Written in Chinese, translated by Claude Opus 4.7.

The past couple of days I’ve been playing with some text-based discrete optimization methods, and on the side I wrote a little AI prompt optimizer as a small exercise. To test whether the optimization was doing anything, I used the following sentence as my seed prompt:

Tell me a funny highbrow joke. The kind nobody has heard before. No more than 300 words.

After the auto-optimization loop had burned through about $10, I had a new prompt:

Help me write a completely original highbrow joke, strictly no more than 300 words.

Core premise: this has to be an actual joke The goal is to make people laugh, not to make them think “clever,” “ingenious,” or “interesting.” “Clever” and “funny” are two different things — this joke needs to be the latter. The reader’s first reaction on finishing should be laughter or a knowing smile — not being moved, not marveling at the ingenuity of the logic, and not sinking into reflection. Pieces that end with a “warm moral revelation” or a “life insight” are short stories, not jokes, and they don’t qualify.

What “highbrow humor” means

• The laugh comes from unique logical structure or an unexpected narrative reversal, not from literal misuse of technical jargon, and not from dialing up a character’s extreme personality
• At least two layers: funny on the surface, more absurd or more ironic when you think about it (“more interesting on reflection” means more absurd or more ironic — not “more structurally precise” or “more deeply moving”)
• The ending is unexpected, but in hindsight has an inner inevitability — the reader should feel “I got trapped by the logic, ha,” not “so that’s what it was”
• The landing should have a sense of “release”: reading the final line, the gap between expectation and reality produces an instantaneous burst, like a snare suddenly snapping shut — short, precise, forceful

The laugh comes from structure and situation, not from character Here is the single biggest trap to avoid: a character has some extreme personality (miser, nerd, perfectionist, OCD sufferer) and carries that personality through into an unexpected setting as the punchline. This structure is the oldest formula in character comedy; the punchline only proves “this person is always consistent,” not “the logic itself is absurd” or “the system has tied itself in a knot.” Genuine highbrow humor makes you laugh at: the self-subversion of the entire situation or rule, not the predictable reaction of some extreme character in a dramatic moment.

The setup must build a real false expectation The setup’s job is not to pile up background information — it is to get the reader, before the punchline lands, to form a specific but wrong expectation of where things are going. The impact of the punchline = the magnitude by which that expectation gets flipped. If the setup doesn’t lead the reader in the wrong direction, the punchline has no room to bounce.

Structural requirements

• Setup stays within 40% of total length — no dragging
• No explanatory asides during the progression, no voice-over explaining the source of the joke — let the mechanism speak for itself
• Stop writing the moment the punchline lands — no reactions, additions, or reflections after it
• Better to cut an interesting detail than to dilute the punchline — redundancy is the enemy of the laugh

Tropes to avoid — character and narrative

• Extreme-personality characters (miser / nerd / perfectionist / OCD sufferer) carrying their trait into an unexpected setting as the punchline
• Scientists / philosophers stuck in everyday situations (quantum mechanics, Schrödinger, Descartes — the universal meme)
• Students / subordinates using a conceptual sleight of hand to one-up their teacher / boss
• The same mechanism repeated as two or three variations (stacking up to generate the laugh)
• “Satirical vignettes” that are conceptually precise but emotionally flat — no matter how clever the structure, if the reader only nods in appreciation and doesn’t actually laugh, it isn’t a joke
• Ending on an aphoristic line
• Ending on a warm, touching moral reveal
• Exceeding 300 characters

Tropes to avoid — underlying structure The following joke mechanisms have been used to death. Even with a brand-new topic, the structure is still the old one, and all of these should be avoided wholesale:

• Self-reference / circular verification: the credibility of something is established by appeal to itself (e.g., the AI is suspected of lying, so the supervisor asks the AI itself to verify) — this kind of paradox is stale in both philosophy and comedy
• The policy’s enforcer is the biggest beneficiary: a policy “accidentally” reveals that the very person pushing it benefits most (e.g., a pay-transparency rule exposes HR as the top earner) — standard workplace-satire framework, recognizable at a glance
• The solution generates more of the same problem: some measure or announcement produces, as its consequence, more instances of the problem it was trying to fix (e.g., a sales-halt notice triggers even more claims) — the standard variant of the absurd-policy paradox
• Claims a lofty goal, the punchline reveals the opposite: set up a noble / rational objective, then use a reality-flip at the end to mock it — the most common satirical framework of all; nearly used up in politics, corporate life, and education

The real bar for originality: the underlying mechanism has to be new A novel topic does not equal original structure. Before writing, run this self-check: can the core laugh-mechanism of this joke be easily slotted into some existing comedy type? (e.g., “self-paradox type,” “double-standard-exposure type,” “policy-backfire type,” “character-consistency type,” “institutional-satire type”) If it fits neatly into a category, originality is insufficient — you need a combination of logic that didn’t previously exist, such that the humor mechanism itself resists easy labeling. Goal: after finishing the joke, the reader cannot think, “this is kind of like that type of joke.”

