inquiry emails in the inbox, and “catches my eye” means whether an email gets opened at all. Every student who has been through an application season understood this long ago. No statement is needed to clarify it.
In one undergraduate’s own words:
“Of course I know I’m not really doing research. But if I don’t get a paper out, I won’t get to keep studying”
You see, students understand everything. She simply had no choice.
The essay also says that if a junior student has several papers submitted or accepted at the same time, “I would be very concerned.” Put these sentences together, and the standard students read becomes: have papers, but not too many; be fast, but don’t look fast; understand your work deeply, but you only have one summer.
It looks like lowering the bar; in fact it raises it. Judging quality instead of quantity is a fine idea. But for a student with no papers yet, where is the quality supposed to show?
2. Slowness is a luxury
My favourite passage in “Slow Down” is about climbing a mountain. Professor Han writes: “Which path you choose does not matter; what matters is moving forward... As long as the path is new to me and lets me gain something, it is worth taking, even if it really turns out to be a dead end.”
It is a beautiful passage. But it rests on an unstated premise: whoever walks into a dead end still has time to walk back out.
For a professor who already has tenure, a dead end is an exploration, and later it can even become a good story about a road that went nowhere. For a third-year undergraduate, a dead end means this: in September the recommended-admission system for graduate school opens and you have nothing to show; in December the application deadlines pass and you take a gap year. Through that gap year, the rent, the living costs and your family’s unspoken worry are yours to carry alone.
Tolerance for error has always depended on where you stand.
For those already established, time is measured in careers: thirty years, so one slow year means little. For those still on the way, time is measured in application seasons and review cycles. Students get one application season a year; young faculty have a tenure clock ticking over their heads. Miss it, and it is gone. No deadline moves back a single day because someone has a healthy mindset.
So when a statement tells students to “slow down”, many people hear something else: “Sit this round out.”
I believe much of the gratitude for the statement is sincere. But what lets people sleep at night is not only being told to slow down. It is also a path that actually leads somewhere.
Those who start fast can stop whenever they like, and can even stand up and call on everyone to slow down. Those who start slow may find, by the time they want to speed up, that they have already been knocked out: no seat at the table, and no one listening.
You need a seat at the table before you can afford to slow down.
3. Papers are a bad ruler, but they are the ruler of people without privilege
I admit that papers are a very bad ruler. They can be padded, they can carry gift authorship, and they can be piled up with AI. All of these criticisms are fair.
But think about it: if not papers, then what?
Interviews? First you have to be picked for one. Recommendation letters? Someone has to be willing to write one, and the writer has to carry weight. “Potential”? Potential is spotted by people who have met you. And who has met you? Students from the same university, people they know, people recommended by acquaintances.
Stop looking at papers, and all that’s left is selection by recommendation. If everyone slows down, the first to pull ahead are usually the well-connected.
Selection by recommendation (juxiaolian, “recommending the filial and incorrupt”) was how the Han dynasty chose officials: by recommendation and reputation. Under the Nine-Rank system of the Wei and Jin dynasties, the result was summed up in the Book of Jin: “No humble families in the upper ranks, no powerful clans in the lower ranks.” Only later came the imperial examinations. The exams were rigid and the eight-legged essay was absurd, but they offered something no recommendation system ever could: a person without connections could be seen, as long as they wrote a good exam.
Today, a paper is that exam. It is something a student from a lesser-known university, a student switching fields, or a student with no academic lineage can earn on their own, and that strangers will recognise. That holds for PhD admissions and for jobs: for research and algorithm roles at big tech companies, papers are still the strongest line on a CV.
Accepting only “high quality” brings another problem. In the fifth section of his essay, Professor Han writes: “A small group of people has pushed the frontier of human knowledge to a new height, while ordinary people without compute or data cannot even reach the known frontier”. The same is true of admissions. If only “high-quality” work counts, the ones recognised will be the few labs and universities with the most resources.
AI is doing the opposite. It lowers the barrier to writing code, running experiments and writing papers. When the barrier drops, the first to benefit are those who could not reach it before.
