Epigrams

Witticisms, parodies, pointed observations, japeries, jocularity, Tom Swifties, examples of nominative determinism, and discursive drollery
humor, psychology, statistics, transhumanism, technology, Silk-Road, Bitcoin, power-analysis, Bayes, insight-porn
2014-04-302019-03-01 in progress certainty: possible importance: 7


Useful sayings

A list of quotes I find myself reg­u­larly using, not nec­es­sar­ily coined by myself.

Phi­los­o­phy:

  • “How would the world look differ­ent if X was true?”

  • “What did I expect to see before I saw X?”

  • “Name three exam­ples.”

  • “Sim­plic­ity does not pre­cede com­plex­i­ty, but fol­lows it.”

  • “By con­ven­tion sweet is sweet, bit­ter is bit­ter, hot hot, cold cold, and color col­or; but in truth there is only atoms and the void.”

  • “If some­one does not believe in fairies, he does not need to teach his chil­dren ‘There are no fairies’; he can omit to teach them the word ‘fairy’.”

  • “He [Omar Khayyam] is an athe­ist, but knows how to inter­pret in ortho­dox style the most diffi­cult pas­sages of the Koran; for every edu­cated man is a the­olo­gian and faith is no req­ui­site.”

  • “‘I don’t speak’, Bijaz said. ‘I oper­ate a machine called lan­guage. It creaks and groans, but is mine own.’”

  • “I am Loyal to the Group of Sev­en­teen.”

  • “Causes are differ­ences which make a differ­ence.”

  • “The would-be sor­cerer alone has faith in the effi­cacy of pure knowl­edge; ratio­nal peo­ple know that things act of them­selves or not at all.”

  • “The enlight­ened man is one with the law of cau­sa­tion.”

  • “Since the begin­ning / not one unusual thing has hap­pened.”

  • “Now, Char­lie, don’t for­get what hap­pened to the man who sud­denly got every­thing he wished for.”

    “What?”

    “He lived hap­pily ever after.”

  • “The fox con­demns the trap, not him­self.”

  • “If the fool would per­sist in his folly he would become wise…You never know what is enough unless you know what is more than enough.”

  • “You think you know when you learn, are more sure when you can write, even more when you can teach, but cer­tain when you can pro­gram.”

  • “In express­ing full func­tion, there are no fixed meth­ods.”

Sta­tis­tics:

Eco­nom­ics:

  • “There is a great deal of ruin in a nation.”

  • “Proof of Trot­sky’s far­sight­ed­ness is that none of his pre­dic­tions have yet come true.”

  • “The opti­mal num­ber of X is not 0.”

  • “Don’t let the per­fect be the enemy of the good.”

  • “If some­thing is worth doing, it’s worth doing poor­ly.”

  • “Never rea­son from a price change.”

  • “Dis­agree and com­mit”

  • “All sta­tis­ti­cal prob­lems are deci­sion prob­lems.”

  • “When two do the same, it’s not the same.”

  • “Beliefs are for actions.”

  • “If peo­ple don’t want to come to the ball­park how are you going to stop them?”

  • :

    • The Iron Law of Eval­u­a­tion: “The expected value of any net impact assess­ment of any large scale social pro­gram is zero.”
    • The Stain­less Steel Law of Eval­u­a­tion: “The bet­ter designed the impact assess­ment of a social pro­gram, the more likely is the result­ing esti­mate of net impact to be zero.”

Secu­rity:

  • :

    • “One can’t pro­ceed from the infor­mal to the for­mal by for­mal means.”
    • “It is eas­ier to write an incor­rect pro­gram than under­stand a cor­rect one.”
  • “Beware of bugs in the above code; I have only proved it cor­rect, not tried it.”

  • “When you have elim­i­nated the impos­si­ble, what­ever remains is often more improb­a­ble than your hav­ing made a mis­take in one of your impos­si­bil­ity proofs.”

  • “The com­plex­ity you add to a com­plex sys­tem to pre­vent fail­ure is itself a major source of fail­ure.”

  • “Attacks only get bet­ter.”

  • “Do not sum­mon up that which you can­not put down.”

  • “The strong do what they will, and the weak suffer what they must.”

Genet­ics:

  • “Every­thing is her­i­ta­ble.”
  • “Organ­isms are adap­ta­tion-ex­ecu­tors, not fit­ness-max­i­miz­ers.”

Tech­nol­ogy:

  • Ama­ra’s Law: “We tend to over­es­ti­mate the effect of a tech­nol­ogy in the short run and under­es­ti­mate the effect in the long run.”

  • Alger­non’s law: “Any sim­ple major enhance­ment to human intel­li­gence is a net evo­lu­tion­ary dis­ad­van­tage.”