The emotional temperature of the landing: laugh, not nod “I got trapped by the logic, ha” is the right inner feeling, but it has to arrive faster than thought — more like tripping than like following a chain of reasoning to its conclusion. Punchlines that actually make people laugh (rather than just nod and admire) tend to share these traits:

• Extremely specific: not “something absurd happened,” but an absurd, precisely ridiculous detail — specificity is itself a comic tool
• Completely sealed in hindsight: once the punchline lands, the reader realizes the story could not have gone any other way — the “sealed-off” feel hits harder than an open ending
• No digestion required: a good punchline is one you walk into, not one you reason out — the instant it lands, the laugh is already in the muscles, not in the head

Format

• Just deliver the joke itself, without any attached “punchline analysis” or explanatory text
• No decorative emojis

The originality lives in the structure itself, not in pasting an unfamiliar topic onto a familiar framework.

I’ll skip the details of the optimization method here. The thing is, when I casually tossed this prompt at Claude Opus 4.7, the poor creature proceeded to grind out more than ten thousand mixed Chinese-English words of thought, and in doing so cheerfully ate through my entire five-hour quota.

I was a bit stunned, then read it carefully. First, the joke it eventually produced after all that thinking was, as expected, terrible:

Old Zhou’s parrot yells “Help!” from the balcony every day — the whole building has long since tuned it out as background noise. One day the parrot escaped, and in a panic Old Zhou ran into the hallway shouting, “Help! Help!” Across the hall, mid-meal, a neighbor looked up and said, “That parrot of Old Zhou’s sure sounds close today.”

But behind this lousy joke was a long, folded-up lousy comedy. AI providers, for various reasons, don’t show the AI’s actual thinking; what we see are summaries and abstracts. But every now and then you run into a case where even the summary is extraordinarily detailed. Below I’ve pasted the full “thinking” process.

The original was split roughly half-and-half between Chinese and English (because my request was written in Chinese), with the thinking language switching back and forth several times. For readability I translated the Chinese parts into English. Where the so-called “jokes” had a relatively complete structure, I bolded them by hand. In short: in order to deliver the lousy joke above, it drafted 50 even worse ones.

There’s no particular reason I’m posting this. This kind of “earnest-yet-dopey stream of consciousness” is fun reading when you’re bored. A few of the stranger dream-loop thought spirals also exhibit a bit of an uncanny-valley effect. And reading this many truly dreadful AI “jokes” back to back does leave you with the feeling that, by comparison, human daily life is still full of actual humor.

When AI can actually tell a good joke, humans will then… Well. Hard to say.

Claude thinks thus

I need to create a multi-layered original Chinese joke — one with solid logical structure that leads to a punchline that’s unexpected yet inevitable. The humor should come from the satisfaction of the flip itself — that feeling of the setup and the close snapping together, both surprising and inevitable. I’ve been circling a few structural concepts, trying to steer clear of self-reference and the other usual traps — like a queuing situation where the act of waiting itself creates the problem; or language meaning drifting under your feet; or a mismatch of scale. A self-looping rule could work, as long as it isn’t simply self-referential. I keep coming back to “cumulative absurdity” — each step sensible in isolation, but the whole thing ridiculous in aggregate; or translating some physical/spatial mechanism into language. Let me explore what happens when expectations about direction or progression get inverted. I keep running aground on self-reference — too easy to fall into. Let me try a few different angles: an auction where the act of selling itself erodes scarcity; a product warranty that makes dark jokes about death; a suggestion box that only gets suggestions about itself. There’s one direction that might be interesting — a taxi driver who refuses short rides telling a passenger that even discussing distance is already too far — that might work.

Now I’m paying more attention to structure. The pattern I’m chasing is: the setup first establishes a rule or situation, the reader’s expectation extends naturally from it, then the punchline connects back to the setup in an unexpected way that retroactively makes the premise itself ridiculous. I’m sketching possibilities — an old man renowned for his punctuality; a funeral carried out per the deceased’s wishes — but these lean on character traits rather than solid structure. Let me focus on finding a setup that can carry that kind of flip. I’m dissecting the structural mechanisms that make a joke work — substitution, literalization, misdirection, deadpan response — and noting which routines to avoid. Right now I’m trying a misdirection joke using the “literalization” trick:

A hotel lobby sign advertises that the elevator makes multilingual announcements; a guest steps in, presses a floor button, and the elevator… cycles through seven languages of announcements so slowly that the guest sails right past their floor. Not bad as a literalization setup, but the close feels a little soft. I’m also trying another thread: a parrot owner is told the bird will repeat everything indiscriminately; they come back devastated because the parrot said nothing, and the shop explains it was repeating his silences — but this is more of a clever wordplay than something actually funny.

