Self-Directed Learning: Science and Technology -- Anti-Involution Survival Engineering for University

Zero: Schooling vs. Learning

The Brutal Reality

The Shanghai Jiao Tong University Survival Handbook (《上海交通大学生存手册》) opens by laying bare the state of Chinese higher education:

"Undergraduate teaching at the vast majority of universities in this country is not on the verge of collapse -- it collapsed long ago."

Eighteen years later, that line still holds.

The department's goal is to secure research funding and build up some "XX discipline." That is how the department survives. As for undergraduates -- what does whether you learned anything have to do with their funding? Whether you sink or swim is nobody's problem. Even if somebody did care, the cost-benefit of looking after master's and PhD students would still be better.

For university teachers, teaching is a "side gig," even a "chore." Their energy all goes into research. Only good research wins funding and titles. Teaching -- especially teaching undergraduates -- is a thankless job: do it well, no reward; do it badly, punishment.

Even "lecturers" without heavy research loads -- don't assume they will teach carefully. Teaching earnestly earns the same salary as phoning it in, so why not slack off? As for whether the students understood anything, what business is that of theirs?

From PPTs older than we are to soporific lectures, from cookie-cutter exams to course content recycled from ten years ago -- that line has kept getting confirmed everywhere for eighteen years and counting.

So if you run into a teacher who is patient and thorough in class, shares the frontier of the field, and tells you how the industry actually stands -- treasure them. They are a scarce resource on campus.

Now, about the training plans. Let me throw out a jaw-dropping conclusion first: the curriculum of university classrooms is at least 5 years behind the industry, and 10 or even 20 years behind is commonplace.

Take big data as an example. Big data started drawing attention from top internet companies around 2008; from 2013 it drew wide attention from government, capital, and media; in 2014 it was written into the Government Work Report for the first time; from 2016 its applications exploded across every major field; and in 2021 the Data Security Law and related regulations came out, tightening oversight. By now, "big data" is hardly an unfamiliar word to us.

The corresponding major, Data Science and Big Data Technology, was approved by the Ministry of Education in 2016, with Peking University, Central South University, and others in the first batch. The earliest its first undergraduates could graduate was 2020; master's students, 2023.

The students of this major perfectly missed the industry's dividend period.

And this is a brand-new major. For those programs with centuries of accumulated science and technology, the lag is far worse. Much of the content is stuck in the early 2000s; the teachers themselves have never touched the industry frontier, and when they have, it goes no further than inviting some "big names" to campus for a lecture.

I once screenshotted a passage from a PPT and showed it to an industry heavyweight. His very first sentence was that half of what that teacher said was wrong, and that they had absolutely never worked on the front line.

If the teachers themselves don't know where the industry stands, how could they possibly avoid producing outdated training plans?

What's more, several hundred students in a major share one training plan, while everyone's minds, knowledge backgrounds, and receptiveness differ. How could that plan possibly work effectively for every single person?

It is like the cafeteria frying one giant wok of food: either undercooked or scorched, never evenly fried. Teachers can only pursue the group optimum; the individual optimum is out of reach.

Here we have to talk about something: the "innovation class." Many 985 universities love to run such classes -- "Excellence Classes," "Academician Classes" -- some even establish a whole college just to herd every "innovation class" on campus into one place.

In their promotional material, these classes come with labels like "integrated bachelor's-master's-PhD track," "advisor system," and "small-class teaching." The favorable policies may be real, but set against the dark side behind them, those perks are, at best, insignificant.

Under the glossy surface of these classes, there are several giant pits:

1. GPA "deflation"

In an innovation class, the homework load is 2x an ordinary class and the exams 3x harder, but the GPA ceiling is still a 4.0. When it comes to graduate recommendation and study-abroad applications, teachers at other schools or admissions officers abroad will not care whether you took "Advanced Quantum Mechanics (all-English instruction)" or plain physics. They only look at numbers. (Your own school is an exception.) The 3.6 you ground out on hell difficulty loses, on the application market, to the breezy 3.8 of an ordinary class. It is a trade in which effort and reward are grossly mismatched.

2. An involution battle royale

Innovation classes, especially in hot majors, gather grind kings (卷王) from every corner of the country. In an ordinary class you might reach the upper-middle with little pressure; here you can only serve as a foil for the grind gods and honestly play the denominator.

Last semester I chatted with a grinder who had transferred into a good class of a hot major. They told me the grinders there average 92+ in every course, snatch national scholarships at will, carry competition certificates thicker than a dictionary, and do not even spare volunteering and activities.

"Even the sports-meet events in our class get fought over, because they add conduct points." They even grind places with near-zero marginal effect.

So if you are not doing well in such a class, it is not that you are incapable -- you were placed in the wrong coordinate system.

3. Research consumables reserve

The department's goal is not to train you into an outstanding engineer or an industry leader; it is to lock you onto the "PhD-at-this-school" track as cheap research labor. Once you discover mid-way that research is not for you, you will find that your dragon-slaying skills are useless on the job market.

Research, too, is a cruel competition where "the success of one general is built upon ten thousand withered bones." Even at the very top of your class, you cannot guarantee that you will be that "one general."

They promise to train you into a scientist; in reality they are training you into a consumable on the research assembly line. The department does not care whether you live or die -- only whether papers and results come out. As for who happens to produce those results: if you won't do it, someone else will.

4. No exit mechanism

Many "innovation classes" run last-place elimination, but almost none offer a voluntary exit. Even if you are not at the bottom, trying to leave or transfer into an ordinary class means enormous administrative pressure and psychological shame. Those who quit are treated as failures; those who stay can only be denominators for the grind kings. So the students who were never happy to begin with sink deeper into depression under the double pressure. One road, blocked at both ends.

The Diminishing Marginal Returns of GPA Grinding

GPA is cash, not an asset.

As your GPA climbs higher, you will notice the cost-effectiveness of grinding it keeps dropping.

The grind kings in hot majors at good schools all average 90+. Suppose your average is 80. The time and energy you spend holding your 80, versus what a 90+ grinder spends holding theirs, are not even in the same weight class.

GPA's usefulness is limited to graduate recommendation and study abroad. But graduate recommendation looks at your GPA ranking, not the GPA itself, and for study abroad you can keep retaking courses to "farm" your GPA.

Between score bands, things are simply different.

• 60-80: this band is the "foundation zone." As long as you master the basic concepts and memorize the formulas, you won't fail, and with some luck you will even get a decent-looking score.
• 80-90: this band is the "improvement zone." Here you need to nail the exam's key points and common traps, and the time invested slowly grows.
• 90-100: this band is the "bubble zone." For one or two extra points you may have to memorize the entire book, or even psychoanalyze the question-setter.

