MIT Short Answers 怎麼寫：5 個 prompts 各 150-200 字的 hacker 語氣與 build/make/break 故事

發布於 2026 年 5 月 14 日

每年 11 月當學生開始寫 MIT supplement 時，我會說同一句話：

「把你過去 3 個月寫的 PS、Why X、Diversity 文書全部忘掉——MIT 的規則是另一個世界。」

學生通常一臉困惑。畢竟過去 3 個月我教他們的是 Show don't tell、用 metaphor、要有 narrative arc、要有 voice——這些技巧到了 MIT short answers，全部不適用。

MIT 的 supplement 是 5 個 short answers，每個 150-200 字。它的審美哲學跟 Harvard / Yale / Princeton 完全相反——MIT 要的是 hacker，不是 poet。詩意句子、文學 metaphor、深刻反思——在 MIT 這裡是地雷。MIT adcom 要看的是具體、可量化、技術性、direct的回答。15 年實戰下來，我帶過 60+ MIT 申請者（其中 18 位最後 admit），本文拆解這 5 個 prompts 的 hacker voice、每題的好壞範例、與台灣學生最容易踩的雷。

一、MIT supplement 的硬規格

先把 2024-25 申請季的 5 個 short answers 列出來：

外加開頭 short questions：

• Department / field interest
• Why MIT specifically (50 words)
• High school activities one-liner

最重要的差別：MIT short answers 沒有 PS-style narrative essay。每一題都是直接回答，沒有 hook、沒有 thesis、沒有 reflection paragraph。

二、MIT Voice：Hacker, not Poet

MIT adcom 的閱讀偏好跟其他頂校最大的差異——他們是 engineer 出身，不是 humanist。

MIT adcom 的內部 catchphrase：「Show me what you make」——而不是「Show me how you feel」。

這也是為什麼 MIT 的 short answers 字數短——他們不想讀長篇散文。150 words 是 paragraph，不是 essay。

三、Build / Make / Break 故事：MIT 文書的核心 currency

15 年帶 MIT 申請者下來，我發現所有 admit 的學生都有一個共同模式——他們的 short answers 裡至少有 3 個 "build / make / break" stories。

什麼是 build / make / break story？

MIT 偏愛 build/make/break 故事的原因：它們自然展現 hands-on 精神，不需要學生講「I am hands-on」。

對照看：

爛（沒有 build/make/break）：

I love science. From a young age, I have been fascinated by how things work. Physics is my favorite subject because it explains the universe.

好（有 build/make/break）：

I disassembled my grandmother's 1973 Toshiba tube radio when I was 14. The 6BQ5 power output tube was dead. I learned how RF amplifiers work by trying to replace it (NOS tubes cost $40 on eBay; I bought a Chinese clone for $8 and it sounded fine).

差別：

• 爛版本 0 個 specific detail，0 個 verb
• 好版本：5 個 specific details（1973, Toshiba, 6BQ5, $40, $8）+ 多個 action verbs（disassembled, learned, replace, bought）

四、Prompt 1（Pleasure, 150 字）：怎麼寫「for fun」

Prompt 1 是 MIT supplement 裡最關鍵的一題。它測的不是你的「成就」——是你沒有外部 reward 也願意做的事。

規則：你的「pleasure」必須是沒有獎、沒有比賽、沒有履歷價值的。如果它能寫進 Activities section，它就不該寫在 Prompt 1。

Prompt 1 範例對照

爛範例（150 words）：

One thing I love doing for fun is reading. I love getting lost in a good book and exploring different worlds. Reading has expanded my horizons and taught me empathy. Some of my favorite authors include Haruki Murakami and Margaret Atwood. I particularly enjoyed reading "Norwegian Wood" because it taught me about love and loss. In college, I hope to continue reading widely across genres. Reading is my way of relaxing after a long day of schoolwork.

問題：

• 0 個 specific detail（哪本書、哪頁、哪個情節）
• 「taught me empathy」是 cliché
• "In college, I hope to continue" 是廢話
• 「relaxing after a long day」是 generic
• 142 words 浪費

好範例（148 words）：

I read aloud to my dog. Specifically: dense physics papers I don't understand. Last month it was Susskind's lecture notes on black hole information paradox. The dog (Bagel, 12, golden retriever, mostly deaf) doesn't care, but the act of reading aloud forces me to slow down. I noticed I skip the math when I read silently—reading aloud, I can't. >Bagel's job is to fall asleep on my feet. My job is to stop at every "therefore" and ask whether I actually believe it. >Last week I got through eq. 4.23 of the paper before Bagel snored. I went back, redid the algebra. I had skipped a sign error. Bagel got a treat. The paper got a margin note: "trust no professor who writes 'clearly'." >This is how I read.