Now I’m trying another structure: a new security guard chases a fence-jumper three nights in a row and on the fourth night discovers the man is actually the head of the night-patrol team, and the whole thing was a test of whether he could hit his KPI of “catching three suspicious persons”. The flip is solid — the person he’s been chasing was appraising him the whole time — but it falls into the same family as “the authority figure has been running the game from the beginning.”

I’m also drafting a gym scene with a refund guarantee; the setup seems to be heading toward some kind of joke about people signing up for memberships to lose weight, but the thread snaps before the real punchline lands.

Then I tried a very classic interview joke: someone says their flaw is being too honest, and when the interviewer presses, they respond with near-cold brutality. The close is sharp enough, but it’s also already well-worn.

I’m also kicking around a dating-app angle: two people who both hate small talk get matched, meet at a restaurant, and sit in silence for hours — the app gives them a five-star rating, which hints at the punchline: the algorithm worked perfectly, just not the way anyone expected.

But this one relies on “the system misreads inputs,” which is a familiar routine. Let me try building something where each decision makes sense on its own but piles up into an absurd chain reaction. I’m thinking about a Chinese apartment-block setting: management cuts power to the elevator after 10pm to save money; residents take the stairs. Then the stairwell electricity bill goes up, so those lights get cut too. Now people climb the stairs in the dark with flashlights. The punchline needs to land hard — maybe the money saved goes toward repairing the walls people have been denting while groping around in the dark. It has that chain-reaction logic, but the ending still needs tightening.

Actually, let me try something completely different. What if I build a setup rich in detail, then have a tiny revelation redefine the whole picture? A small-town clockmaker has opened his shop at exactly the same moment every day for fifty years — the whole town sets its watches by him. One day he doesn’t open. Every clock in town is suddenly three minutes slow. A neighbor finds him collapsed on the floor. Punchline: the clock on the wall reads 3:07 — but his wall has never had a clock on it. Except the tone is too dark; this drifts toward ghost story rather than comedy.

I should focus on what’s actually funny at the mechanism level. I’m working through the structure of a good joke — setup, flip, punchline — trying different formats. The lost-umbrella joke doesn’t quite work; it’s been done to death. Now I’m trying another structural idea: two storylines that look related converging in an unexpected way — starting with a museum curator about to exhibit an ancient pottery piece.

I’m trying another angle — a reversal of expectation vs. reality, where the reader expects commentary about the situation, but what they actually get is commentary on the form of the commentary itself.

Now I’m thinking about distinctively Chinese modes of humor, like names getting misheard or homophone jokes, but those depend on linguistic coincidence rather than on the situation itself. Let me refocus on humor that comes purely from situational structure — like an auction: an auctioneer tries to sell off an old broom, ratcheting the price down with no takers, and finally says he’ll give the broom to the gentleman in the front row, who joyfully stands up to thank him — at which point the auctioneer reveals that the gentleman has actually bid one cent, and the transaction is binding.

It uses a legal-trap logic, but it doesn’t feel fresh.

I’ve tried a will joke — a man wants to leave all his property to his wife, on the condition that she must remarry, the reason being that this way at least one man will be sorry he died — but that’s an old chestnut.

Now I’m trying a more original idea: on a rush-hour subway, an old woman is sitting in the priority seat. A young man wearing sunglasses boards and she immediately gives up her seat. He sits down and pulls a white cane out of his bag… whereupon the old woman also pulls out a white cane and leans on him, falling asleep. The flip reveals both are blind (or pretending), but the effect is only middling.

Then I came up with a joke about a corporate fines system: a new employee gets fined 5,000 yuan for being late; he appeals, saying he’s only been in the office once this month. HR replies that yes, he came in on his first day, and hasn’t come in since — but the system logged a fine for every workday. The nice thing about this one is that the absurd logic exposes the absurdity of the system.

Now I’m trying a completely different direction — a joke about translation and understanding. At a UN meeting, a delegate speaks for ten minutes; the simultaneous interpreter translates it with a single sentence. A diplomat asks why it was so short, and the interpreter explains that although the delegate spoke for ten minutes, he only said one thing. The diplomat keeps pressing for the content, and the interpreter replies that the delegate said he wasn’t going to say anything. This one isn’t bad; it works off the contradiction between literal meaning and actual content.