  • Good­hart’s law: “When a mea­sure becomes a tar­get, it ceases to be a good mea­sure.”

  • Perlis 1982:

    • “One man’s con­stant is another man’s vari­able.”
    • “Every­thing should be built top-down, except the first time.”
    • “Beware of the Tur­ing tar-pit in which every­thing is pos­si­ble but noth­ing of inter­est is easy.”
    • “Most peo­ple find the con­cept of pro­gram­ming obvi­ous, but the doing impos­si­ble.”
  • “The future is already here, it’s just unevenly dis­trib­uted.”

  • Karpa­thy’s warn­ing: “Neural net­works want to work.”

Misc:

  • Cowen’s Sec­ond Law: “There is a lit­er­a­ture on every­thing.”
  • “Authors write things down to for­get them.”
  • “When two thieves meet, they need no intro­duc­tion.”
  • “Noth­ing in psy­chol­ogy makes sense but in the light of indi­vid­ual differ­ences.”

Statistics

  • “Often, I must cal­cu­late oth­er­wise than I think. That is called diplo­ma­cy.”

  • Sur­pris­ing lin­ear mod­els work at all, when they are the worst func­tional lan­guage ever: only oper­a­tor *, n vari­ables, n assign­ments, & out­put=­sum.

  • When cri­tiquing a paper, go for the jugu­lar: any part with the words “pre­sum­ably”, “obvi­ously”, “past research”, or “stud­ies show”. (If a sen­tence can­not sur­vive hav­ing the word “just” removed from it, it does not deserve to.)

  • Tip: try to read the appen­dixes of research papers as well. Sur­pris­ingly often big results, not just caveats & flaws, will be buried in them

  • If your local meta-an­a­lyst or method­ol­o­gist is not either an alco­holic or an insuffer­able jerk, they may be doing it wrong.

  • There’s always some­one who can make a per­pet­ual motion machine com­pli­cated enough that you, per­son­al­ly, can’t fig­ure it out.

  • “Remem­ber, cit­i­zens, full­text is our first line of defense against sci­ence jour­nal­ism!”

    • We can sleep soundly because rough men stand ready with blogs will­ing to do vio­lence to papers on our behalf.1
  • When does mea­sure­ment error mat­ter? If you can’t write “X cor­re­lates with Y” as “X cor­re­lates with scrib­bled sur­vey ask­ing about Y” with­out qualm.

  • The use of con­fi­dence inter­vals rather than p-val­ues is a clear improve­ment; it makes our diffi­cul­ties van­ish like smoke in a fog.

  • p-value test­ing is weird, but with mul­ti­ple cor­rec­tion, it gets even weird­er: the more you mea­sure & mod­el, the less you know.

  • Power vs error vs sam­ple size: painful trade­off of sta­tis­tics. Sins of omis­sion, com­mis­sion, & risk aver­sion—­painful trade­off of life?

  • Not doing power analy­sis with infor­ma­tive pri­ors is search­ing a haystack of unknown size for a nee­dle which prob­a­bly isn’t there.

  • Power analy­sis: if there is enough light to find the nee­dles in the haystack; vari­ance com­po­nent analy­sis: if there are any nee­dles at all.

    • Her­i­tabil­ity is like toss­ing a haystack into a fur­nace & frac­tion­at­ing the slag: you know the ratio of haystack­:needle; but not where the nee­dles were, how big the haystack was, whether many small nee­dles painted yel­low or a few big spears, how sharp the nee­dles were, or your neigh­bor’s per­cent­ages.
  • It’d be nice if peo­ple cared about sys­tem­atic vs sam­pling error, Bayesian meth­ods, infor­ma­tive pri­ors & design more than once every 4 years.

  • “If you ever decide unbi­ased­ness isn’t too high price to pay for sav­ing pos­te­ri­ors from entropy, let Bayes know. He’ll be ready.” /人◕‿‿◕人\

  • Remem­ber: p-val­ues are not pos­te­rior prob­a­bil­i­ties are not effect sizes are not util­i­ties are not profits are not deci­sions.

  • All analy­ses are ulti­mately deci­sion analy­ses. We can ignore that for con­ve­nience, but we must never for­get that.

  • Our beliefs are con­tin­u­ous, but our actions dis­crete. This leads to analy­sis paral­y­sis; can one exe­cute whole­heart­edly on +EV actions with P = 1% just as well as when P = 99%?

  • “Dis­agree and com­mit” to courses of action— pos­te­rior for solv­ing the of life.

  • Beware geeks bear­ing per­cents.

  • To mine for for­got­ten gold in a field, look for its most aus­tere & math­e­mat­i­cal sub­field­—­math­e­mati­cians are ter­ri­ble at mar­ket­ing.

