科学与数学jacobbuckman.com
请多来点明目张胆的学术造假
Please commit more blatant academic fraud
2021 — 2025，它一次次回到首页
黑点是上首页（≥ 50 分）的那次投稿，空心点是没冲上去的投稿，点越大分数越高。
导读
机器学习研究者 Jacob Buckman 在这篇博文里指出：真正侵蚀学术的不是偶尔被抓的公然造假，而是日常的、细微的"灰色造假"——反复调参直到结果称心、只挑有利的基线对比、把误差棒重叠的差异包装成显著。他反过来主张：与其让这种隐性操纵继续，不如造假造得更明目张胆一些，因为越显眼越容易被识破、被纠正。
为什么经久不衰：它精准命名了科研激励结构里那个人人心知肚明却少有人明说的灰色地带，任何领域做过研究的人都能对号入座。
HN 评论精选
"把一个领域搞得更糟、直到它突然直面清算然后脱胎换骨"这种想法……我不知道人们是从哪儿得来的。历史上从来没有哪个领域靠这个成功过。真正会发生的倒是：领域逐渐分裂成严谨与不严谨两个阵营，就像循证医学，或者化学与炼金术。视领域而定，市场里胜出的可能是任何一方。医学研究和造桥大体严谨，编程大体不严谨。AI/ML 内部从相当不错到一塌糊涂都有，圈内人基本心里有数；反倒是圈外人以为直接拿报告指标最高的那套方法就能得到好结果，这很让人沮丧。
tlb查看英文原文
The idea that a field can be reformed by making it worse until it suddenly faces a reckoning and emerges much better is ... I don't know where people keep getting the idea that this might work. It has never worked in any field ever in history. The thing that can happen: fields gradually split into rigorous and non-rigorous camps. Like with evidence-based medicine, or chemistry/alchemy. Depending on the field, either might prevail in the market. Medical research and bridge-building are mostly rigorous, programming is mostly non-rigorous. AI/ML has a range of rigor levels, from fairly good to total crap. I think people in the field have a reasonable idea which is which. It's frustrating to people outside the field that think they can just take the technique with highest reported performance numbers and expect good results.
造假成本极低、后果几乎为零，而且被查出来也往往不点名，所以问题远比作者说的普遍。
值得注意：绝大多数情况下，即使被抓住，责任人和问题论文也不会被点名。主流期刊里有编辑心知肚明的造假论文至今还在被引用。学术造假的代价很小，而不造假的代价往往是没有教职、没有学术工作。文章结尾那种"99.99% 的人是清白的"说法并不成立，这种事极为普遍。更准确的说法是："因为每个人都参与共谋于这种微妙的造假，所以没人愿意承认它的存在。"
blagie查看英文原文
It's worth noting: In most of these cases, even when caught, the guilty parties and the bad papers remain unnamed. There are known-to-the-editors fraudulent papers in major journals still being cited. The consequences of committing academic fraud are minor, and the consequences of not doing so generally mean no academic jobs or tenure. There isn't a 99.99% innocent claim here (as at the end of the article). This stuff is widespread. This is much more accurate: "because everybody is complicit in this subtle fraud, nobody is willing to acknowledge its existence" Littman is at Brown, where the majority of the CS department engages in this !@#$%. I'm not trying to single out Brown. Parts of MIT, Stanford, etc. are even worse.