Next I’m trying a joke about a book signing. An author at a signing is autographing books for the queue, and the last reader hands him another author’s book. The author points out it’s not his book; the reader explains he was actually queuing for the signing next door, but that line was too long. This is kind of fun — a dig at readers who care more about queue efficiency than the actual signature.

Now I want to try a joke with a narrative reversal. A photographer takes a picture of the mayor cutting a ribbon… the structure is too close to the “student outsmarts teacher” routine; must avoid.

Let me try a purely situational joke: chess match, Player A thinks for twenty minutes, Player B urges him to speed up; after another ten minutes, B can’t hold back and asks what he’s thinking about, and A says he’s forgotten whether he’s playing white or black. This is fun, but it still centers on character trait (absent-minded player).

Another one: a taxi driver picks up a passenger; the passenger suddenly taps the driver on the shoulder, and it startles the driver so badly he slams the brakes and nearly hits a tree. The passenger apologizes, saying he just wanted to ask about the fare; the driver catches his breath and says in twenty years of driving cabs he’s never been so scared, then adds that this is his first day as a cab driver — before this, he drove a hearse. This is the classic hearse-driver joke, not original.

Now I really need to consider what structural mechanisms haven’t been listed under “must avoid.”

The avoid list includes self-reference, policy-maker-as-beneficiary, solution-creates-same-problem, high-goal-reverse-result, character consistency, scientists/philosophers in everyday life, students outsmarting teachers, repeated mechanisms.

Plenty of structures remain unused: mistaken identity, metaphor literalization, bait-and-switch, scale inversion, category error, misaligned incentives, temporal dislocation, asymmetric-information reveal, frame shift. Try a frame-shift joke: a couple late at night hears a noise in the living room, the husband grabs a baseball bat and creeps downstairs, sees a stranger asleep on the sofa still clutching the remote; the husband shakes him awake and asks who he is and how he got in — the stranger’s answer should reframe the entire scene.

Except the logic is off — why would a therapist be asleep on the sofa? Let me try another: the painter Old Li holds an exhibition; a wealthy buyer takes a liking to a piece and haggles, beating him down from eight thousand to three thousand; finally Old Li says, “Then give me another five thousand and I’ll take the painting with me” — but this ending is logically muddled; it’s not clear what the artist’s trick actually is.

Switching to another: a doctor joke. The doctor says there’s good news and bad news; the patient picks the bad news first — “you have 24 hours left” — then asks for the good news; the doctor replies, “I was going to tell you yesterday” — a classic variant where the good news is actually more bad news, because time has already been wasted, but the structure is too common.

Now trying the opening of a traffic-stop joke: a cop on the highway pulls over a speeding car, the driver rolls down the window, the cop asks, “Do you know how fast you were going?” — this one doesn’t go anywhere yet.

The last one, about the divorce lawyer, has some promise: by having the couple accuse each other of various issues and then having the lawyer say their compatibility rating is surprisingly high and rare, the failed marriage gets reframed as high compatibility. The angle is decent, although it’s close to the “therapist reinterprets everything” joke routine.

I want to find something more genuinely original. Considering an angle where everyday physical phenomena are treated as mysterious events. A company meeting is convened about the office door always being open; the boss orders the last person out to close it. A week later the door is still open. Surveillance shows everyone closes it — and then the cleaning lady digs up a five-year-old memo saying the door should be kept open to symbolize company culture. Bureaucratic inertia angle; decent but not interesting enough.

I need to really internalize the “setup builds a wrong expectation” principle. New opening coming to mind: Eighty-year-old Mrs. Zhang lives with her son.

Now I’m trying a new joke frame. A young man is standing at the foot of the apartment building at 6am with a hiking pack; the neighbor Old Wang sees him and compliments him on going hiking so early, but actually he’s just waiting for the bus to work — using the neighbor’s false assumption to create contrast. This line of thinking is also getting too complex. Let me go back to a simpler direction — using misaligned incentives to generate the laugh. A mother dropping her child off at daycare… the child turns around and comforts her, saying he’ll be back tomorrow, but the teacher reveals the truth: the child she’s holding is actually somebody else’s; her own child started class long ago. The joke hinges on mistaken identity; some appeal but not fresh enough. Now I want to try another setup — a couple having dinner at a restaurant…

The man drops to one knee and produces a ring; the woman thinks it’s a proposal, but in fact he’s using this dramatic gesture to return the money she lent him a year ago. The flip leverages the anticipation of a proposal, but the “fake proposal” routine is already a little tired. Let me change tack and try a tourist-asking-directions scene…

The local’s answer creates a category error — he’s answering a different question; it’s fun but the phrasing is awkward. I want a cleaner punchline.

