HackerTube English?
表示、直觉与反馈 →

换一种表示,哪些事实才会显出来?

均值、频谱和几何图各自把某些结构推到眼前,也会藏起别的东西。不要只问哪张图直观,先问想比较的是什么。

本文的问题 · 42
  1. 全球进步与国内财富集中可以同时成立
  2. 可视化既呈现数据,也塑造理解
  3. 改变表示,可以直接看见特定问题的解
  4. 先理解频率分量,再区分采样与量化
  5. 精度是否足够取决于误差落在哪里
  6. 从套用面积公式走向可迁移的推导模型
  7. 无穷多项、有限答案:两种算法的边界在哪里?
  8. 画同一个句子,哪张图方便量距离?
  9. 把同一群人按国家、收入和多数宗教各排一次
  10. 积木能表示十亿人,也能表示一个预测吗?
  11. 数组和 AWK 各替你写了哪层循环?
  12. 哪一层中间表示还认得原来的运算?
  13. 这杯子是相对于谁在移动?
  14. 动作越顺,为什么越不像手绘?
  15. 同一份输入,错开以后会发生什么?
  16. 还原出一张清晰的图,为什么数字会错?
  17. 一笔钱的数字能代表多少实际工作?
  18. 人数都画在一条线上,会隐藏哪种变化?
  19. 一张地图里,哪些边界允许叠在一起?
  20. 在同一张地图上标两个集合
  21. 怎样理解同一讲者的第二次解释?
  22. 为什么同一段视频的字幕与配音各说各的?
  23. 让沙子承重和让沙子流动,改变的是哪项约束?
  24. 一盏红灯亮起来,后车还缺哪条信息?
  25. 信号频率相同,何时才算真正同步?
  26. 这些热还有能量,为什么不能无限做同样多的计算?
  27. 唱针读到的究竟是振动,还是随沟槽改变的电容?
  28. 每一刻只有一道光或一种色,图像怎样连起来?
  29. 为什么同一块橙色会变成棕色?为什么亮度能变成距离?
  30. 两个群体的平均值,能替你读出一个人的值吗?
  31. 点亮一个点,与指向一个像素
  32. 保持滤波器不动,让信号移过来
  33. 改口型、改剪辑顺序,观众多知道了什么?
  34. 相差一分,什么时候变成了两类人?
  35. 参数已经走到一半,路程也走完一半了吗?
  36. 把熟悉的编程媒介暂时拿走,会想到什么?
  37. 编辑器看到的是一行字符,还是一棵树?
  38. 同一份表示能留下什么,又能换掉什么?
  39. 走同样一步,视野扩大了多少?
  40. 把公式移一格,哪个引用会跟着动?
  41. 把画面写成一个函数以后,哪里还可以单独改?
  42. 轴转了一圈,刻度读出的数从哪来?

全球进步与国内财富集中可以同时成立

Rosling 展示国家健康和收入均值的长期改善,同时明确这些均值掩盖国内差距;美国财富分配视频把镜头缩至一国内部的财富份额,揭示平均图景看不到的集中。

对照原文与时间戳 · 2
1% of America has 40% of all the nation's wealth
Wealth Inequality in America · ▶ 4:40–4:57
Inequalities within countries these bubbles show country averages, but I can split them
Hans Rosling's 200 Countries, 200 Years, 4 Minutes - The Joy of Stats - BBC · ▶ 3:25–3:41

财富存量与人均收入、寿命是不同指标,两个视频的数据时期也不同,不能拿份额直接反驳全球健康进步。

可视化既呈现数据,也塑造理解

Matplotlib 演讲把色图称作数据与大脑之间的接口;Rosling 用动态气泡图让国家长期变化可见;二者提示可视编码决定观众能察觉哪些模式。

对照原文与时间戳 · 2
an interface between your data and your brain
A Better Default Colormap for Matplotlib | SciPy 2015 | Nathaniel Smith and Stéfan van der Walt · ▶ 0:40–0:47
having the data is not enough
Hans Rosling's 200 Countries, 200 Years, 4 Minutes - The Joy of Stats - BBC · ▶ 0:10–0:17

Rosling 同时提醒国家平均值掩盖内部差距;动画和色彩都不能替代对数据定义及分布的审视。

改变表示,可以直接看见特定问题的解

Kay以画圆和数据结构说明恰当表示能承担推理工作;Strang给出具体例子:三平面交点难以看清,但把同一方程改写为列向量组合后,可看出右端就是第三列,因此系数为(0,0,1)。这是表示改变认知成本的实例。

对照原文与时间戳 · 2
choose the appropriate data structure before you start tinkering around with the algorithm.
Alan Kay: Doing with Images Makes Symbols (Full Version) · ▶ 71:12–71:16
that right-hand side that I'm trying to get is actually one of these columns.
Lec 1 | MIT 18.06 Linear Algebra, Spring 2005 · ▶ 24:38–24:45

Strang明确说此例经过特意选择;列视图也不总能肉眼读出解,一般求解仍需要消元等方法。此比较不声称列视图普遍优于行视图。

每个像素都被碰到,就等于填满了平面吗?