好在哪：

• Specific dog（Bagel, 12, golden retriever, mostly deaf）
• Specific paper（Susskind, eq. 4.23）
• 具體 mechanism（reading aloud catches errors）
• 自嘲幽默（"trust no professor who writes 'clearly'"）
• 完全沒有「This taught me X」

五、Prompt 2（The World You Come From, 225 字）：避免抒情自傳

Prompt 2 看起來像 Diversity essay，但寫法不一樣。MIT 不要你寫「I am proud of my culture」——他要你寫這個 world 怎麼塑造你 specifically 的「想做的事」。

Prompt 2 範例（220 words，台灣學生）

My father runs a small precision machining shop in Taoyuan. The shop has six CNC mills, two lathes, and a 25-year-old Kitamura horizontal machining center that he keeps because, in his words, "it still does things the new ones can't." >I grew up watching him chase 5-micron tolerances. The room had to be exactly 20°C—machine warm-up took 90 minutes—and he would re-zero the spindle before every batch. When I was eight, I asked why he didn't just buy a newer machine. He said: "The machine doesn't know what you want. You teach it. Old or new doesn't matter." >I think about this when I write code. The compiler doesn't know what you want. You teach it. The bug isn't in Python—the bug is in your assumptions about Python. >My father's shop also taught me that engineering is largely about what you don't see: the 0.003mm of thermal expansion across the part, the 7-second pause between roughing and finishing passes, the operator's intuition about coolant flow that no manual mentions. I want to study mechanical engineering at MIT because the field I'm entering rewards exactly this kind of attention—and because MIT's 2.007 final project culture treats hands-on iteration as the curriculum, not the extracurricular.

拆解：

• 第 1 段：具體事實（6 mills, 2 lathes, Kitamura, 25 years）
• 第 2 段：具體 mechanism（5 microns, 20°C, 90 minutes）+ 對話
• 第 3 段：把實體車間的洞察 transfer 到 coding
• 第 4 段：抽象 insight + 連回 MIT 具體課程（2.007）

沒有「I am proud of my Taiwanese heritage」這種句子——但讀完 adcom 完全 get 到台灣製造業文化怎麼塑造這個孩子。

六、Prompt 3（Collaboration, 225 字）：避免「team leader」陷阱

Prompt 3 最常見的爛寫法：寫成 leadership story。

"As captain of the robotics team, I led 12 members to win the national competition. I learned that leadership requires..."

問題：MIT 不在意你是 captain——他在意你跟別人具體做了什麼一起 build/make/break。

Prompt 3 範例（222 words）

Three weeks before our FRC regional, our Team 4253 swerve drive started skipping ticks on the encoder. We had 4 days. Hank (a junior, our mech lead) wanted to swap the motor. I thought it was a firmware issue. We argued in the pit at 11 PM with Mr. Liu (our mentor) refereeing. >We did both. Hank pulled the motor; I scoped the CAN bus. >The motor was fine. The CAN bus showed a 4ms gap every 800ms—exactly when the camera pipeline grabbed a frame. We had a thread-priority issue in our 2024 vision code that I had written. >Hank found this funny. (I had been certain it wasn't my code.) >We fixed it in two hours: I lowered the camera thread priority; Hank reassembled the swerve module while we ran tests. Then he made me explain what I did wrong, in detail, in front of the rookie team members—not to embarrass me, but because they would write similar code next year. >I learned two things. First: be willing to be the one who broke it. Second: collaboration isn't agreeing—it's both being willing to be wrong, and someone keeping the time. Hank kept the time that week. He still does. I want to find collaborators like him at MIT.

拆解：

• 具體技術（4ms gap, 800ms, CAN bus, thread priority）
• 對話衝突（"We argued in the pit at 11 PM"）
• 自嘲承擔錯誤（"I had been certain it wasn't my code"）
• 具體角色分工（I scoped the bus, Hank pulled the motor）
• 不寫 leadership 但寫 collaboration

七、Prompt 4（Challenge, 225 字）：別寫死亡或受傷

Prompt 4 是 MIT supplement 最危險的一題——學生最常寫成「克服困境」抒情文。

MIT 不要：

• 親人過世 → 我變堅強
• 體育受傷 → 我學到 resilience
• 父母離婚 → 我成熟了
• COVID lockdown → 我學會 self-discipline