Going from 65 to 70 takes maybe 5 extra hours; going from 90 to 95 might not be enough even with 50. And once those 5 points get weighted into the GPA, the effect may show up as a mere 0.2.

Crazier still: for those crumbs of points, people will not even spare the class-rep post of a fluff course -- another place with near-zero marginal effect. To keep others from getting that crumb, they would rather spend ten times the time and energy.

That 0.2 might matter right at the graduate-recommendation cutoff, but in recommendation interviews and graduate-entrance re-examinations its payoff is 0. Teachers do not care what makes a 95 stronger than a 90; but at one glance they can tell whether you are someone who has "completed a real project" or someone who "only knows how to do problem sets."

If someone treats the policy's scoring rubric as gospel, then they are the biggest victim of that very policy. And the reality is that in hot majors, everyone fights bloody battles in the "bubble zone" just to avoid falling behind -- even the "winners" of this system cannot escape becoming its victims.

After all, exams are the game we know best: clear rules, comfortable and safe; you just keep playing the high-school way. Everyone crowding onto one single-log bridge -- of course it turns into a grind.

The rational strategy is to "cut your losses": once you reach your target line, stop the extra investment and merely "hold steady" so you do not fall out of the safe zone.

How to hold steady is what the rest of this document covers.

Is Attending Lectures the Same as Learning?

Schooling and learning are two different concepts.

Our learning path from childhood on has been "preview -- attend class -- take notes -- do homework -- take exams." Along this path we sit in the position of passive reception; nothing is actively constructed.

From elementary school to middle school to high school, and then university, we have repeated this path over and over. For the first 18 years we had no choice. But now we have the internet, we have AI, and oceans of usable resources are waiting for us.

Having read everything above, you probably already believe that university teachers have no intention of teaching seriously, and that we can absolutely choose to self-study.

But for many people, so-called "self-study" merely means finding an online course and re-watching the lecture they failed to understand offline, online. They still have not left the passive-reception path.

Why are we so attached to this illusion? Because it delivers a lethal hit of pleasure.

Whether in class or watching an online course, your thoughts follow the teacher's logic. The derivations go smoothly, the worked examples seem doable, everything feels fine.

In reality, you are only being "carried along." Those roads were walked once with the teacher leading you -- you never truly walked them yourself. You feel like you understood; that feeling is just the comfort of a brain massage.

Ask yourself to restate what was just covered, or to solve one problem, and you will stall instantly. Because you never actually owned that logic.

Input only, no output: it clicks during the lecture, it dies on the problem set. Using lectures to dodge output is nothing but a cowardly escape.

Real "learning" is painful; it is full of chaos and uncertainty, like tearing muscle at the gym so it can regrow. Lectures are comfortable and full of certainty.

Most people would rather read the book ten times than derive a formula once; would rather rewatch an online course five times than solve one problem independently.

And this inertia -- "chase certainty," "avoid pain," "performative diligence" -- is precisely the trap the schooling system was carefully designed to spring on you.

Schooling is a "finite game." The goal is out-grinding others or satisfying metrics. Everyone grinds, so you get swept into grinding too; it is kill-or-be-killed, the evaluation system is single, and the game ends the moment the results come out.

Learning is an "infinite game." The goal is building cognition and solving problems, driven by curiosity and passion. The evaluation system is diverse and immediate, and the game's greatest purpose is to keep the game going.

So "schooling" is merely a job we are forced to do, while "learning" is the cause worth investing in.

Yet what we face is a zero-sum situation: school not only refuses to teach you anything useful, it also tries to occupy all of your time, so that you have none left for real learning.

Just look at school: meaningless activities, performative group projects, dull team-building. All of them plunder your attention, keeping you running on fumes, unable to settle down and construct knowledge actively.

So for the sake of learning, we must resist the school's existing model.

De-Studentization

As the last stop before entering society, what university should be doing is "de-studentizing" you.

The dean once shared an insight with me: the time undergraduates spend on coursework should decrease with each year of study, and the time freed up should go toward "things that can be written into a resume."

Graduates I have interviewed told me that what you learn in university is less than a tenth of what jobs require -- and that is engineering majors who found jobs in their own field.

HR sorts students into three tiers: those with hardcore project experience; those with strong all-around ability but high training costs; and those with nothing but GPA and student-leader titles.

You graduate at 22. If the only new things you can put on a resume are course scores, then those 4 years were a blank page.

So what can be written into a resume? How should the 4 years be divided?

• Freshman year: everyone is a newbie at this point and cannot do much beyond coursework. Put 80% of your time into courses like Advanced Math (calculus) and College Physics, and grind those foundational scores as high as possible. Courses like calculus, physics, and probability theory carry a lot of credits; chewing them down builds a huge GPA cushion. A brutal truth: later major courses do not carry that many credits, so your freshman GPA ranking is almost your GPA ranking for the entire 4 years. So the first year deserves extra sweat on coursework, leaving room for error in the following 2 years.

• Sophomore year: the time spent on coursework should drop below 60% of the required course hours you must sit exams for. After a year of feeling your way around campus, you have some knowledge base, some feel for how exams work, and some information channels. The focus now is "exploration." Contact advisors and join a research group, form a team for competitions, get into clubs, learn skills, or -- like me -- write a practical PDF document. In short, you must have output, achievements beyond coursework, even if the results themselves do not look pretty.

• Junior year: this is the year to settle your direction. Employment, graduate-entrance exams, graduate recommendation, study abroad, even entrepreneurship -- every road is laid out in front of you, and once you choose, you go all in. For employment, boldly hunt internships; for entrance exams, burrow into the library and grind problems; for further study, attack competitions and publish papers; for entrepreneurship, run your project like crazy. Do not imagine you can bloom on multiple fronts -- most people struggle to walk even one road well. Courses at this stage are mostly whatever; maybe one of them carries high credits. If you are still, like a freshman, afraid to skip class for the full-attendance points or staying up late to make your homework excellent, then you are a "strategic giant baby."

• Senior year: everyone scatters onto their own paths. Courses are almost gone; each person drifts farther along their own road, collecting the results that belong to them. Nothing more to say.

A pretty GPA with nothing else shining looks perfect but is extremely fragile. When the people competing on the same stage as you all hold transcripts averaging 90+, how exactly are you going to stand out?

Coursework is your "cash flow" -- keeping it above the safety line is enough. Projects and competitions are your "assets" -- they decide your future worth. As the years go up, you must frantically convert cash flow into assets, instead of hugging a pile of depreciating cash (GPA) and feeling smug about it.

Beyond Review

In university, there is no relationship between learning well and testing well.

Since schooling is just a job we cannot avoid, our mindset should shift: trade the shortest time and the least energy for the expected score.

I did the math: this semester I spent about 45 hours on "schooling," while the required courses with exams totaled 162 class hours. On less than a third of that time, I still managed 80+ in every course and held first place in GPA.