把希尔伯特折线画得足够密,它会经过屏幕上的每个格子。可放大一个格子,折线之外仍有空白。数学上填满正方形的曲线是整个序列的极限,不是其中最精细的那张图。无限和的讲解给了另一种检查直觉的方法:只说部分和「靠近过」1 不够;给定再小的误差,也要能找到此后一直留在误差内的位置。两种极限的对象不同,作用都是把「越来越像」换成可检验的条件。

原文与时间戳 · 2
The output of the Hilbert curve function evaluated on this input is just defined to be the limit of those points.
Hilbert's Curve: Is infinite math useful? · ▶ 13:51–13:59
no matter how small the distance you choose, at some point down the list, all the numbers start falling within that distance of x.
What does it feel like to invent math? · ▶ 4:09–4:25

海岸线的方格计数则问一个经验问题:换一种格子大小,被碰到的格子数怎样变?在对数坐标上测得的斜率可能随尺度区间改变。只拿一段近似直线,不能证明真实海岸在任意细的尺度上都服从同一规律。下次看到一张细节丰富的图,可以先问:这是一个有限样本,一次尺度比较,还是一个关于极限的命题?

原文与时间戳 · 1
this dimension, at least as I've described it so far using the box counting method, can sometimes change based on how far zoomed in you are.
Fractals are typically not self-similar · ▶ 15:38–15:51

先理解频率分量,再区分采样与量化

傅里叶变换演示如何从混合信号中分离频率成分;数字音频演示强调采样与量化是不同步骤;连读可避免把频谱分析、采样重建和量化误差混为一谈。

对照原文与时间戳 · 2
It pulls out the original frequencies from their jumbled up sums
But what is the Fourier Transform? A visual introduction. · ▶ 10:29–10:33
the sampling step is perfect it loses nothing
D/A and A/D | Digital Show and Tell (Monty Montgomery @ xiph.org) · ▶ 9:38–9:47

两支视频的教学重点不同;傅里叶变换本身并不保证任意采样信号都能无损重建。

精度是否足够取决于误差落在哪里

模拟计算视频以神经网络可容忍小幅误差解释低功耗推理机会;数字音频视频则把量化误差细分为噪声与输入相关失真,并用dither改变误差性质。两者共同要求按任务评估误差,而非只比较位数。

对照原文与时间戳 · 2
neural networks don't need the precision of digital computers.
Future Computers Will Be Radically Different (Analog Computing) · ▶ 14:41–14:57
dither is specially constructed noise that substitutes for the noise produced by simple quantization
D/A and A/D | Digital Show and Tell (Monty Montgomery @ xiph.org) · ▶ 12:09–12:28

模拟推理误差与音频量化噪声不是同一机制,不能把音频抖动方案直接套到神经网络。

从套用面积公式走向可迁移的推导模型

Mazur主张学习应形成能迁移到新情境的心智模型;微积分视频以圆环近似、矩形求和与曲线下面积重建圆面积公式,使一次计算转化为可迁移的积分思路。

对照原文与时间戳 · 2
build mental models that you can use in other contexts
Confessions of a Converted Lecturer: Eric Mazur · ▶ 27:08–27:19
the sum, thought of as the aggregate area of those rectangles, approaches the area under the graph.
The essence of calculus · ▶ 8:07–8:18

视频推导不是同伴教学实验,不能由此推断具有同样的课堂学习增益。

无穷多项、有限答案:两种算法的边界在哪里?

追赶问题可以列出无限多段路,也可以设双方同时到达某点并直接解方程。指数函数的级数也能截到有限项做近似,再用单位圆在角度 π 处的位置检查结果。这里容易养成一个坏习惯:凡是式子画得顺,就把它带到更远的输入。ζ 函数原来的级数有收敛域,解析延拓讲的是满足额外条件的函数怎样延伸;它没有使原来发散的加法过程在新位置变得收敛。比较两种表示之前,先在各自旁边写下适用的输入范围。

原文与时间戳 · 3
since they both are at this point at the same time
Preface | MIT Calculus Revisited: Single Variable Calculus · ▶ 29:16–29:19
you can chop off the series pretty early to get an approximate value
What is Euler's formula actually saying? | Ep. 4 Lockdown live math · ▶ 8:57–9:07
extending an analytic function in the only way possible that's still analytic
But what is the Riemann zeta function? Visualizing analytic continuation · ▶ 16:04–16:12

画同一个句子,哪张图方便量距离?