Of course papers are a bad ruler. But until a better one is built and made public, please don’t break the only ruler we have.
4. The cure for padding is verification
There is one sentence in the statement that I have read many times: “Finishing a ‘top-conference paper’ in two or three months, in the great majority of cases, means: the problem was not posed by you, the method was not designed by you, and the paper was not even written by you.”
I only want to add one question: then how did that paper get submitted?
Professor Han recalls: “Back when I was doing my PhD, if there were still some doubts about a piece of research, advisors generally would not let us submit it.”
Exactly. Whether to submit is rarely an undergraduate’s call alone. For an undergraduate’s first paper, the problem usually comes from the advisor and the method is usually settled under a senior student’s guidance. In any normal research group, that is called mentoring. Every paper dismissed as “padding” was let through, in turn, by a student, an advisor, collaborators and reviewers.
So if padding really should cost points, those points should not come only out of an undergraduate’s CV.
Someone used DBLP to count the signatories’ papers published in 2025: the median is 20. There is nothing wrong with that number in itself. Running a big group, twenty papers a year is normal, and collaborative papers are counted more than once. What it shows is something else: these rules reward output, for students and for teachers alike.
Of course, I also accept what the statement gets right. AI-generated papers keep multiplying, reviewing keeps getting harder, some students really cannot explain their own papers, and foundational courses are squeezed thinner and thinner. Everyone around you is competing on papers and contests, and since AI arrived, that unsteady feeling keeps growing. Many people share it.
But the root of padding is counting without checking. The cure is called verification. Slowness alone will not cure it.
I also want to say a fair word for speed: speed has always been part of research training. Professor Han himself wrote that research is a cycle of trial and error: analyse why, then try again. An undergraduate who completes one full loop in a summer, from idea to experiment to writing to rejection, may well learn more than in a year of sitting in class. Rather than waiting on courses that lag behind, it is better to get into a lab early and learn something real.
Many students would love an advisor who guides them every day. Left to figure everything out alone, how could they not be slow? Very often, what gets called “slow” simply means no one is guiding you.
AI got where it is today precisely by being fast: papers go up as preprints the day they are finished, code is open-sourced the same day, and within weeks others reproduce an idea and improve it. In the fifth section of his essay, Professor Han asks: “Is the frontier of the mountain we are climbing still where it was?”
That is exactly why students don’t dare to slow down: the mountain is moving. Spend two slow years on a problem, and by the time you finish, a large model may have solved it in passing. Professor Han, who works on 3D, probably feels this more deeply than any of us.
5. Asking the weakest to stop first does the least good
This is the heart of the matter. We are in a textbook prisoner’s dilemma: everyone knows padding is bad, but everyone else is padding, and if you don’t, you are eliminated.
What does a moral appeal do in a prisoner’s dilemma? It cannot make everyone stop at once. It only singles out the most honest and most obedient, and then they lose out.
One person said their group agreed to boycott NeurIPS this year and submitted nothing, while other groups kept submitting and kept getting in. That is how a one-sided “slow down” really ends.
Even if you do slow down, reviewing will not slow down with you. Work you spent two years polishing can still be rejected in minutes by a reviewer whose comments were written by AI.
There are only two ways out of a prisoner’s dilemma: change the rules, or get everyone to stop at the same time. The signatories are able to push for the first. The second simply cannot be achieved by asking students to restrain themselves.
So the real appeal should go to the Natural Science Foundation, the Ministry of Science and Technology and the universities: how evaluation should change, and how young people should be treated. Many of the signatories have the standing to make that appeal. People are products of their environment, and changing the environment should not start with the weakest students.
I also want to speak up for teachers. Assistant professors without a talent title are not well paid, universities set yearly funding targets, and the tenure clock will not stop for a statement. Professors are in the same position: grants to apply for, titles to compete for, and the futures of dozens of students in their group to worry about. Teachers and students are pushed along by the same evaluation system. That is exactly why this statement should be sent upward.