So everything below rests on one assumption: you do not plan to attend lectures and have decided to self-study the whole way through. You want to free up that 2/3 of your time for the things you actually want to do, no matter how heretical those things look in the eyes of the outside world.

You do not need to learn everything the way you did in high school.

University finals are nothing like high school exams. High school is selective testing; university is qualifying testing. University exams do not exist to filter geniuses but to verify qualified products. Our goal is not a perfect 100 but the best value for effort: 85+, or just passing.

So from the very start, drop the high-school playbook of "grind through problem sets furiously." In university you have time, you have the power to choose, and you can freely reach the internet and AI. Using every tool you can reach to maximize efficiency is how this game is played. It is exactly this playbook that got me first place in major GPA even while cramming.

Many courses were designed from the start to allow cramming.

If you do not believe it, search Bilibili for a speedrun course of that class. The vast majority of public courses and a good share of major courses can be found. But have you ever wondered: teachers keep insisting you study steadily all semester, so why do these speedrun courses rack up several million views? Is it really just because most students do not study during the term?

At minimum, such courses prove that learning an entire course in a few hours is entirely feasible.

Maybe you resent teachers for "watering down" their lectures, hate the baffling homework, and only discover near the final that you have learned nothing. No need to panic: high school exams are a marathon, while university is interval running.

High school requires steady accumulation all term long; university requires yanking your memory to its peak right before the exam.

High school knowledge points are simple, but the question types vary endlessly and the tricks abound, which is what makes it "hard." University has many knowledge points but simple question types: what you study is what gets tested, there are few variations, and they generally do not take detours or set traps.

For teachers, failing students also means trouble for themselves. So most teachers will give you as many points as they can, and even if their estimate of your score is off, do not worry too much -- your score usually lands higher than you expected.

So in university, the core skill is not "how to learn well" but "how to test well."

One: The Day-to-Day Work

Burying your head in books is diligence at the tactical level and laziness at the strategic level.

You can observe the high-GPA people around you and sort them roughly into two types: those who grind hard from the first day of term all the way to finals, and those who are exceptionally good at taking exams.

Which type do you choose?

I am not saying burying your head in books is wrong -- only that we have a more efficient way. And I suspect that if you were truly the head-down grinder type, you probably would not have read this far.

"Skipping Class"

There is no positive correlation between attendance and grades -- the correlation may even be negative.

That teachers do not teach seriously is already a fact we all take for granted. But then, what are you doing in the classroom?

Listening carefully? You will learn nothing. If you had digested the content above, this thought would not even occur to you.

Scrolling your phone, slacking off? Look up and glance around: most of your classmates have their heads down, staring at that glowing screen. Watching Douyin, playing Honor of Kings, or doing something else. They are merely driven by the instinct of "don't want to listen," with zero concept of "why not listen."

Body in prison, mind in exile.

Doing your own thing? Sounds good -- but doing what? Homework? Your own online courses? The PPT for a group project? Or your own project?

If you are certain that the knowledge density of the teacher's lecture is lower than the knowledge density of reading the material yourself, then skip without hesitation, or practice "invisible skipping": your body sits in the classroom while you do your own thing. Body checked in, soul offline.

I personally support skipping class, and as class monitor I kept the cost of "legal skipping" -- taking leave -- remarkably low. Write yourself a leave note, give me a heads-up, no signatures needed, and the leave was granted.

There are far, far more interesting things in this world than sitting in a lecture.

As the saying goes: if you skip, don't fear; if you fear, don't skip. Skipping class is an art of risk management -- while your body escapes, minimize your loss of routine points as much as possible.

Figure out the teacher's roll-call pattern, have your roommates answer for you, sign the sheet for you. When there is a policy from above, there is a countermeasure from below. If you want to skip, there is no class you cannot skip.

Only show up when necessary, and even when you do show up, decisively do your own thing. The classroom is our "miscellaneous-tasks processing center."

But what about in-class roll calls?

If you choose physical skipping: treat your roommates to a meal in advance, give them a little benefit, state your request clearly, and ask them for the favor. A mouth that has eaten goes short, a hand that has taken goes soft.

If you choose mental skipping: just saying "I don't know" will not hurt you. You feel mortified, but most people will forget your awkward moment right after class. If you are truly terrified, mentally check out during the roll call and use AI to generate an answer.

Even if you do get marked at a roll call, that is merely the price of 90 minutes of freely disposable time. When a lecture's payoff is far below its opportunity cost, not skipping is the real waste.

Skipping class is not a "bad student" label -- it is an investment, and losing routine points is merely the penalty clause of the deal.

Let me also share a method for defeating Rain Classroom (雨课堂) random roll calls:

Scan the QR code to check in as normal, but during roll-call windows, disconnect every one of your devices from the network -- on a computer you can log out of WeChat. The core is keeping WeChat offline, and then you will not get picked. This trick works great in big lectures with huge enrollment, especially the various politics courses.

As for how to make up what you missed after skipping, that will be covered in detail later.

Impression-Score Management

That most teachers will fish students up at the end does not mean being fished up is owed to you.

If you were the teacher, facing a student you have zero impression of, who never does homework and whose exam is a disaster -- would you feel any urge to fish them up?

Teachers are human too. It is perfectly natural for students they remember, or remember fondly, to receive more routine points and looser grading on the final.

Even if routine points carry little weight, they can decide whether your final score is 59 or 60, whether your GPA is 3.7 or 4.0.

Moreover, compared with grinding through final review, routine points are absurdly easy to earn. Do not look down on these points. Your score is added up point by point -- not deducted point by point.

The detailed rules for routine points may leave little room to maneuver, but rules are dead and people are alive. Leaving a good impression on the teacher never hurts, and it can pay off when you least expect it.

You do not have to listen in every lecture, but in the key sessions -- the first and the last -- visibly make your presence felt.

"Class monitor" and "study commissioner" are two identities that make presence easy to show. Approach the teacher with this identity and they have no reason to turn you away; besides, many teachers can only put names to these two roles.

If you hold neither role, no problem. Use AI in advance to prepare a question that could only come from someone who previewed the material, and ask it right after class. The teacher will think you are "eager to learn" and remember your face and name.

That is fine -- we are asking the question for the sake of asking the question itself.

If you missed that chance, there are still options. Whenever you do attend class, sit in the first row. Even if you are doing something else on your laptop, the teacher will default to "the ones in the front row are the good students."

The emotional value that sitting in the front row delivers to a teacher far exceeds sleeping in the last row.

In the second semester of my freshman year, I sat in the first row of every session of high-credit courses like calculus. Those teachers could all call me by name, and would even often call me up to the board to explain problems.

Learning Rhythm

Why is a university lecture about 90 minutes (45 x 2)?