Gibson 讨论英语等语言时,可以把句子画成逐层组合的短语,也可以直接把彼此依赖的词连起来。在第二张图上,沿原句数两个相关词相隔几位变得直接。先明确要比较的是成分组成还是依赖跨度,再决定采用哪张图;仅凭某种树画起来更简洁,还不能推断人的理解机制就是那棵树。

原文与时间戳 · 2
These are sequences of words which are connected and the connections are dependencies between the words.
Edward Gibson: Human Language, Psycholinguistics, Syntax, Grammar & LLMs | Lex Fridman Podcast #426 · ▶ 9:03–9:47 · 段落范围
the lengths of dependencies between words, which isn’t so obvious in the phrase structure
Edward Gibson: Human Language, Psycholinguistics, Syntax, Grammar & LLMs | Lex Fridman Podcast #426 · ▶ 45:52–46:21 · 段落范围

把同一群人按国家、收入和多数宗教各排一次

先看一国的平均点,再把它拆成五个收入组。Rosling 在乌干达的例子里让一个平均点变成了一段内部差距。Rönnlund 则把各国的家庭按收入排在一起,让读者比较同一种炉灶或寝具。另一次演讲里,Rosling 给国家按多数宗教上色:同色国家仍覆盖多种生育率。三种重排都让一个被压平的维度出现。照片可以说明某个家庭怎样生活,国家级散点可以揭示分组内部的差异;由后者直接推断个人信仰怎样导致个人选择,就换了分析单位。

原文与时间戳 · 3
split Uganda there's quite a difference within Uganda these are the quintiles of Uganda the richest 20% of Uganda's are there the poorest are down there if I
The best stats you've ever seen | Hans Rosling · ▶ 14:13–14:26
But look at these stoves. Very different, right, because it depends on what income level you have, how you're going to cook your food.
See how the rest of the world lives, organized by income | Anna Rosling Rönnlund · ▶ 7:49–7:56
you see countries all the way. But there's no major difference. There's no major difference between these religions. There is a difference with income.
Religions and babies | Hans Rosling · ▶ 6:23–6:32

积木能表示十亿人,也能表示一个预测吗?

Rosling 的积木容易记:一盒十亿人,每挪一盒,量级都不变。但讲者随后给出的未来情景还取决于儿童生存和教育条件。另一场观众问答中,两位 Rosling 指出,人们会用最困难地区的经历替代多数人的情况。读这种图,可以把「每盒代表多少人」「盒子里有哪些不同群体」「哪项条件使未来要多摆一盒」分开回答。第一问是刻度,后两问都不是数盒子能解决的。

原文与时间戳 · 3
This box contains one billion people.
Hans Rosling: Global population growth, box by box · ▶ 0:53–0:55
So the only way of really getting world population [growth] to stop is to continue to improve child survival to 90 percent.
Hans Rosling: Global population growth, box by box · ▶ 8:22–8:31
Now, we miss the majority. When you answer, you answer according to the worst places, and there you are right, but you miss the majority.
How not to be ignorant about the world | Hans and Ola Rosling · ▶ 4:28–4:38

数组和 AWK 各替你写了哪层循环?

想把每一行符合条件的金额翻倍。Kernighan 的 AWK 会逐文件、逐行匹配,程序里只出现模式与动作。换成数值数组,Oliphant 回忆的 Numeric 让整个数组的运算成为可直接表达的对象。两种写法都缩短代码,但各自保留了不同的问题:文本里什么算一条记录,数值里哪些元素占据相同位置。改变表示之前,先问原来那层循环是纯粹重复,还是悄悄做了对齐。

原文与时间戳 · 4
it reads through each of the lines
Brian Kernighan: UNIX, C, AWK, AMPL, and Go Programming | Lex Fridman Podcast #109 · ▶ 37:26–37:28
you just write the pattern and the action
Brian Kernighan: UNIX, C, AWK, AMPL, and Go Programming | Lex Fridman Podcast #109 · ▶ 37:48–37:49
Numeric filled the gap of having an array object
Travis Oliphant: NumPy, SciPy, Anaconda, Python & Scientific Programming | Lex Fridman Podcast #224 · ▶ 10:48–10:50
Numeric had math and a basic way to think in arrays
Travis Oliphant: NumPy, SciPy, Anaconda, Python & Scientific Programming | Lex Fridman Podcast #224 · ▶ 11:17–11:20

哪一层中间表示还认得原来的运算?

把 JavaScript 里的条件表达式交给编译器,先要回答哪些值算假。Lattner 解释,这种语言自己的语义可以留在前端,之后把运算交给更通用的中间部分优化。谈 MLIR 时,他从相反方向指出,各领域从头做一整套编译器也很昂贵。试着为张量乘法挑一个接口:降到足够低才能共享现有优化,但若太早丢掉「这是矩阵乘法」这条信息,后面的领域工具就得重新认出它。

原文与时间戳 · 4
it's much more language independent
Chris Lattner: Compilers, LLVM, Swift, TPU, and ML Accelerators | Lex Fridman Podcast #21 · ▶ 20:46–20:48
that middle part can be shared across all those
Chris Lattner: Compilers, LLVM, Swift, TPU, and ML Accelerators | Lex Fridman Podcast #21 · ▶ 21:00–21:04
building the whole thing from scratch for each domain
Chris Lattner: The Future of Computing and Programming Languages | Lex Fridman Podcast #131 · ▶ 67:16–67:20
build these domain-specific compilers in a much faster way
Chris Lattner: The Future of Computing and Programming Languages | Lex Fridman Podcast #131 · ▶ 68:34–68:41

这杯子是相对于谁在移动?