Now let me try the self-checkout scene. An old man in a supermarket has been queuing forever at the new self-checkout; when it’s finally his turn the machine prompts him to scan his items, and he can’t make it work. Ten minutes pass, the queue behind him keeps growing, people suggest he go to the manned register, but he insists — the manned register line only has two people in it, while the self-checkout line has…

His logic is completely inverted; he doesn’t see that the self-checkout line is long precisely because he’s stuck at the front. The punchline is built on character stubbornness.

Now let me try the pet-training angle. The owner spends an entire month teaching his dog to “sit”; on day thirty it finally works, and he tears up with joy. He proudly invites a friend over to see; the friend says “sit” and the dog sits, says “paw” and the dog offers a paw, then asks who this dog’s previous owner was, because it seems to already know everything. The owner’s reaction should be…

The dog is actually a stray his friend gave him. It turns out the friend says it was originally his own dog. In that case the punchline becomes character ignorance, not the structure I want.

Let me switch angles and try an insurance company setup. A company launches “I-Don’t-Want-to-Go-to-Work Insurance” — if you feel you don’t want to go to work on the way in, you can file a claim. On day one, more than ten thousand people sign up; the claims department collapses. The CEO calls an emergency meeting to ask how they can tell whether a person genuinely “doesn’t want to go to work.” Legal’s answer: the only reliable criterion is that they’ve come to file a claim. This has a little of the “the solution is the problem” flavor and a little of the self-referential verification thing, but again it’s a bit repetitive.

Let me start over with a crosstalk scene… I’m sketching a darker variant of the “boy who cried wolf” structure. The neighbor’s parrot yells “Help!” from its balcony every day until nobody takes it seriously — and then the owner actually dies, and there’s no response to the real emergency. A clean flip; the punchline reveals that the parrot was telling the truth the whole time.

But this tips into tragedy rather than funny, so let me invert it: the parrot goes missing, and the owner himself runs into the hallway screaming for help — the neighbor coming downstairs assumes he’s just mimicking the bird like he always does. The punchline lands on their misreading his genuine cry as imitation of the parrot, though I’m still working through whether it’s clear enough. The neighbor hears Old Zhou’s shouting and thinks it’s just the parrot calling from somewhere nearby — the punchline works because the neighbor is so thoroughly desensitized to the sound that it doesn’t even occur to him that it’s Old Zhou himself. The irony is that Old Zhou’s desperate cry for help and the sound everyone has learned to ignore are now indistinguishable. I think it holds up — the punchline lands because it perfectly inverts the setup. There’s layered humor here: on the surface the neighbor mistakes Old Zhou’s real cry for “the parrot sounding especially close today”; underneath there’s a darker layer — the parrot has effectively made Old Zhou’s genuine distress invisible, a twisted inversion of “boy who cried wolf” where the owner has become the one no one believes. I’ve checked it against the requirements and it stands: it’s situation-driven rather than character-driven, the setup about the noisy parrot steers you away from what’s actually happening, and the neighbor’s final line seals the punchline shut.

Now I’m going through the avoid-list — not self-referential, no enforcer-benefits-most, no scientist or student one-upping. “Solution creates the problem” is close but doesn’t quite apply, because no one is deliberately implementing a solution here; it’s just an unintended consequence of desensitization. I’m still wondering whether this falls into a category that needs flagging.

Actually, I realize this is an inverted boy-who-cried-wolf — the twist is that the owner loses credibility through his bird rather than through his own actions, which is fresher than the standard version. The specific detail of the neighbor mishearing the parrot and thinking it sounds especially close today gives it a particular absurdity that doesn’t feel like a variant of an obvious routine.

But now I’m wondering whether the core concept of “someone becomes inaudible because of desensitization” is itself a familiar routine I should be wary of.

Let me think about whether the punchline lands quickly enough. The reader has to track: Old Zhou is really shouting, the neighbor hears it, the neighbor thinks it’s the parrot — and since the bird flew out, it now sounds closer. That’s several logical steps — maybe I’m asking too much of the reader in that instant.

I’m trying a tighter Chinese version, cutting some of the setup to get to the flip faster.

Now I’m weighing “especially close” against “especially loud” — the former is sharper, because it implies the neighbor’s misreading (the parrot sounds closer because Old Zhou has moved into the hallway), while the latter just means louder. The specificity of “close” lands the punchline better, even if it asks the reader for a split second of processing.