Because it makes scheduling convenient for the academic affairs office, letting you trek to the teaching building only once or twice a week. That is all.

The school timetable was never designed for better absorption by your brain -- it is merely a compromise with the teachers' free hours. Yet the human attention limit is around 40 minutes. Setting course length right at the limit is like setting off on a road trip with exactly enough gas and zero margin: the slightest detour strands you. If you could genuinely get through a lecture without slacking for a single minute, you probably would not be reading this.

Factor in forgetting, too. When you show up the next Tuesday for Thursday's course, you have already forgotten the previous lecture completely. The teacher has to spend the first 15 minutes reviewing for you, then force-feed you new knowledge with whatever time remains.

Repeat this for a dozen-odd weeks, each time rewriting data into a freshly "formatted" brain. The efficiency is abysmal.

Likewise, different subject types suit different learning rhythms. Memorization-and-recitation content thrives on short, high-frequency stimulation, while mathematical-logical content needs one large block of time for deep immersion.

The school makes no distinction between these two types of courses. It stuffs every course indiscriminately into the 45 x 2 timetable and then expects us to actually learn something.

That alone is absurd enough.

Attending class is a form of "brain massage": the teacher pretends to have explained clearly, the students pretend to have understood, and everyone feels comfortable.

But without forced overload, there is no real consolidation. Everything beneficial is uncomfortable. Chewing through formulas, memorizing provisions, debugging code -- these tasks carry built-in discomfort. And it is exactly that discomfort which lights up new neural connections.

For memorization-and-recitation subjects, such as the various politics courses, our operating logic is "small meals, many times."

Invest about an hour a day, but not in one block: split it into 3 chunks of 20 minutes each, dedicated to reciting this content, repeatedly and frequently stimulating the relevant neurons.

This is relatively relaxed, keeps you from being crushed by so-called "quota" pressure, and since you memorize while attention is at its sharpest, the brain absorbs the most and forgets the least.

Reciting vocabulary once a week for 7 straight hours is self-deception: after the first hour your brain has already refused to write anything in. Reciting 45 minutes a day, split into three sittings, letting those neurons fire over and over -- that is the play that obeys biology.

For highly procedural and structural knowledge, such as calculus, mechanics, and coding, we need the "immersive experience."

Here, any fragmented interruption breaks our continuity -- even standing up for a bathroom trip can wipe out the line of thought you just had.

If you are deriving a complicated mechanics formula, or configuring a machine-learning environment, do not interrupt yourself with the so-called "Pomodoro Technique." This kind of knowledge has an enormous cold-start cost, like bringing water to a boil: if you turn off the flame every few minutes, the water will never boil.

In these moments, what you need is to go to the library, put down your phone, and give yourself an absolutely immersive 3-hour block to reach a flow state.

If you happen to have a free afternoon and want to study something to make finals easier, you do not have to hammer away at one subject -- you can "interleave":

First do about 90 minutes of high-intensity computation (for example, working through College Physics example problems), then rest 10 minutes. You can stand up and move around, use the bathroom, but absolutely no phone.

The moment you open your phone, dopamine levels in your brain spike instantly and it becomes hard to return to a focused state. And by the time you put the phone down and get ready to continue, an hour has probably slipped by.

For the next hour, review memorization-type content: recite some politics, do a couple of English reading passages. This activates the brain's "language" region while the "logic" region rests.

After another 10-minute break, finish with organizing and output. By switching between different subject types, mental fatigue drops dramatically.

Information Channels

The final exam is not only a contest of learning ability -- it is also a contest of information channels.

What information exactly do you need?

1. Textbook: all knowledge and all exam content originate in the textbook, but do not read it word by word. Treat the textbook as a "dictionary" -- look things up wherever you are unclear.
2. PPT: the PPT is a condensed textbook, gathering every knowledge point that could possibly be tested. The teacher's PPT may be older than you are, but that does not stop you from reading it carefully.
3. Notes and exam experience from seniors: if you know seniors with well-made notes, try hard to get their notes, and probe them about the exam's difficulty and review tips. When you ask, send a small red packet as thanks -- do not just say "thank you." Whether they accept is their business; you must show a "no freeloading" attitude.
4. Past exam papers: the most important of all. The vast majority of university final exams do not change for 10 years, especially in STEM. There is a good chance the teacher takes a past problem and changes a few numbers, or swaps in a fresh scenario shell. Even if the question types shift somewhat, you will still know roughly which knowledge points get tested and at what difficulty. If conditions allow, prepare at least 3 sets.
5. Key-points lecture: some teachers devote a whole lecture near the end of term to marking what matters. Even if you have not listened all term, seize this lecture. Listen with total concentration: what is emphasized, what will not be tested, what shows up as small questions, what shows up as big ones. Whatever is still unclear, chase the teacher after class and keep asking -- ideally until the teacher runs out of patience.

With this information in hand, the preparation for review is basically complete. If you are missing some of it, no problem -- we still have AI. Space is limited, so I will not cover it here; how to use AI is a topic we may update later.

None of this means attending lectures during the term is unimportant. If you are aiming for graduate recommendation or study abroad, or you do not want enormous pressure at finals, listening carefully during the term is still very necessary.

Which courses deserve careful listening?

• High-credit courses. As a rule, courses worth 3 or more credits are fairly hard: calculus, linear algebra, College Physics, and so on. The credit values for each course are listed in your college's training plan on its official website -- every major spells them out.
• Courses whose speedrun videos have huge view counts. At the start of term, go online and look for speedrun courses to gauge their popularity, so you know where you stand. These courses are harder and likely need repeated listening to master.

Although I choose the speedrun route, I still study certain key courses seriously. Serious study has another benefit: with a good impression, the teacher may give you more routine points. Everyone knows routine points can drag a 59 up to a 60 -- but what if they could drag an 89 up to a 90, too?

The 90+ formula: careful study during the term + efficient cramming at the end + high-quality information channels.

Distributed Computing

Those 90+ grind kings are your external computers.

The zero-sum gaokao left us no room to cooperate -- we could only fight our way out alone. But in university, everyone has room to cooperate.

Especially those students with upper-middle grades who listen attentively in class, keep detailed notes, and take homework seriously. They are the best resource channels.

Outsiders might think they are grinders, but in reality, most excellent people only care about themselves. Their attention is on "making myself better," not "crushing so-and-so down the rankings."

So at this point, if you volunteer yourself and tell them "I won't drag you down," or even "I have things I can trade with you," they will usually welcome you.

Use your strengths -- an information edge, mastery of AI tools, or a treated meal -- to trade for their notes and homework. I said trade. Trade. A goddamn trade.

Do not open with "hey bro/sis, can I take a look at your notes?" Nobody likes being freeloaded off. They may not refuse, but one-sided value extraction never lasts. Value exchange is the most stable model.