闭上眼,手指沿杯壁挪动一点。Hawkins 的假说把预测下一次触感所需的位置钉在杯子上。把杯子连同手一起平移,房间坐标全变了,手指相对于杯子的位置却可以相同。Bach 谈的自我模型把另一个关系放到前景:一个身体试图调节自己与环境的哪种关系?两人提出的理论各需各的证据。它们给读者的共同练习是指出模型坐标属于谁,以及哪种移动应当改变预测。

原文与时间戳 · 4
the location of my finger relative to that cup
Jeff Hawkins: Thousand Brains Theory of Intelligence | Lex Fridman Podcast #25 · ▶ 35:59–36:04
a reference frame that's anchored to the cup
Jeff Hawkins: Thousand Brains Theory of Intelligence | Lex Fridman Podcast #25 · ▶ 36:13–36:17
It's a tool for modeling
Joscha Bach: Artificial Consciousness and the Nature of Reality | Lex Fridman Podcast #101 · ▶ 44:50–44:52
it basically has regulation targets that it tries to get to
Joscha Bach: Artificial Consciousness and the Nature of Reality | Lex Fridman Podcast #101 · ▶ 45:08–45:12

动作越顺,为什么越不像手绘?

先让一个角色贴墙侧滑。Rosen 用身体真实移动的方向带动转身。在另一个攀爬示范中,用逆向运动学约束手脚的接触点。《罪恶装备 Xrd》偏要破坏另一种连续性:关掉帧间插值,还在每个关键帧稍微掰动网格。因为一个严格保持形状的物体穿过透视空间,观众会立刻看出它是三维刚体。两边都请观众看动作,但一个要保住着地感,另一个要保住画师选择姿态的痕迹。

原文与时间戳 · 2
his rotation is reactive to the velocity
Animation Bootcamp: An Indie Approach to Procedural Animation · ▶ 4:49–4:51
deform the mesh every keyframe to add imperfection
GuiltyGearXrd's Art Style : The X Factor Between 2D and 3D · ▶ 31:58–32:02

同一份输入,错开以后会发生什么?

瀑布底下的泡沫共用一条颜色渐变。所有色块同相变化就会闪烁,让它们从序列的不同位置起步才有翻涌感。《毁灭战士》的配乐换了媒介,同一个正弦波可以分流到不同的处理链上,产生一批不同音色。在这两处,先固定输入,再单独听一路、看一块,有助于找出丰富性究竟来自输入、变换,还是几路变换之间的相对关系。

原文与时间戳 · 2
the gradient is all happening in all of these little cells at the same time
8 Bit & '8 Bitish' Graphics-Outside the Box · ▶ 57:58–58:04
That computer split that signal out into four different paths.
DOOM: Behind the Music · ▶ 17:39–17:43

还原出一张清晰的图,为什么数字会错?

在 JPEG 的 8×8 小块里,每个系数记录一个余弦图案的贡献。量化把附近的系数舍入到同一档,重新绘图时会失去一点细节。Kriesel 的扫描仪案例有另一种后果:把长得相似的字形当作同一图块重复使用。若把两个不同字符归进一组,输出照样锐利,文本的值却改变了。检查压缩是否可接受要先选验收题:让人认出一件毛衣,还是正确抄下工程图上的每个数字?

原文与时间戳 · 3
representing the contribution of each of these individual blocks to the whole.
JPEG DCT, Discrete Cosine Transform (JPEG Pt2)- Computerphile · ▶ 4:51–4:57
round to the nearest integer.
JPEG DCT, Discrete Cosine Transform (JPEG Pt2)- Computerphile · ▶ 9:02–9:04
Das Pattern Matching findet das I ist eben ähnlich zum kleinen L
David Kriesel: Traue keinem Scan, den du nicht selbst gefälscht hast · ▶ 29:19–29:24

一笔钱的数字能代表多少实际工作?

Felix 拆开按揭月供,区分已经转成权益的本金与难以收回的支出。Jancovici 拆开生产价值,追问哪些物料、机器和机械功让产品成为可能。两人讨论的不是同一市场。可共同使用的是检验单位的方法:当你把两个金额并排比较时,先各自写出它代表的资产变化或实物过程。

原文与时间戳 · 4
a combination of interest and a principal repayment
Renting vs. Buying a Home: The 5% Rule · ▶ 1:24–1:28
variables like tax rates and portfolio asset mix
Renting vs. Buying a Home: The 5% Rule · ▶ 7:37–7:43
we need resources to transform
Jancovici : Can we save energy, jobs and growth at the same time ? 08/01/2018 · ▶ 34:44–34:47
amount of money that we can distribute
Jancovici : Can we save energy, jobs and growth at the same time ? 08/01/2018 · ▶ 60:23–60:24

人数都画在一条线上,会隐藏哪种变化?