6. We are launching a “Fast Research” initiative
So a few of us who do research want to launch an initiative called “Fast Research”. It does not divide students from teachers: undergraduates, graduate students, postdocs and young faculty are all welcome, and so are the professors who agree with it.
Let us be clear: Fast Research is not about padding. Our goal is research that is both fast and high in quality. Fast means starting earlier, verifying sooner and iterating faster. Quality is the floor.
To ourselves, we commit to three things:
Fast, but verifiable. Open code, reproducible experiments, and a clear account of what we did ourselves and what AI did.
Fast, but explainable. We can answer for every design decision in a paper that carries our name.
Fast, but honest. No gift authorship, no ghostwriting, no fabricated data.
The teachers who signed the Slow Research statement carry more weight than anyone in changing the rules. We have six very concrete requests for you:
One: publish your admission criteria. Write them down, as concretely as possible. Whatever can be written down is a standard; whatever can’t is only an impression.
Two: judge by representative work. Ask applicants to list their three most representative pieces of work, and look only at those three: the contribution, the quality, and whether the applicant can explain them. If only three count, padding the count is pointless, and someone with three good pieces of work loses nothing by being fast.
Three: after each admission season, publish the distribution of paper counts among admitted students. Anonymised is fine. A person is judged by deeds, not intentions; we would like to see the deeds.
Four: as corresponding authors, don’t sign off on papers that aren’t ready. Guarding the last name on the author list does far more good than telling students not to be anxious.
Five: push for change upstream. Credit toward recommended admission, scholarship decisions, grant and promotion reviews: these rules govern students, and they govern teachers too. On this, your voices carry further than ours.
Six: build a platform for verifiable work. Attach code, weights and experimental details to papers, open them to peer verification, and let admissions, review and promotion look at that instead of counting CCF-A papers. That would be closer to real slow research than any statement.
These requests are also for teachers themselves. If evaluation looked at representative work and verifiable contributions, young faculty would not need to pad their publication counts, and professors would not need to judge by numbers across hundreds of inquiry emails and tens of thousands of submissions.
We don’t need teachers to slow down. We need change to come a little faster.
Please write “slow” into the rules, not just into pep talks.
If the rules won’t move fast, we have to.
7. Walking fast through life
In Professor Han’s mountain metaphor there is also an old man. He stops you halfway up and asks: “Beside the third pavilion you passed, there is a rare rose. What colour is it?”
I love that question. But on the real mountain, before you get halfway up, you meet a guard at the gate. He doesn’t ask about roses. He only checks your ticket: “Do you have a first-author paper?”
It’s not that we don’t want to look at the roses. We are just afraid that by the time we’ve looked, the gate will be closed.
At the end of his essay, Professor Han describes two children of three or four by the roadside: one kept crying, the other kept saying sorry. What the crying child wanted was his mother’s embrace. The child saying sorry found that apologising changed nothing, and in the end he cried too: “what he wanted, the other could not give either.”
Reading this, I felt we are like the crying child. What we want is a clearly written rule and a real chance to be seen. “Slow down” or “you haven’t really fallen behind”: however gentle the words, they cannot replace those things.
The child saying sorry is like the teachers who signed. I am willing to believe you mean it. You are trapped in the same system; you want to give, but you can’t.
Then let’s stop apologising to each other where we stand. Let’s go together to those who can give: the people who write the evaluation rules, who set admission policy, who decide on grants.
Professor Han wrote “Slow Down” during the ICLR submission deadline. He says that, unlike before, he did not keep watching his students’ progress but chose to trust them, and felt at peace.
I believe that peace is real. But I also ask teachers to see that on the other side of that peace, students’ progress bars are still running.
It is now the fall 2027 application season, and many people are still running experiments, revising papers and writing inquiry emails tonight. To you I want to say:
Go faster. You haven’t done anything wrong. You simply understood the rules earlier than anyone.
As long as the work is solid and stands up to checking, speed is not a mistake. Fast and good: that is what we want.
The ones who most need to slow down and think are the people who write the rules.
P.S. Every researcher is welcome to sign and comment on this initiative. No real name required. Sign ›