Also do not complain in front of them about "this course is so hard" or "I understand nothing" (unless it is a script you designed). Nobody likes someone who spreads negativity without acting.

Once you have the materials? Scan and copy the notes, copy the homework outright. Where you copy something you do not understand, have AI explain it. In university, copying homework is actually a way to raise efficiency.

The top-ranked student in my major once let slip that he copies homework all the time too, especially in "recitation courses" -- it saves a lot of time.

So where do the top students' "external computers" come from?

Beyond individuals, study groups are also an efficient way to learn. Group projects finish slowly precisely because nobody steps up to lead the room, assign tasks, and push progress.

So you have to be the one who organizes the whole thing. You have to push everyone.

Three to 4 weeks before the exam, gather the upper-middle students, the average ones, and those who just want to pass. No need to invite the grind kings -- they almost certainly have their own review plan and will not want to be interrupted.

Then, for the harder subjects -- analog electronics, theoretical mechanics, say -- each person studies 1 chapter, and 2 days later everyone reconvenes and teaches each other, answering whatever the others cannot. The act of explaining is itself learning.

This way, each person learns 8 chapters in the time it takes to learn 1. An 8x efficiency lever, stronger than going it alone.

That is distributed computing. Others handle front-end data collection; you handle back-end cramming.

Two: Subject Triage -- Know Your Opponent

The school applies a one-size-fits-all cut to course hours, and then applies a one-size-fits-all cut to review methods, too.

Whatever the major, whatever the subject, the review method we are fed is "read the book -- grind problems -- memorize answers." Calculus works that way, and so do the politics courses.

The biggest problem with this model is how crude it is. It assumes every type of subject can be handled the same way, but reality is certainly nothing of the sort.

Subjects differ in type. Some people can easily score 95+ in calculus but barely pass politics; others can effortlessly memorize codes and provisions but blue-screen the moment numbers appear.

Three Types of Subjects

Different subject types call for different review methods. But before choosing a method, we need to classify the subject:

1. Mathematical-logical Mainly tests formulas, reasoning, and computation. Obvious markers: lots of computation on the paper, a single correct answer, and a high share of points for working steps. Typical examples: calculus, linear algebra, mechanics, circuits.

2. Memorization-recitation Mainly tests memory and understanding of key concepts, codes, and provisions. Obvious markers: term explanations, short answers, and essays on the paper; answers are open but with standard scoring points. Typical examples: politics courses, law.

3. Hybrid Shows the features of both types above at once. Obvious markers: term explanations, short-answer questions, and computation problems all appear on the same paper, and the same concept appears in different question types in different ways. Typical examples: economics, and most major courses.

This step matters, especially for hybrid subjects: only when you have the type sorted out can you avoid reviewing like a headless fly.

The subject is hybrid, but each question type is singular. It is rare for one problem to test both your memory of a concept and your computation at once.

If you can get past papers, or the teacher has told you the question types, the first thing to do is judge each type's weight.

Take one of my own major courses as an example: it was 60% memorization + 40% math-logic, and a few special chapters tested both concepts and computation. So I chose this rhythm: 3 hours in the morning attacking computation problems, and 30 minutes x 3 sittings after dinner memorizing concepts.

A Trap

For hybrid subjects, one concept can very likely be tested in more than one way.

The term explanation tests your memory of the concept; the short-answer question tests your grasp of where the concept applies; the computation problem tests how the concept slots into a formula.

The same concept: you may be fluent at plugging it into formulas, yet when you see it in a short-answer question your mind goes blank.

And the most awkward case is when you have covered most of the concept's tested angles, forgotten one -- and that one is exactly what shows up on the final.

So for concepts in these subjects, do a fast pass before the exam: what is this concept, what does it look like inside a short-answer question, and what does it look like inside a computation problem.

Three: Concepts -- Lossless Compression of Knowledge

First Principles

The mediocre student memorizes 100 question templates; the smart student masters 1 concept.

High school pays little attention to concepts: a quick sketch, then straight into problems. That runs completely counter to the underlying logic of learning. Whether you are learning knowledge or a skill, concepts are always the deepest, most foundational layer. Everything built on top is organized around them.

This is especially true in STEM. Every problem is an unfolding of basic concepts and laws under different boundary conditions.

For STEM, as important as concepts are derivations. The logic chain a derivation displays shows where a formula comes from, and understanding that logic beats force-memorizing the formula.

Within a derivation, understanding the logic chain matters more than the arithmetic. Calculation is a mechanical process; understanding requires deep thought.

Most people learn concepts the "language-class way": brute-force memorization. STEM concepts should be learned the "parameter way": understand what every parameter means.

Take Taylor's formula: failing to compute each coefficient is not a big deal. The key is understanding that the formula approximates a function using addition and multiplication, that the approximation proceeds by repeatedly differentiating, and how the remainder term is pinned down.

To test your understanding of concepts and formulas, test your grasp of a concept the way you would test code:

1. Zero/infinity test: what happens if some parameter goes to zero / tends to infinity?
2. Reverse test: if you flip cause and effect, does the theorem still hold?
3. Extreme test: under what conditions does this formula absolutely not apply?

If you cannot answer these questions, ask AI directly to help you understand.

For subjects leaning toward memorization and recitation, we can memorize and understand like this:

1. Dimensionality reduction of the text: take a definition or an argument, delete every adjective and modifier, and keep only the 3-5 scoring keywords.
2. Plain-language substitution: imagine you must explain this concept clearly to a kindergartner, and explain the keywords and terms using everyday words you know well.

Back to review. At the start you can go through the book against its table of contents, or have AI summarize you a list of concepts (which is not the same as a list of knowledge points). First make sure your understanding is sound, then memorize as much as you can.

If you have ample time, this step should take the lion's share. Only with the foundation laid can the skyscraper slowly rise.

Three: Internalization -- Reverse-Engineering the Exam Paper

Reverse Engineering

If doing homework is only about producing the right answer, why not just hand it to a homework-search app and AI?

The traditional problem-grinding path is "read the problem -- solve it -- check the answer -- correct it." While solving problems is a decent way to check your mastery, in university, learning well and testing well have nothing to do with each other.

Stopping at merely completing the assignment has no effect beyond building familiarity -- it is linear, inefficient diligence. However complicated the problem, it remains a mechanical process.

If you have read this far, you surely no longer want the traditional "sea of problems" tactic -- you want to internalize and master efficiently.

So what we must do is work backward from the answer to the question-setter's intent.

The Answer-Backtracking Method

1. Problem selection

Internalization starts from worked examples and homework.

From a review standpoint, worked examples and homework are of moderate difficulty and test only simple applications of knowledge points, so they will not slap you with extreme difficulty right away.