人口总数仍在上升,同时每个家庭的孩子却可能减少。影片让人把曲线分解成年龄结构和生育率。讨论创新的另一段动画也拆开「人变多」:有机会做研究的人数,与买得起成果的人数,可以沿不同方向变化。给任意一条总量曲线注上两个会独立变化的分量,再试着预测下一年。

原文与时间戳 · 4
Fewer babies were conceived
Overpopulation – The Human Explosion Explained · ▶ 3:10–3:13
far fewer children on average than their parents
Overpopulation – The Human Explosion Explained · ▶ 4:01–4:05
more people have the freedom and education to contribute
A Selfish Argument for Making the World a Better Place – Egoistic Altruism · ▶ 3:48–3:56
increase the size of the market for innovations
A Selfish Argument for Making the World a Better Place – Egoistic Altruism · ▶ 4:05–4:08

一张地图里,哪些边界允许叠在一起?

先在英国示意图上圈出大不列颠岛,再圈出联合王国。北爱尔兰会让两条线分开。Grey 另一期节目用一圈圈「轨道」讲脱欧谈判,那里的距离表示一组制度承诺,与海岸线无关。给每根线写一个可检查的图例:陆地、主权还是条约权利?不写清楚,叠两张图时很容易凭形状猜出了一个新的结论。

原文与时间戳 · 4
Great Britain is the largest island
The UK Is a Country Made of Countries · ▶ 1:18–1:20
the United Kingdom of Great Britain and Northern Ireland
The UK Is a Country Made of Countries · ▶ 1:07–1:10
Free movement of goods, free movement of services
The EU Has Orbits, and the UK Fell Out of All of Them · ▶ 1:07–1:12
a list of deal-breakers
The EU Has Orbits, and the UK Fell Out of All of Them · ▶ 0:42–0:48

在同一张地图上标两个集合

Grey 用荷兰的省名和王国的范围提醒人留意「这个地名数了谁」。他的欧洲组织图又把通行安排与成员资格分开。同一个瑞士可以在两张图表中占不同的位置。把有争议的名称换成一个可判断是否属于它的集合,重叠与差异就能直接检查。两次示例同出一位作者。

原文与时间戳 · 3
the Netherlands is divided into twelve provinces
The Netherlands Is Not 'Holland.' It Never Was. · ▶ 0:29–0:35
the Schengen Area is a bit mind-blowing
Europe Is Actually Five Different Groups of Countries · ▶ 1:57–2:04
part of Schengen but not part of the European Union
Europe Is Actually Five Different Groups of Countries · ▶ 2:17–2:21

怎样理解同一讲者的第二次解释?

Jancovici 在一场演讲中用机器做功解释生产能力,在另一场课里把物理变化和货币计价都落在资源加工这个动作上。把两处并排有助于辨清他说「价值」时究竟在用哪个单位。两场都是同一讲者的阐述;想验证这种对应能否用于比较不同经济体,还需要能独立核对的数据。

原文与时间戳 · 3
As energy quantifies transformation
1 - L'énergie - Cours des Mines 2019 - Jancovici - [EN subtitles available] · 字幕译文 · ▶ 36:52–36:58
we count the moment when we transform
1 - L'énergie - Cours des Mines 2019 - Jancovici - [EN subtitles available] · 字幕译文 · ▶ 128:13–128:22
we need resources to transform
Jancovici : Can we save energy, jobs and growth at the same time ? 08/01/2018 · ▶ 34:44–34:47

为什么同一段视频的字幕与配音各说各的?

把印地语的公司名放到句末,配音演员的嘴型就更容易和原片合上。字幕则需要让观众在镜头切换前读完,可以采用另一种句法。看两个版本时,先把原句承担的事实和笑点写下来,再分别验证能否读完、口型能否对上、含义有无变化;把两份译文逐词比对会漏掉前两项。

原文与时间戳 · 2
match lip movements as closely as possible.
Why don't subtitles match dubbing? · ▶ 1:07–1:11
they have to fit the words on the screen.
Why don't subtitles match dubbing? · ▶ 7:07–7:09

让沙子承重和让沙子流动,改变的是哪项约束?

送气前后用的是同一槽沙。空气改变颗粒之间力的传递,演示里有的物体下沉、有的上浮。与加筋土受横向约束后承重的案例对读,能避免给「沙」贴一个固定的强弱标签。把边界画出来,分别标注空气从哪里进入、颗粒能向哪里移动,再讨论同一材料会给出什么结果。

原文与时间戳 · 2
With air on, the top surface is nearly frictionless.
Liquid Sand Hot Tub- Fluidized air bed · ▶ 1:07–1:10
some stuff floats in the sand and other stuff sinks.
Liquid Sand Hot Tub- Fluidized air bed · ▶ 5:36–5:40

一盏红灯亮起来,后车还缺哪条信息?

把刹车和转向合在同一个灯泡上,在日常状态下许多驾驶者能读懂。故意把问题换成一只灯失效,或者双闪开启时又踩刹车,同一表示能区分的状态就减少。旧电视加颜色时遇到类似的兼容条件:给新设备传递颜色,还得给旧设备留下可理解的亮度。检验兼容时,用同时发生的动作和部分失效的设备构造两个难例。

原文与时间戳 · 4
if you have a brake light go out, now you’ve also lost a turn signal!
The Senseless Ambiguity of North American Turn Signals · ▶ 4:10–4:14
the hazards can be on and the brake lights work completely normally.
The Senseless Ambiguity of North American Turn Signals · ▶ 5:24–5:28
The reason for the fussing about had to do with compatibility.
How Analog Color TV Works: The Beginnings · ▶ 0:28–0:32
Picking just one color would produce a very unnatural image on the black and white set.
How Analog Color TV Works: The Beginnings · ▶ 12:24–12:28

信号频率相同,何时才算真正同步?