  • RCTs : tra­di­tional West­ern med­i­cine :: Nixon record­ings : pres­i­dents :: DNA test­ing : courts :: priest abuse : Catholic Church :: cellphones/body cam­eras : police :: Wik­ileaks : geopol­i­tics :: Soviet archives : espi­onage accu­sa­tions :: Snow­den : com­puter secu­rity :: GWASes : can­di­date-gene stud­ies : FLOSS :: micro­econ­o­mists : Bit­coin : macro­econ­o­mists :: DL : ML :: coro­n­avirus : …

  • We will know AI has suc­ceeded not when we anthro­po­mor­phize machi­nes, but when we unan­thro­po­mor­phize humans.

  • The life of a neu­ron in Heb­bian learn­ing is oft nasty, brutish & short; cells must wire & fire together to sur­vive, thus cre­at­ing Lev­AIthan.

  • When I was a lit­tle kid, I thought cars told their dri­vers where to go (us­ing their turn sig­nal­s). It took a while, but even­tu­al­ly, I was right.

  • Why do humans have such large costly brains? Cats and chimps can see or walk as well as us.—Be­cause the deci­sions we make least with lit­tle feed­back or rewards based on big data over a life­time can mat­ter the most. We may be at our most human while fill­ing out the paper­work for life insur­ance. (With deep learn­ing, the new is not that AI finds hard­est what we find the eas­i­est, but that AI needs to be small­est—NN para­me­ter coun­t—what we need to be biggest!)

  • The news offers a fixed selec­tion of the most extreme out­lier (or should that be ‘out­liar’?) dat­a­points, but the global pop­u­la­tion becomes larger every day. Thus my corol­lary to :

    Human extremes are not only weirder than we sup­pose, they are weirder than we can sup­pose.

  • Sci­ence increas­ingly requires pre­cise pre­dic­tions which can fail, a will­ing­ness to fail, and shar­ing of the data demon­strat­ing fail­ure. Like hon­esty & sim­plic­i­ty, these are prices the most pres­ti­gious, wealth­i­est, and pow­er­ful find most diffi­cult to pay.

Biology

  • When­ever I feel maybe the FDA & big blind clin­i­cal tri­als are unnec­es­sary, I lie down & read Longecity threads until the feel­ing pass­es.
  • Count on it: the most humane orga­ni­za­tions oft have the most inhu­mane results. Save us from those who pre­fer being or seem­ing good to doing or giv­ing well!
  • Paleo? Atkins? Low-carb? Mediter­ranean? Maybe it is time to step back and ask: what should be the epis­te­mol­ogy of food?
  • Cor­re­la­tion ≠ cau­sa­tion: in causal infer­ence class­es, the first les­son taught, and the last les­son learnt.
  • Obser­va­tional sci­ence on the cheap is often nei­ther sci­ence nor cheap.
  • Every nor­mal man must be tempted now & then to sharpen knives, hoist a black flag, and run amok, shout­ing “No cau­sa­tion with­out ran­dom­iza­tion!”
  • If you con­sider them in terms of QALYs lost, school is at least as harm­ful to your health as smok­ing—and far more expen­sive.
  • Paul Krug­man notes you can look like an insane rav­ing loon just recit­ing eco­nom­ics text­book con­tents in a loud voice; like­wise, .
  • Some ask if behav­ioral genet­ics has made me nihilis­tic and feel­ing all is futile. But I have to believe in free will—my whole fam­ily does.
  • Àrxi vu (/ˈärˌkīv ˈvuː/; n.): when you read a new­ly-pub­lished paper and are haunted by the feel­ing you read a longer, bet­ter, ver­sion 2 years ago on ArXiv/BioRxiv.
  • On genetic engi­neer­ing and AI risk, opti­mist: “we can’t afford to worry about risks with sur­vivors”; pes­simist: “we can’t afford worry about risks with­out sur­vivors”.
  • Dis­ap­point­ments: when 23, read­ing a paper, real­iz­ing: “this is Peak Human. You may not like it, but this is it”. I had grown up: now differ­ences were quan­ti­ta­tive, not qual­i­ta­tive. (Never again would I see leaps like object per­ma­nence, the­ory of mind, numbers/counting, read­ing, the future, death, sar­casm, log­ic, game the­o­ry, pro­gram­ming, sta­tis­tics… All it took to reach the human fron­tier was a mere 20 years.)

Technology

Programming

  • Max­ims of reli­able com­plex sys­tems:

    • it is eas­ier to invent a buggy “reli­able” sys­tem than it is to under­stand a bugfree reli­able sys­tem.

      Corol­lary: no one under­stands a work­ing com­plex sys­tem.