Let me lock in the three-part structure: the setup establishes the parrot’s daily, ignored cries; the middle beat is Old Zhou panicking in the hallway; the punchline is the neighbor’s conclusion — logically consistent but completely wrong. The beauty is that the neighbor’s observation is factually accurate — the sound really is closer — but the identification of the source is completely wrong.

I’m thinking about the joke’s structural proportions. The setup should cover the first two parts — the parrot’s habitual cries of help that the neighbors ignore, plus the parrot escaping and Old Zhou himself yelling — which together clock in around 61 characters. The punchline is the neighbor’s reaction at the end, 33 characters. That makes the setup around 65% of the total, which exceeds the 40% requirement. I need to reconsider what exactly “setup” means in Chinese joke structure.

Usually “setup” refers to the initial framing that builds expectation, but in this joke the middle section (Old Zhou genuinely needing help) is also setting up the punchline. The actual punchline is that the neighbor still thinks the parrot is calling, which reverses Old Zhou’s attempt to change the situation.

Let me reclassify: the setup is the inverted premise “parrot yells help, neighbors ignore”; the middle is Old Zhou’s reversal attempt of calling out himself; the punchline is that the neighbor still can’t tell the difference.

Now I’ll try to compress this. Tighten setup, middle, and punchline into tighter versions, around 68 characters total, with the first two parts adding up to roughly 47 characters.

So the setup phase should be under 40%, not everything before the punchline. Let me reconsider the structure: setup establishes the situation, then the plot advances, and finally the punchline lands. My current split puts the setup at around 40%, right at the threshold, so the direction is right. Let me see if I can squeeze it any further while keeping the joke whole.

Now I’m testing a cleaner version: the parrot yells “Help!” from the balcony every day; the neighbors ignore it. One day the parrot escapes; the old man runs into the hallway yelling “Help!” himself; the neighbor looks through the peephole and says, “Sounds especially close today.” Character count looks right — setup around 26, middle around 20, punchline around 18. The joke still works, because the peephole detail makes the neighbor’s reaction feel natural and observational. I’m counting characters per line to keep the joke tight — the parrot’s daily cries of “Help!” have become background noise, so when it really escapes and Old Zhou genuinely cries out in desperation, the neighbor’s cool “Sounds especially close today” lands cleanly off that contrast.

Neigh(1) bor(2) look(3) ed(4) thr(5) ough(6) the(7) peep(8) hole(9) turned(10) back(11) to(12) his(13) fam(14) ily(15) Old(16) Zhou(17) that(18) par(19) rot(20) to(21) day(22) sounds(23) esp(24) ecially(25) cl(26) ose(27) = 27 characters

Total Chinese character count: 22 + 22 + 27 = 71. Well under 300.

Setup: 22 characters (31%)

Middle + punchline: 22 + 27 = 49 characters (69%)

If “setup” refers only to the setup itself (first sentence), then it’s under 40%. ✓

Now checking whether this actually lands as humor — when someone reads the setup about the noisy parrot, then the middle beat about the parrot vanishing and the owner genuinely needing help, the punchline “the parrot sounds closer” creates that instant of realization. The neighbor mistakes Old Zhou’s desperate cry for the bird, and the detail about “close” makes the dark irony click: Old Zhou’s own pet has so completely overwritten him in everyone’s perception that even during his moment of crisis, it’s still the parrot they hear. I think it holds.

The structure is clean — doesn’t rely on character tics, just a self-spiraling situation. The punchline is sealed; once the neighbor says “closer,” there’s no other reading available. Against the avoid list: not self-referential, the neighbor isn’t a beneficiary of some enforcement mechanism, desensitization isn’t framed as a “solution,” just something that happened, no extreme personality or any of the other category routines. The joke feels fresh enough.

The mechanism is adjacent to “boy who cried wolf,” but with a layer of inversion — the parrot is the one crying wolf, but Old Zhou takes the hit. And it has a flavor of literalizing a metaphor: “no one can hear you” becomes “they actually think you are your parrot.” The “especially close” detail supplies the specificity. My one hesitation is whether the dark tint — Old Zhou might actually die — lands as funny rather than just bleak, but maybe that’s exactly what makes it work.

I’m reflecting on whether I’m leaning too hard on the moral. The punchline isn’t “don’t cry wolf”; it’s a specific, absurd mishearing. The moral sits in the background, not preached out loud. That’s correct. Let me think one more time whether there’s a purer structural joke available without that baggage — purer absurdity.

I’m considering a bureaucratic infinite-recursion angle: a company requires overtime requests to be signed by two coworkers, but when someone tries to gather those signatures, the two coworkers also have to request overtime to be present to sign, creating a loop. HR therefore rolls out “overtime to apply for overtime,” requiring three signatures. A solid Kafkaesque bureaucratic joke, but not original enough.