看到邻车闪灯与自己同时亮起,再等几个周期,看相位是否逐渐错开。符合一个频率范围只说明每分钟闪多少次大致合格,不能保证在何时开始闪。若想让另一辆车永远同时亮,除了更准的振荡器,还得说明两边用哪个事件建立相位参照。

原文与时间戳 · 3
between 60 and 120 flashes per minute.
Why it's not possible to synchronize turn signals (but also absolutely is) · ▶ 20:59–21:04
turn signal flashers just weren’t precision devices
Why it's not possible to synchronize turn signals (but also absolutely is) · ▶ 32:33–32:39
the hazards can be on and the brake lights work completely normally.
The Senseless Ambiguity of North American Turn Signals · ▶ 5:24–5:28

这些热还有能量,为什么不能无限做同样多的计算?

套层计算机的假设把内层排出的热作为外层输入。需要同时标注能量流和温度差,因为外层不能按照内层完全相同的条件工作。想把热以更低温度辐射出去,又会遇到散热面积。为这张图每一层补上温度、表面积和一个现实中尚未建造的假设,比只写「余热被循环」更能看清思想实验的边界。

原文与时间戳 · 2
the next layer absorb that waste heat and run more calculations
Matrioshka Brains: Star-Powered Computers · ▶ 3:27–3:37
a sphere’s surface area goes with the square of its width
Matrioshka Brains: Star-Powered Computers · ▶ 6:26–6:30

唱针读到的究竟是振动,还是随沟槽改变的电容?

第一眼看到影碟和唱针,很容易照搬留声机的解释。CED 的读出器却把电极与沟槽形成的电容变化当作信号。把类比分成三步检查:接触部件怎样运动,哪种物理量被转换,哪一段电路恢复图像。外形能提示要找什么,却不能替代第二步的测量。

原文与时间戳 · 2
a titanium electrode which senses changes in capacitance
The CED: RCA's Very Late, Very Weird Video Gamble (Pt. 1) · ▶ 17:07–17:12
brought about by the depth of the groove.
The CED: RCA's Very Late, Very Weird Video Gamble (Pt. 1) · ▶ 17:12–17:15

每一刻只有一道光或一种色,图像怎样连起来?

普通 CRT 让一束光在时间上依次写出不同位置的亮度。JVC 的方案再加一个维度,把几幅单色分量画面按顺序送出,让不同颜色的液晶快门配合。RGB 演示则换一个问题:同名的颜色也可能来自不同光谱。先说明眼前的颜色由哪些时刻与波长组成,才知道相同的颜色名称是否意味着相同的输入。

原文与时间戳 · 6
starts the next line. But remember, it skipped a line.
Lines of Light: How Analog Television Works · ▶ 10:31–10:36
The higher the line goes, the brighter that part of the scan line
Lines of Light: How Analog Television Works · ▶ 13:23–13:27
Liquid Crystal Color Shutter.
LCCS: The LCD / CRT Hybrid from JVC · ▶ 1:11–1:13
three monochrome images in sequence
LCCS: The LCD / CRT Hybrid from JVC · ▶ 4:01–4:03
non-spectral colors.
The Weird World in RGB · ▶ 15:04–15:08
if I light the room I’m in with dichromatic magenta light
The Weird World in RGB · ▶ 20:03–20:08

为什么同一块橙色会变成棕色?为什么亮度能变成距离?

Watson 把橙色方块的背景从黑渐变到白,观众可以盯着方块,判断颜色名称何时变化。Krauss 在另一个问题上需要另一类参照:天体看起来暗,既可能是因为它远,也可能是它本身不够亮。他用标准烛光讲怎样引入已知的本征亮度。这两段的机制有意分开:前者改变视觉背景来观察知觉,后者校准光源以推算距离。共同的读法是先把参照项写进问题里。

原文与时间戳 · 4
It’s orange, with context.
Brown; color is weird · ▶ 3:24–3:26
Let’s fade from black and into white.
Brown; color is weird · ▶ 8:31–8:33
a standard candle something whose intrinsic brightness we understand
'A Universe From Nothing' by Lawrence Krauss, AAI 2009 · ▶ 14:06–14:13
these exploding stars are great standard candles
'A Universe From Nothing' by Lawrence Krauss, AAI 2009 · ▶ 17:34–17:40

两个群体的平均值,能替你读出一个人的值吗?