The worked examples on the PPT and the homework deserve serious attention. Teachers are busy with research and have no time to write exam papers; many final-exam questions are most likely worked examples or exercises with changed numbers, sometimes outright copies of problems from the textbook -- and past final questions may well be hiding in there too.

2. Read the answer first

Do not solve it yourself first -- read the answer first. Look at three things: the answer's structure, the scoring points, and the discipline's terminology. First locate what you do not understand, then work backward from the final result. Where did this number come from? From the data of the previous step. Why did that step use this formula? Because the problem gave that condition.

Backtrack step by step until you arrive at the problem's original conditions, and then at the knowledge point being tested. You will discover that the problem is just a clumsy bundling of conditions, and which knowledge point each problem tests becomes visible at a glance.

3. Independent replay

Being able to follow the answer does not mean you can solve the problem. Cover the answer and independently write out the problem's logic chain once. Skip the actual arithmetic if you like -- grasping the core logic is what matters. Wherever you get stuck, that sticking point is your review priority.

Repeat this process until you have eliminated every sticking point.

4. One problem, three passes

Do not do 3 problems -- do 1 problem 3 times.

First pass: read the answer and make the logic click. Second pass: solve it independently, producing the output yourself, to check whether you truly understand. Third pass: mutate the problem. Ask yourself: if this condition were removed, could the problem still be solved? If it asked for B instead of A, would the solution change? The moment you can produce variants of a problem, you are riding the question-setter.

Four: Precise Output -- Strike Where You Aim

Extreme Score Orientation

Only extreme score orientation produces extreme output efficiency.

This step is where those prepared past papers come in. Go to the library, set a timer, put down your phone, no music, no mid-exam bathroom trips -- simulate the real exam scene as faithfully as possible, and when you finish, grade yourself as strictly as you can. Be honest with yourself.

Your first paper will usually go poorly. That is fine -- the first paper's main job is to expose the holes in your review. You can even be brutal about it: leave blank anything you are unsure of.

While grading, record your sticking points and common errors. Ask AI about anything you do not understand -- AI is highly intelligent and on call around the clock. AI does hallucinate now and then, but that should not stop you from using it boldly. Squeeze this paper dry by every means available; do not touch a new paper until this one is fully digested.

A few days later, find time for the second paper. Follow the same procedure, simulate the exam environment again, and check whether the earlier problems got fixed and whether new ones appeared. If a mistake you made before shows up again, go back to where you first made it and redo that one problem. Repeatedly digesting one problem beats skimming over 100 problems like a dragonfly on water.

The third paper is best taken 3 days before the exam. By now your review is mostly done, and this run deepens impressions and keeps your touch sharp while catching any last holes.

By this stage you should have a sheet of paper summarizing question-type approaches, common errors, and other pitfalls.

The Cheat-Sheet Mindset

Disclaimer: the following is a review technique, not an incitement to cheat on exams.

If the teacher suddenly allowed you one A4 sheet of notes, what would you write on it?

Basic concepts? Formulas? Key and difficult points? Or problem-solving patterns?

Whatever you write, to shrink the content as far as possible you must compress ruthlessly, distill formulas, and summarize patterns. The process itself forces you to produce output.

If it were me, I would also consult the A4 sheet produced by the past-paper work above and write in every pit I have personally fallen into, giving myself early warning for the mistakes I am likely to make.

Once these things are written down, your brain remembers them automatically, and you will not even need to bring the sheet into the exam. So it is best done the night before the exam -- write it, then go straight to sleep to consolidate the memory.

The most advanced form of cheating is photocopying the textbook into your own head.

The Grading Algorithm

Teachers hate grading papers. They have no time to admire your prose; their eyes sweep like scanners, hunting for scoring points. But they will also stretch as far as they can from the standard answer to keep you from failing -- a failing student means trouble for both sides.

Your answers are written for a "manual keyword-retrieval algorithm."

Treat answering like writing code, and match that algorithm:

1. Never leave blanks. The vast majority of teachers will give you whatever points they can. Whether or not you know the answer, first translate the problem's conditions into mathematical notation -- for example, "at rest" can be written as v=0. Just writing these down often freeloads 1-2 points. If you write nothing at all, the teacher cannot give you points even if they want to.
2. Answer point by point. This goes for both the humanities and STEM. Lay out your answer line by line, and explain as clearly as possible. It makes grading easier for the teacher, too.
3. Formula first. Before any actual computation, write the original formula. This shows the teacher that you did not guess your way through -- you know a clear logic chain. As long as the formula is right, you collect most of the points even if every bit of computation is wrong. If you bury the formula inside the calculation, the teacher cannot see your reasoning, and all you can do is pray.

Five: Exam-Room Control Engineering

The Minus-20 Law

Energy always loses something in transmission -- and so do points.

Review is the "input-side power," while the exam is the "output-side power." In between lies a chain of losses: walking out of the dorm, arriving at the exam hall, meeting unfamiliar problems, sweaty palms, a fogged-up brain, a mistyped calculator key. That loss is about 20 points.

Review aiming at 100, and you usually land in the low 80s; aim at 80, and you usually land in the low 60s, scraping past the line; aim at 60, and you usually land in the low 40s -- start waiting for the make-up exam and retake.

If you want to pass, you absolutely cannot fixate only on the "just let me pass" content -- you must leave a safety margin for those 20 points of loss.

In the 5 minutes before the exam, your brain is still foggy. Do not rush into the problems. After writing your name and student ID, skim the whole paper to get familiar with the questions, and while your memory is still fresh, write the formulas down on scratch paper immediately, right now.

The Exam Procedure

Maybe you are this kind of person: well prepared during review, but the moment you enter the exam hall your palms sweat and your heart races; the instant the paper lands, your mind goes blank; you scribble through the paper in a panic, and can only regret it at the moment you hand it in.

I am not a completely nerves-of-steel player either. If you have met me offline, you know I get very nervous giving speeches. But for exams I have distilled a procedure for your reference:

1. Change your bed sheets and duvet cover 3 days before the exam. If it has been over a month, change them once so you sleep more comfortably. Sleep underwrites all your energy. Do not fantasize about sacrificing sleep for review -- unless you have only 3 days to cram.
2. The night before, read carefully through that sheet covered in solution approaches, common errors, and pitfalls, and anticipate what could plausibly go wrong the next day. Write them down so your mind has expectations. Fear comes from the unknown. If you have already thought through many scenarios in advance, they will not rattle you as much when they actually happen.
3. Prepare some tissues. Students with severe sweaty hands should pay extra attention here. Clothing alone cannot dry hand sweat; keep some paper to wipe with or lay under the exam paper, and you will feel much more at ease.
4. No reading in the last 30 minutes. If you have honestly completed every step above, there is no need -- more reading only breeds panic. Sit with your eyes closed and meditate, settle yourself, and cut the mental noise.
5. Skip any problem you have no idea about within 2 minutes. Even though university finals have relatively uniform question types, that does not mean zero variation, and you cannot predict whether the next problem will click for you. Grinding away at one problem only makes you more rattled; secure the points you can get first, and do not lose the big for the small.
6. Keep your scratch paper neat too. If you finish early and want to check your work, or you worry about mistakes, reviewing the scratch paper is a highly effective way to catch latent arithmetic errors.
7. No post-exam discussion. The moment the paper is handed in, nothing about it can change; discussing it is not only pointless, it wrecks your mindset. What if you check answers afterward and find you missed a bunch? Give up on yourself? Console yourself? Finals do not finish in one day. Do not let one subject's slip-up poison the rest.