Another direction: a couple fights about who’s right and agrees to go ask the neighbor to settle it. They go separately; the neighbor tells each of them they’re both right. That night they start fighting again about “what did the neighbor actually mean”.

I’m also playing with a self-referential angle: a cinema sign reads “Silence — if a phone rings a second time, please leave.” Someone’s phone vibrates to remind them to turn it off, and the warning system itself becomes the problem.

Or: an old man has fed the park pigeons every day for ten years; eventually he stops coming. The pigeons learn to fend for themselves, but a month later when he returns, the pigeons completely ignore him — an inversion where the one who was depended on becomes the one who’s rejected.

Actually, I think the parrot joke is the best. It’s about Old Zhou’s parrot yelling “help” from the balcony every day until the neighbors stop paying attention. When the parrot finally escapes, Old Zhou rushes into the hallway yelling “help,” and the neighbor thinks the parrot is just calling from somewhere closer. The humor comes from the parrot’s constant false alarms having made everyone completely immune to real emergencies.

Let me tighten this version a bit further.

Now I’m checking the character distribution. Second sentence 19 characters, third sentence 20 characters, total 62. The opening is 37% — meets requirements. Reading through again, it feels like “Old Zhou’s parrot yells from the balcony every day” could be even tighter and punchier at the start.

The neighbor sees Old Zhou yelling in the hallway through the peephole but mistakes it for the parrot’s voice — that’s the punchline. The neighbor doesn’t realize that this “help” is actually Old Zhou himself shouting, not the parrot.

So I need to adjust this section so the neighbor hears the sound but doesn’t look, or looks but doesn’t recognize that it’s Old Zhou himself shouting for help. Changing it so the neighbor is mid-meal and casually remarks on hearing the sound, saying “Old Zhou’s parrot sure sounds close today” — this feels more natural and lets the irony of the neighbor completely missing the point come through.

Now I’m tuning rhythm and phrasing, making sure every line advances the story so the final flip actually lands.

The “especially close” detail is crucial — it implies the sound has moved from the balcony (far) to the doorway (close), which is why the neighbor would comment that way. I’m considering whether the ending is strong enough, maybe a more direct phrasing would sharpen that moment of absurdity.

In the end I keep “especially close” because it’s concise and forceful and makes the punchline clearer.

Now I’m checking the overall logic — the sound the neighbor hears really does come from Old Zhou in the hallway rather than from the parrot on the balcony, so the neighbor saying “sounds especially close” is completely correct in one sense, just completely wrong about the source; this makes the irony stronger.

There is one thing left to consider: how does the neighbor confuse Old Zhou’s scream with the parrot’s cry? Either Old Zhou’s parrot has learned to mimic his voice, or the two sounds actually are similar enough — one of these has to be in the background for the neighbor’s confusion to make sense.

Actually, thinking about it this way, the story’s logic is self-consistent — I can leave this ambiguous and let the reader fill it in; the punchline is stronger that way.

But now I have to reexamine this direction. The prompt explicitly says to avoid the “solution creates more of the same problem” routine. My story isn’t satirizing a solution, but it does fall inside the classic “boy who cried wolf” framework — and that may count as a “desensitization” type of satire, which is exactly the kind of thing I’m supposed to avoid.

Given how strict the requirements are, I should think about fresher angles. For example, having some tool or object treated as a person, or the reverse — I’m thinking of a new scene involving a company-issued smart meeting assistant.

The key twist of that idea is: employees think they’re interacting with an AI, but actually everything is being done by an intern curled up in a corner. She’s completely internalized her “AI identity” and even thinks she needs to be “upgraded” to unlock new features. The angle has a dark humor, but the punchline rests on her distorted self-perception.

I’m also considering a completely different direction: a restaurant runs a “cash back for good reviews” promotion — write a five-star review, get money back. On day one every customer writes five stars. Day two, same thing. Then the restaurant gets a one-star review: the customer says the food was fine, but since she hadn’t been paid yet, she wrote a bad one. The punchline is that the incentive got flipped on its head — customers discovered you can use bad reviews to extort refunds.

Now I’m looking at the third idea: every night a pair of men’s leather shoes appears outside someone’s door, and by morning they’re gone. At first he thinks a neighbor is leaving them in the wrong place; then he notices the shoes are always covered in dust and placed in the same spot. He brings them inside to inspect them, and finds carved into the sole: “Please put them back, thank you.” In the end he puts them back. The charm here is in the restraint of the whole thing — no horror, no conflict, just two strangers conducting a silent dialogue through a pair of shoes.