在对 Damore 的访谈中,主持人描述了备忘录里一幅带重叠区域的分布图。可以把争论暂时放下,做一个只涉及读图的检验:在重叠区画一个点,遮住颜色,再问仅凭横轴位置能否把它唯一分到某一组。即使这张图的均值已知,作出个体判断仍需要更多信息。这个练习不依赖两位讲者对差异成因或公司政策的判断。

原文与时间戳 · 2
the amount of overlap
2017/08/08: James Damore and his Google Memo on Diversity (complete) · ▶ 30:54–30:55
what happens if the distribution is ignored
2017/08/08: James Damore and his Google Memo on Diversity (complete) · ▶ 31:16–31:17

点亮一个点,与指向一个像素

Watson 的节日灯实验先改变发光材料,再给暖白光加有色外罩。看显像管时,红绿蓝荧光点也各自限制可发出的颜色;模拟信号决定一束电子扫过时究竟照亮哪些点。两段可以用一个问题串起来:眼前能数出的小部件,是输入信息的一个地址,还是产生颜色的一种物理材料?在显像管表面数出多少点,并不能反推出送进去的画面有多少独立像素。

原文与时间戳 · 4
an ordinary warm white LED
LED Experiments: Making Holiday Lights Less Garish · ▶ 2:33–2:38
warm white LEDs with colored glass
LED Experiments: Making Holiday Lights Less Garish · ▶ 5:00–5:04
These are actually called phosphor dots.
These Are Not Pixels: Revisited · ▶ 6:09–6:11
the shape of the image defines which phosphor dots are lit.
These Are Not Pixels: Revisited · ▶ 14:04–14:09

保持滤波器不动,让信号移过来

超外差收音机可以用同一个固定中频滤波器收多个电台。旋钮改动本地振荡频率,混频后让选中的台落在滤波器擅长处理的位置。与其笼统说「换了一个坐标」,不如拿纸画两处位置:调台前后,空中广播的频率在哪里,送入滤波器的目标成分又在哪里?只有后者要保持稳定。

原文与时间戳 · 2
the difference between the two frequencies.
The Superheterodyne Radio: No really, that's its name · ▶ 2:31–2:35
the filter more precise because it is set to a fixed frequency
The Superheterodyne Radio: No really, that's its name · ▶ 6:10–6:14

改口型、改剪辑顺序,观众多知道了什么?

BBC 保持音轨不动,换掉口型,观众报告的音节就可能改变。Passengers 的剪辑实验保持原片素材,换掉观众首先跟随的人物,让一段早已拍好的行动成为待解的谜。两者分别改变感官线索与叙事信息顺序。若要找出是哪一步产生差异,可先只听原音,再逐段列出同一行动之前观众分别知道什么。

原文与时间戳 · 4
the sound hasn't changed
Try this bizarre audio illusion! 👁️👂😮 - BBC · ▶ 0:58–1:02
what we see overrides what we hear
Try this bizarre audio illusion! 👁️👂😮 - BBC · ▶ 1:32–1:38
the story is infused from the start with mystery.
Passengers, Rearranged · ▶ 4:52–4:58
from Lawrence's perspective
Passengers, Rearranged · ▶ 5:51–5:54

相差一分,什么时候变成了两类人?

Sapolsky 让学生回想考试里的 65 分和 66 分。如果及格线恰好在中间,两个相邻表现会得到截然不同的名称。把这对分数画在一条连续数轴上,再拿纸盖住及格线:两点的距离变了吗?另一场访谈讨论分布有重叠时,从组名能否推断个体。两种图可以做同一个检查:先保留观测值,再加上分类标签,看解释从哪里开始依赖那条线。

在行为生物学课上,Sapolsky 更进一步:从准确描述行为出发,问行为发生前半秒生物体内有什么活动。这样的追问才开始寻找机制。把观察分成两组或给分数改名,已经能改变判断,却还没有解释一次动作是怎样发生的。

原文与时间戳 · 3
between getting a 65 on a test and a 66 on a test.
1. Introduction to Human Behavioral Biology · ▶ 15:10–15:14
in that organism a half second before that behavior occurred
1. Introduction to Human Behavioral Biology · ▶ 22:48–22:52
what happens if the distribution is ignored
2017/08/08: James Damore and his Google Memo on Diversity (complete) · ▶ 31:16–31:17

参数已经走到一半,路程也走完一半了吗?

在一条 Bézier 曲线上均匀增加 t,看起来像是在匀速前进;换一组控制点,物体却会忽快忽慢。Holmér 把同一段运动换成距离来观察,问题就显出来了:相等的参数间隔不对应相等的路程。她再建立累计长度与 t 的采样表,用目标距离反查参数,然后仍然用原曲线求位置。曲线没有换,改变的是沿曲线前进的刻度。读者可以先预测两个等间隔 t 之间哪段更长,再观察导数的长度;这比只看路径是否平滑更能解释速度变化。查表提供的是近似,增加采样也有计算代价。

原文与时间戳 · 2
equally spaced t values are not equally spaced by distance
The Beauty of Bézier Curves · ▶ 16:01–16:09
generate what's called a lookup table
The Beauty of Bézier Curves · ▶ 17:19–17:24

把熟悉的编程媒介暂时拿走,会想到什么?