Six: The Ultimate Weapon -- Agent-Based Interactive Learning

Absorb everything above and you can triple your efficiency. The approach that follows raises it by an order of magnitude and completely flattens traditional review.

"The first 18 weeks tell you I'm in college; the last 2 weeks tell you why I got into college" will no longer be a joke about final-week cramming -- it will simply be the way things are.

In the AI era, refusing to use AI is squandering a treasure. Yet most people's AI usage stops at the web chat window: searching for answers to problems, or asking it to explain some concept.

And what they tell the AI is just empty talk like "explain concept XX to me," so what they get back is naturally encyclopedic, hollow, dreary preaching.

Any web-based AI carries two limitations: 1, it can only solve problems one at a time, never a whole class of problems at once. 2, a large model's context capacity is limited -- once the conversation runs too long or too many files pile up, it gets dumber.

So as this kind of interaction goes on, the results keep degrading and the AI's output grows more and more "moronic" -- missing key concepts, botching calculations, with these amateur errors appearing again and again.

Our approach is to use an agent: first hand it a set of operating instructions, then have it read local files and generate markdown documents, turning the conversation into nodes that double as prompts for the next round. It also stops blindly pulling in random files and memories that produce bizarre answers.

As for what teachers say -- "even with AI, the real thinking you must do yourself" -- treat such correct platitudes as a bit of entertainment. They almost certainly cannot use AI properly themselves.

1. Identity Definition

AI should not be "a smarter Baidu" -- it should be "a senior collaborator."

AI is a probabilistic prediction model. It has swallowed the corpus of the entire internet, yet it has no idea whether you listened carefully in class today, and no idea of your course teacher's question-setting habits.

Without being told these things, it can only output the "least likely to be wrong" average -- and the average means mediocrity and patronizing preaching.

First tell the AI "who you are." Once its identity is defined, it can give answers that hit the vital point instead of useless, bloated platitudes.

And in the first conversation, you must tell it not only who it is, but also "who I am": for example, I have only 3 days to prepare for this exam, and I know nothing about XX.

Even a ridiculous identity like "I am a PhD student of subpar intelligence" beats saying nothing.

Never let the AI decide which face to show you. If you do not define it, it will fob you off with maximum mediocrity.

2. Context Hegemony

Even after telling the AI "who you are" and "who I am," its output may still miss the point and fail to solve your actual final-exam questions.

AI holds all the knowledge in the world, but it lacks exactly one thing: the context of this very moment. It does not know your textbook edition, the difficulty of your lectures, or your teacher's question-setting style.

Garbage in, garbage out -- in AI interaction this is absolute truth.

It is like a PhD student fluent in every subject who knows nothing about your situation; you have to feed it your background. So never open with "summarize the key points of calculus for me." Instead, throw in everything: the textbook you used, the homework you wrote, the past papers you worked.

You can use the following directory structure as a reference:

course-root/
  reference/
    textbook/
    ppt/
    photos/
    homework/
    past-papers/
    answer-keys/
    senior-notes/
  progress/

With these materials, the AI has context. It can study for the knowledge points and the exam you actually face, and its answers drift off-target far less often.

Tell it "I want to review chapter 4," and the AI will not hand you content that looks important but is never tested -- it will genuinely summarize the testable points from your files.

Putting reference materials in folders has one more benefit: the AI is not influenced by its "memory" feature, will not cross-contaminate subjects, and will not recall bizarre things where memory has no business being involved.

Add the prompts from above, and the AI knows not only "who I am" but also "who you are." Knowing both sides, its answers strike the vital point far more reliably.

3. You Are the Master of the AI

A week before the gaokao, I was walking along the river when I stumbled on a tarot stall. I had a reading done; most details have faded, but one sentence I remember vividly:

"You must seize the initiative and never let go, just like the Queen gripping her scepter."

AI can summarize for us, break things down for us, generate practice questions for us -- but the decision-making power over learning must stay in our hands. Hand everything over to the AI and you may well be led straight into a pit.

The driving can be delegated, but where the car is headed must be decided by us.

When we normally use AI to help solve problems, we probably end with vague phrases like "what do you think?" or "help me break this down." The AI then outputs mountains of steps -- maybe correct, maybe diverging from the reference answer -- but one thing is always certain: nothing stands out, and you have no idea where to start.

Same line as before: garbage in, garbage out. The input was an unclear instruction, so for "safety" the AI's output naturally avoids emphasizing anything. It is afraid its "key points" are not your actual test points -- exactly like how, in class, you are afraid of answering wrong when suddenly called on.

Command the AI with clear, forceful instructions, such as "which 3 terms must appear in this short answer?" or "explain this theorem to me in words a high schooler could understand."

Then the AI stops fearing a wrong output and, following your requirements, dares to output boldly.

Never forget: you are the driver of this game called learning; AI is merely your navigator and your engine. Do not let the tool turn master and make strategic plans on your behalf.

4. Agent Setup

First, download an agent tool -- Claude Code, Codex, Trae, or WorkBuddy, whatever. Anything that can read local files and install skills will do.

Then open PowerShell on your computer, confirm Node.js is installed, and enter the following command:

npx -y skills add Wh1te358/skills-by-Wh1te -g --all

With that, the skill is installed. Organize your materials in the format above and use it as follows:

1. Survival outline. You input the subject type, target score, time remaining, quality of materials, and exam question types. The Agent judges whether the target is realistic, generates a survival outline, ranks chapters by priority, lists the must-test concepts and formulas, and builds you an initial review plan.

2. Concept compression. For mathematical-logical content it uses parameter tests, deepening your grasp of concepts by varying parameters. For memorization content it uses keywords, keeping only keywords to guarantee you hit the scoring points.

3. Reverse breakdown of past papers. It does not start from "explaining knowledge points" but works backward from answers and scoring points: which formula written first earns the step points, which problem trigger word maps to which operation, which step is most likely to stall you, and the minimum you should write when you are completely lost.