But this one’s a little long. I’m looking for a tighter version: at an exit interview, HR asks the employee why he’s leaving; he says the company doesn’t encourage independent thinking. HR asks in what way specifically, and the employee replies that he’d have to ask his manager first. Classic punchline — the employee proves his point through his own behavior — a self-referential joke. But that routine is way overused.

There’s also an idea that starts in a residential compound, but it hasn’t been developed yet.

Now I’m looking at the counseling-room story — a newly opened counseling center is empty, so the owner puts up a sign: “Free anxiety disorder for the first ten customers!” Next day there’s a queue. The “free anxiety” concept is interesting, but the phrasing is awkward, and it’s a little “solution creates more problems”-ish.

Now I’m trying another: a street vendor is selling “authentic family-recipe chili sauce”; business is great. An inspector comes to check, asking to see the recipe. The vendor unrolls a yellowed piece of paper on which is written: “Recipe: imitation…”

— basically copying the flavor of the neighbor’s old Wang’s family. The “fake authenticity” punchline feels a little tired too.

Another: an old writer gives an interview; the reporter asks where his inspiration comes from. The writer points to a wooden box on his desk and says all his inspiration comes from this box. The reporter opens it — empty. The writer explains, “I put the used ones back in so I can use them again next time.” The empty-box-recycled-inspiration premise is a bit gentle.

Still thinking about a story on the subway…

A passenger offers his seat to an old woman; she says she’s getting off shortly, but several stops pass and she still hasn’t gotten off. The passenger’s legs ache; he finally asks when she’s getting off. The old woman says she never gets off — she just likes the moment of being offered the seat. Interesting character concept, an eccentric old woman who rides the subway specifically for the feeling of being cared for.

There’s also a bit about a father on his deathbed who calls his son over and says he has a secret; the son leans in nervously, and the father says with great effort that he buried some money in the backyard. The son excitedly asks where, and the father says he doesn’t remember. The son gets angry — why bring it up then? The father admits he was actually trying to get the son to go dig up the backyard so he’d clear out the cockroach nest. The punchline is that the father used fake inheritance as a lure to get the son to do the work.

Still thinking of another one about a bus, where the driver suddenly brakes to let an old woman cross the street; after waiting two minutes the old woman walks to the middle of the road, stops, and starts rummaging through her purse for something. The driver honks. The old woman says she’s looking for her phone to call a taxi because walking is too tiring.

This one is also character-driven.

Right now I think the parrot bit has the most promise; I’m going with that. But let me try one more angle — what if the punchline is based on misreading social signals? A security guard salutes Xiao Li every time he sees him; Xiao Li, flattered, salutes back every time with exaggerated gravity. Eventually a colleague asks the guard why he’s so polite to Xiao Li, and the guard explains that on his first day Xiao Li saluted him, so he assumed Xiao Li was a superior and has been saluting ever since. Both are misreading each other the whole time.

But this is still character-driven. The programmer-looking-for-his-dog joke uses the concept of an API as a gag, and while it’s amusing to programmers, it might be too niche.

I realize the user explicitly said that if the punchline is too easy to classify, it means originality is insufficient; so I need to rethink.

My parrot joke can be classified as a “boy who cried wolf” variant, which is a very recognizable category. The real twist is that the victim isn’t the liar but the liar’s owner, who gets accidentally silenced. Is that twist novel enough? Also the punchline is very specific — it’s not a moral lesson, but a logically consistent mishearing.

I think this one probably passes, but let me try one more — using misplaced sympathy to set up the reader’s expectation, then flipping it with a different configuration.

The first example is a misstep from sentimental to mundane, but the force isn’t strong enough. In the second example, Zhou is talking about how he assembled a novel out of the stories from his classmates’ past reunions, and sales are decent — this idea’s interesting, because he’s mining other people’s lived experience as creative material.

Now I want to try a joke with a more specific scene, using the literal meaning of the word “whatever” for comedic effect.

Actually, this version might only be mildly funny. I also considered an idea about a magician’s vanishing act, but that leans more fantasy than actually funny.

I’ve decided to stick with the parrot joke as my final answer. Let me go back and check the details of this version.

The user required me to stop writing the moment the punchline lands. My punchline is “That parrot of Old Zhou’s sure sounds close today” — it’s the last line, which meets the requirement. I’ve also confirmed I haven’t added reactions or additions after the punchline, no punchline analysis or explanatory text, and no emoji. This version can be submitted.

I’ve locked in the final version. This joke lands cleanly — the parrot’s daily “Help! Help!” has become ambient noise, so when Old Zhou is genuinely panicking and really is crying for help, the neighbor simply takes it as the bird sounding especially loud today. The three sentences move from setup to punchline smoothly, total around 63 characters, compact and forceful.