Victor 的历史角色扮演把观众暂时放回一种选择尚未固定的处境:程序能否通过直接操作数据形成,目标能否交给求解器,代码与其他信息能否用不同空间视图组织?这些例子并不证明图形表示胜过文本,也不保证一种并行模型适合所有硬件。它们给表示问题补了一层条件:若只学会一种写法,很容易把它当作计算本身。读完可以选一个任务,分别写下现有媒介让什么关系容易表达,以及哪种关系需要绕路;再提出一个替代表示需要保留的约束。演讲里的「四十年后」是 2013 年对 1973 年的回望式表演,不是本文对未来的预测。

原文与时间戳 · 2
it can be hard to adopt new ways of thinking
Bret Victor The Future of Programming · ▶ 1:46–1:54
you stop looking around for other ways of doing things
Bret Victor The Future of Programming · ▶ 31:34–31:38

编辑器看到的是一行字符,还是一棵树?

Paredit 的演示把 Lisp 代码看成一棵可以操作的树。移入或移出邻近表达式时,编辑动作可以围绕分组工作。Guy Steele 在讨论语言论文的记号时也遇到类似的选择:两套替换记号的惯例方向相反,单看括号与斜线无法判断哪一项被替换。

两份材料分别处理编辑动作和阅读动作。试着在同一个例子里标出操作的对象、边界和方向,比较树形结构、文字约定和记号各自能把哪些信息直接交给读者。

原文与时间戳 · 2
as a tree to be manipulated
Emacs Rocks! Episode 14: Paredit · ▶ 0:44–0:53
you're not sure which way the substitution is going
Invited Talk - Guy Steele · ▶ 35:24–35:28

同一份表示能留下什么,又能换掉什么?

道格拉斯·恩格尔巴特用购物清单演示视图控制。同一份清单可以展开层级,也可以按任务只显示一部分。Alex Evans 回顾 Dreams 的早期渲染探索时,描述了另一种分离。造型仍可用距离场表示,研究者可以改试用小斑点来绘制它。

清单例子改变的是读者看见哪些条目,渲染例子改变的是已有造型怎样呈现。比较两种视图前,先指出哪一份数据应当相同,以及改变后的画面需要保留其中的哪些信息。

原文与时间戳 · 2
what we call view control
The Mother of All Demos, presented by Douglas Engelbart (1968) · ▶ 6:03–6:08
finding something that married an art style that stood apart and a technical plan
Alex Evans at Umbra Ignite 2015: Learning From Failure · ▶ 27:11–27:22

走同样一步,视野扩大了多少?

《十的次方》的旁白规定,每一步把视野宽度乘以十。按这个规则,从一米到十米和从十米到一百米各走一步,增加的宽度分别是九米和九十米。影片用同一尺度从野餐处向外看,再向手里的微观世界深入。比较这些画幅时,先分清两件事:每步的倍率是多少,每步增加或减少的实际宽度是多少。

原文与时间戳 · 2
field of view will be 10 times wider
Powers of Ten™ (1977) · ▶ 0:49–0:55
our journey has taken us through 40 powers of 10
Powers of Ten™ (1977) · ▶ 8:24–8:31

把公式移一格,哪个引用会跟着动?

Spolsky 用 R1C1 表示法把行和列的引用分别写清楚。复制公式时,相对引用保留和新位置的距离,绝对引用仍指向原来固定的行或列。他还建议给可能变化的输入起名,让读表的人直接看出公式用了什么。检查公式是否适合复制时,可以先指出哪些输入应当随新位置变化,哪些在复制后还应来自同一个地方。

原文与时间戳 · 2
both the row and the column can be relative or absolute
You Suck at Excel - Joel Spolsky (Reupload) · ▶ 10:50–10:56
give it a name and refer to it by name
You Suck at Excel - Joel Spolsky (Reupload) · ▶ 43:38–43:44

把画面写成一个函数以后,哪里还可以单独改?

Quilez 在风景篇把地形、光照和颜色拆开,最后合成一个从画布坐标得到颜色的函数。在人物篇,他把造型、着色和动画分段讲解,并说明实际创作时几方面经常一起推进。两篇合读,可以问一个具体问题:改的是形状、光照还是颜色?找到负责这一步的部分之后,还要回到合成结果,观察改动如何与其他部分一起起作用。

原文与时间戳 · 2
composed into one large and singular function
I painted a Landscape with Mathematics · ▶ 40:52–41:00
normally all the aspects of the painting do progress at the same time
Painting a Character with Maths · ▶ 9:38–9:47

轴转了一圈,刻度读出的数从哪来?

机械计算教学片先给轴转一圈规定一个数值,再用齿轮和其他机构改变或组合这种运动。旁白也提醒,算出的结果还取决于输入资料的质量。读一个机械计算例子时,可以沿着同一条链追问:物理运动被赋予了什么数值,中途做了哪项运算,最后的输出能追溯到哪项输入?这样可以把机构按规则运转和输入能否代表现实分别检查。

原文与时间戳 · 2
One revolution of a shaft is assigned a numerical value.
Mechanical Computer (All Parts) - Basic Mechanisms In Fire Control Computers · ▶ 2:20–2:27
computer accuracy depends on the quality of the information it receives
Mechanical Computer (All Parts) - Basic Mechanisms In Fire Control Computers · ▶ 1:34–1:44

读完再想一步

把同一批数据画两次。两张图各能回答什么,又各自不能回答什么?