4. Knowledge graph. STEM exams love testing the connections between knowledge points. The skill outputs: 知识节点表.md (knowledge node table), 知识连接图.md (knowledge connection graph), and 综合题型地图.md (comprehensive question-type map).

5. A4 cheat sheet. This step is not for cheating -- it forces you to compress high-frequency knowledge points, essential formulas, common errors, and scoring points.

I developed this skill myself and have open-sourced it on GitHub, at the following link:

https://github.com/Wh1te358/skills-by-Wh1te/tree/codex/exam-hacker-skill/exam-hacker

If there is any feature you would like this skill to have, leave a comment in the issues -- I take every suggestion seriously.

Seven: If You Only Have Three Days

You Only Have 24 Hours

You do not have 3 days. You have 24 hours.

But the good news is you do not need to learn the whole book -- only the whole exam paper.

The final is imminent and you have not even started "previewing." You have not read Self-Directed Learning: Science and Technology (《自学科学与技术》) either, so you decide to use the dumb method and push yourself: 8 hours of study a day. Bilibili speedrun courses mostly run 3-5 hours, so we take the midpoint of 4 hours. Suppose some parts still do not stick on the first pass, so you listen again and sketch out a rough outline against the book. Day one's 8 hours are gone.

Day two: you dig up past papers, or have AI generate a set. First attempt, 2 hours of simulation, and the result is unspeakable -- you start doubting whether you can even pass.

To avoid failing, you decide to squeeze dry whatever is easiest to score. The simple fill-in-the-blanks and multiple-choice get chewed down one by one; the big problems at the front fall step by step; the hard stuff at the back is strategically abandoned. 4 hours used. And for insurance, you even memorize the answer format for big problems, so that even if your mind goes blank in the hall you can still harvest some step points. Another 2 hours.

Day three: despite all the preparation, you are still panicked, convinced it is truly too late. But you still have 8 hours today -- you can still rescue this. At the very end, you go over the outlines and papers from the previous two days again and again, still feel uneasy, so you copy out a full set of answers by hand to find your touch. The morning's 4 hours are earnestly spent. But this review does not calm you -- it only makes you more panicked -- so you simply spend the afternoon's 4 hours gaming, and glance one last time at your prepared material before bed.

For passing, this is completely enough. Completely, completely, completely enough.

All right. The accounting is done. Now for the hands-on part.

Just-In-Time

First, the mindset: with only 3 days left, do not hope for a high score. Unless the teacher is kind enough to fish you up hard, be mentally prepared to scrape by with a 60-something. For those 90+ grind kings, a course in the 60s might be a devastating blow -- which is exactly why I hope such students skip this section.

Counting 6 chapters per course, 24 hours works out to about 4 hours per chapter. Within those 4 hours, even trying to "understand" one difficult concept is a waste of time. In the 3-day cramming mode, understanding is a luxury; memory is the necessity.

Three days before the exam, "seeking deep understanding" is the fastest road to failing. Be an emotionless problem-solving machine. You may forget all this knowledge the moment the exam ends -- but that is a post-exam matter.

Our principle: do not learn what will not be tested; brute-force memorize what is hard to understand; abandon what carries few points.

First prepare the textbook and the PPT -- but do not read either front to back. They are for the table of contents and for lookup. Spend 3 hours rapidly producing a knowledge outline against the textbook and PPT.

Then pull out the past papers and mark every point that has ever been tested onto that outline. The denser the marks, the more important the review; everywhere left blank, strategically abandon.

If you want AI to do the summarizing, put the textbook or PPT into the folder and tell it: "I have only 2 hours to review this chapter. Skip all background knowledge and derivations and tell me directly: 1. What are the 3 formulas that are guaranteed to be tested here? 2. When facing this question type, what is usually the first step of the standard solution?"

Whatever you do not understand, memorize outright. Drop the "why" and keep only the "what." See "find the limit," and L'Hôpital's rule launches immediately; see "extremum," and your reflex fires: set the derivative equal to zero.

Cut the same question type from the last 3 years of papers and paste them together; read 10 in a row. You will find they all look identical -- only the numbers changed.

Memorize the shape of the solution steps, and do not ask why. In the exam hall, even if you cannot finish it, writing down the overall steps collects most of the points.

Decisively abandon that hardest 20% of finale problems, and spend the time on the 80% of basic ones. Better a full score on easy problems than zero on the hard ones.

Eight: The Very Last

"Once you get into university, it gets easy." We all heard that line growing up, more or less. The truth is that even in university, the moment you want to take it easy, the atmosphere around you sweeps you up and you have no choice but to grind.

The better the school and the better the major, the fiercer the grinding and the more extreme the grade inflation.

A while ago I chatted with a high school classmate in Electronic Information at Nanjing University. They vented to me that the average-score cutoff for graduate recommendation had reached 91 -- and two years ago, that number was 83.

They sat right at the edge of missing recommendation, worrying about their prospects at every moment, even toying with taking the entrance exam to cross over to Southeast University next door.

Teachers, parents, and society expect us to be "outstanding," yet nobody has ever asked us what we ourselves want.

Before the gaokao, we were trained into excellent answerers. After the gaokao, we are forced to become question-setters. And the 4 years of university are exactly the transition from answerer to question-setter.

We stop waiting for others to hand us exam papers; we start finding problems ourselves, defining problems ourselves, iterating and trial-and-erroring ourselves.

So the purpose of this document is to "steal" time away from coursework for you -- time to practice setting questions. If you graduate from university with nothing but a slightly larger pile of useless knowledge than you had in high school, that is the saddest ending of all.

Here, everything I wanted to say has been said. I hope this document gives you something to take away, and inspires you to look at final exams -- and at university as a whole -- in a new light. For any comments or suggestions, you can give feedback through the survey linked at the end of the document or through this project's GitHub Issue.

I am walking the university road together with you -- you are not alone.

And finally, at the very last: may you pass your finals smoothly and get the score you want. And with the time you steal back, go walk a legendary road that belongs to you alone.

Acknowledgments

Thanks to Teacher Hu Nan (胡楠), Associate Dean of the School of Civil Engineering and Transportation at South China University of Technology (SCUT). The occasion for this book was Teacher Hu reaching out to me to pass review techniques for College Physics on to younger students. So I simply wrote the whole methodology into a PDF document, and iterated it step by step into what it is today.

Thanks to South China University of Technology for sparing my life. Even when freshman me was arrogant enough to mouth off at heaven and earth alike, the school chose tolerance.

Thanks to Gemini and ChatGPT for their tireless support of my writing. I suddenly feel rather lucky to have been born in this era. Without AI, the progress of this e-book would certainly have slowed to a crawl.

(A blank is reserved here to thank my future partner. The day I find them, I will fill it in.)

Reader Feedback

Scan the QR code below to fill in the reader feedback questionnaire for Self-Directed Learning: Science and Technology.