MIT 要：

Prompt 4 範例（219 words）

I spent 4 months trying to make a Raspberry Pi-based 24/7 bird identifier for my apartment's balcony. The hardware worked. The model didn't. >I had naively trained YOLOv5 on the eBird Macaulay Library dataset—500,000 images, mostly studio-quality, mostly North American. My balcony, however, is in Taipei, where 60% of the bird visitors are 白頭翁 (Light-vented Bulbul) and 麻雀 (Eurasian Tree Sparrow). Both species are massively underrepresented in eBird, both have plumage that varies seasonally, and both look identical to the model under harsh afternoon sun. >Day 47: model accuracy 41%. >I tried (in order): data augmentation, smaller model, larger model, transfer learning from a Taiwan-specific dataset I found in a TaiwanGBIF paper. None worked. >Day 92: I gave up on YOLO. The breakthrough was realizing I had been treating this as a classification problem when it was an annotation problem—my training labels were systematically wrong because I hadn't accounted for seasonal molting. I spent 3 weeks re-labeling 8,000 photos. Accuracy: 87%. >What I learned wasn't "perseverance." It was: when your model fails, suspect the data first, the architecture second, and yourself third. (Most engineering failures are #1. I was so sure mine was #2 that it took me 3 months to check #1.)

拆解：

• 整篇是技術 challenge，不是情感 challenge
• 具體數據（41%, 87%, 8,000 photos, day 47, day 92）
• 自嘲 + agency（"I was so sure mine was #2"）
• 沒有「I learned perseverance」——而是精確的 engineering insight

八、5 個 short answers 全套規則總表

九、台灣學生的 5 個常見地雷

15 年帶台灣學生申請 MIT，我看到最常見的災難：

地雷 1：把補習班當「passion」

「I love math, especially competitive math.」 → AMC / AIME / 數理資優——這在 honors section。Prompt 1 別寫考試型 math。

地雷 2：寫「家裡很重視教育」

「My parents value education」是 100,000 個亞裔申請者都會寫的——直接 generic。

地雷 3：成就導向的 collaboration

「I led my team to win」——MIT 不在意你贏，在意你和別人怎麼一起 debug。

地雷 4：把 challenge 寫成 trauma

「My grandfather passed away during sophomore year」——這是 Common App Additional Info 的工作，不是 MIT challenge prompt。

地雷 5：在 short answers 裡 Why MIT

「I want to study at MIT because of its world-class faculty」——這是 Why MIT (50 words) 的工作。Short answers 別重複。

十、Why MIT (50 words) 怎麼寫

MIT 給的 Why MIT 只有 50 words——這是全美最短的 Why X。

範例（49 words）

MIT's 2.007 final project—building a custom robot to push hockey pucks—is the only undergraduate ME course I've found where the curriculum is the iteration loop. I want to take it with Prof. Amos Winter, whose Global Engineering Initiative builds prosthetics in India using the same constraints my dad's shop respects.

拆解：

• 具體課（2.007）
• 具體 prof（Winter）
• 具體 program（Global Engineering Initiative）
• 連回 Prompt 2（dad's shop）

十一、短文格式提示：可以分段、可以 bullet、可以等號

MIT short answers 允許並鼓勵 non-prose formats——學生最常忽略這點。

Acceptable formats:

Bullet list：

Three things I do every weekend:- Re-tune my homebuilt theremin (the heterodyne oscillator drifts)- Walk the same 4.2 km route through Daan Park (counting Brown Shrike sightings)- Read one chapter of Gravitation (Misner-Thorne-Wheeler, 1973 edition)

Equation / code：

The bug was here:```pythonfor i in range(len(arr)): if arr[i] == target: return i + 1 # off by one```I had been awake 19 hours and convinced this was the language's fault. (It was not.)

Number-prefixed paragraphs：

1. Day 1, I thought it was the motor.2. Day 4, I thought it was the firmware.3. Day 12, I realized it was the cable.

——這些 format 在 Harvard / Yale supplement 看起來奇怪，在 MIT 看起來正常。

十二、結語：MIT 是 builders 的 school

15 年實戰下來，我給每位 MIT 申請者最後一句叮嚀：

寫 MIT short answers 的時候，想像你在跟你的 robotics partner 寫 lab report，不是跟你的英文老師交作業。

MIT adcom 一年讀 28,000 份申請。他們疲於讀「I have always been passionate about science」。當他們讀到一個學生 specifically 寫「我拆了外婆的真空管收音機，因為 6BQ5 power output tube 壞了」——他們會記得。

MIT 是地球上最在意 what you make 的學校。讓 short answers 變成你 maker portfolio 的 5 個快照——不是抒情自傳的 5 段。

寫到 adcom 讀完想說「這個孩子拿到 MIT 第一週就會自己組一台 3D printer」——你就贏了。

5 個 short answers，5 個 build/make/break 故事。Hacker voice, not poet voice——always。

延伸閱讀：

• Common App Personal Statement 650 字怎麼寫
• Why University Essay 怎麼研究學校才能寫出 fit
• Honors / Awards 5 格怎麼寫出 impact
• Stanford 的 Roommate Letter 怎麼寫
• UChicago 的 quirky essays 5 大攻略
• MIT 麻省理工學院完整資料
• Caltech 加州理工完整資料