从业者按"扯淡程度"给论文分级：一半的 CS 论文是跑到结果满意为止，而同行评审反而成了共谋。
我在机器学习领域管理一个应用研究组，方向是 AutoML 和合成数据生成，工作是读最新文献并做成应用。我筛论文的第一件事就是按"扯淡 / 可能是扯淡 / 大概是真的"分类。看计算机科学的论文，一半都有这个问题：反复跑仿真，直到作者跑出自己想要的结果。有几千次引用的论文里公然出错的多得数不清——它们不按仿真次数做校正，也不控制其他变量。这不限于 CS，生物、经济、环境科学都一样。最糟的是这些还都经过同行评审，也就是说同行们都认可这是正确做法，所以我几乎不信任学术界。
lettergram查看英文原文
I manage an applied research group in the machine learning space, particularly AutoML and synthetic data generation. That means is we read the latest literature and create applications. Sometimes we build new research, most of the time we patent something new. But the goal is to get research to production as fast and as impactful as possible. The first thing I do when sorting research (papers, blog posts, presentations, etc) is sort the research by “B.S.”, “maybe B.S.”, “probably real”. If you look at research papers in computer science, half are what have the following problem: “running a simulation over and over until the authors get the results They like”. I can’t tell you how many papers with thousands of citations are just blatantly wrong. They don’t run corrections based on the number of simulations they run, they don’t take into account other variables, etc This isn’t limited to CS either. Biology, economics, environmental sciences, etc all suffer the same fate. The worst part about it, is it’s PEER REVIEWED. Meaning, others agree this is the way to do things, which is why I don’t trust academia almost at all.
一位退学的博士生证实：套路化写作加 p 值洗白就足以发表，这种环境把人逼走。
很高兴看到这篇文章。我退掉了自己的 AI 方向博士项目，部分原因就是作者和我都观察到的这种"日常造假"。论文读多了就能看出套路：提出新方法、和基线对比、讨论结果差异。有些人只是照着套路走，并不批判性地思考每一步。你会看到两张图上的点号称服从不同分布，可误差棒甚至分布本身都重叠着，但一个 p 值说明它们显著，于是这就够发表了。在这样的环境里分享自己的研究让我很焦虑，也感到压力要把工作包装成所有问题的最优解，而现实要微妙得多。
aabaker99查看英文原文
I'm really happy to see this article here. I quit my PhD where I was working on AI partly because of the "mundane day-to-day fraud" that the author and I observed in the field. Once you've read enough papers, you can easily see a kind of recipe to an academic paper. Present a new method, compare against a baseline, discuss the differences in the results found. To me, some researchers just follow this recipe and do not think about the steps critically. You see figures that plot two categories of points that are purportedly following different distributions but the error bars (or even more of the distributions) overlap. But a p-value says they are significant. So that's good enough to publish! I felt a lot of anxiety about sharing my research in this environment and felt pressure to make my work seem like it was the best solution for every problem when in reality there are more nuances.
历次讨论
同类还值得读
如果月球只有 1 像素：一个极其精确的太阳系模型（2014）
If the moon were only 1 pixel: A tediously accurate solar system model (2014)
一个网页版太阳系模型，以“如果月亮只有 1 像素”的真实比例呈现：你必须一路横向滚动，才能体会行星之间空旷到何等程度。
科学与数学上首页 7 次 · 2014–2025joshworth.com编辑精选沉浸式线性代数（2016）
Immersive Linear Algebra (2016)
《Immersive Linear Algebra》是一本完全免费、开放在线的线性代数“超文本教科书”，其最大特色是大量可交互的图形：读者可以直接拖动、操作图中的向量、矩阵和变换…
科学与数学上首页 5 次 · 2015–2024immersivemath.com编辑精选如何将贝果切成两个相连的半环（2009）
How to Slice a Bagel into Two Linked Halves (2009)
数学家 George Hart 的经典教程，演示如何用一刀把贝果切成两个相互扣住的环，并附上切割步骤与数学原理。
科学与数学上首页 5 次 · 2009–2022georgehart.com“犹太问题”：一组用于歧视的数学题
Jewish problems
一篇 arXiv 论文，整理并讨论苏联时期犹太学生在申请大学数学系时被问到的特殊数学难题：这些题往往有简单解法却极难想到，被用作面试中拒绝犹太考生的手段。
科学与数学上首页 5 次 · 2011–2021arxiv.org你的身体并非为长久而造：人类死亡率数据的一课
Your body wasn’t built to last: a lesson from human mortality rates
文章从人类死亡率随年龄变化的统计规律出发，解释身体并非为长久使用而设计：年轻时死亡风险较低，之后随年龄近似指数上升，像复杂系统在不断累积损伤。
科学与数学上首页 5 次 · 2009–2018gravityandlevity.wordpress.com