    • the com­plex­ity you add to a com­plex sys­tem to pre­vent fail­ure is itself a major source of fail­ure.

    • Like in secu­ri­ty, the worst pos­si­ble state for a power cord, USB cable, or data­base is half work­ing.

    • Prov­ing prop­er­ties about sys­tems reveals bugs; it does­n’t mat­ter what prop­erty you prove! Sim­i­lar­ly, a lit­tle fuzz test­ing goes a long way: your sys­tem is prob­a­bly not robust even to a cat walk­ing across a key­board.

  • The soul of mod­ern man is so fallen & muti­lated that he can feel despair only when trapped in depen­dency hell.

  • One pro­gram­mer’s depen­dency solu­tion is anoth­er’s ver­sion con­flict.

  • Whether to check pre­con­di­tions before a loop, or after: this is to define san­ity and insan­i­ty.

  • Pro­gram­mers’ chairs and key­boards cause RSIs of the body; but what RSIs of the mind?

  • Will a just & mer­ci­ful devel­oper con­demn good pro­grams to bad smart­phones?

  • We have abstracted away from for-loops over arrays; but what idiom will abstract away the loops of our lives?

  • Back­ups are con­fronting one’s fal­li­bil­ity & the tran­sience of the world; we should not be sur­prised so few can do it.

  • When was the last time you saw Amer­i­can pro­gram­mers worry about Indi­ans? A les­son there. But don’t ask an Indian devel­oper what, ask an Amer­i­can.

  • The pre-In­ter­net PC was the real Wild West: a lone pro­gram­mer and his com­piler against hordes of quan­daries.

  • A pause on my key­board for thought—and how peace­ful it must be in the CPU, as the nanosec­onds slowly tick by…

  • “Bugs in work­ing code are moments for reflec­tion: how much we take on faith, because it seems to work!”

  • “Help­ing new­bies requires an active mem­o­ry—of all the times we our­selves failed to read the fuck­ing man­u­al.”

  • Is pro­gram­ming Chris­t­ian or Bud­dhist? Ask your­self how many of your pro­grams have the : , , & .

  • “Once a pro­gram has taken on a defi­nite form, it does not lose it until dele­tion.”

  • How can we fear machines will sep­a­rate humans when even in their source code, we can read the stamp of per­son­al­ity & style?

  • “Data, and the for­mats by which data are com­mu­ni­cat­ed, inevitably cre­ate a sys­tem per­me­ated by illu­sions.”

  • “A truck dri­ver may drive for 20 years with­out ever improv­ing; it must be con­fessed that many pro­gram­mers do lit­tle bet­ter.”

  • “Any­one sat­is­fied by last year’s code is not learn­ing enough.”

  • “I do not like this algorithm/language/tool.” “Why?” “I am not up to it.”—any­one, ever?

  • “The Inter­net is embar­rassed by the brows­er.”

  • “The HTML page is a stark data for­mat: every­where it goes, there is dupli­ca­tion of process. It is per­fect for hid­ing infor­ma­tion.”

  • There is but one con­stant in every pro­gram; and it is [see VM table entry 0x000007FEFC831010 → 0x00885ED010 → fault­ing in page…]

  • When you become frus­trated with com­put­ers, please remem­ber they are only clev­erly arranged sand. (When you become frus­trated with peo­ple…)

  • “Uproot your func­tions from their ground and the dan­gling roots will be seen. More func­tions!”

  • “I think there is a world mar­ket for maybe five com­put­er­s—­Google, Ama­zon, Face­book, Microsoft, and Baidu.”

  • A haiku:

    The sum­mer garbage—
    the sole rem­nant of many
    bright engi­neers’ dreams

  • All that is nec­es­sary for entropy to tri­umph is for good men to do some­thing.

  • We do not go because the traffic light becomes green; the traffic light becomes green because we go.

Languages

  • To test whether a lan­guage despises its users, merely see whether ‘if (a = 1) {…}’ is valid.

  • Web browser devel­op­ers are con­demned to rein­vent the OS, poor­ly. What is to be done about this? What could ever have been done about this?

    • …it is clear our tech­nol­ogy exceeded our human­i­ty. I do not know in what W3C stan­dard WWWIII.htm will be writ­ten, but WWIV will be writ­ten in Web Assem­bler.
  • Mozilla now has its own com­mu­ni­ty, browser, cloud, lan­guage, user­land, and in a final con­ces­sion to the inevitable—­Fire­fox OS!

  • Time-shar­ing, mul­ti­-user main­frames, main­frame VMs, PCs, dat­a­cen­ters, dat­a­cen­ter VMs, Dock­er, micro-ser­vices: the Wheel of Rein­car­na­tion yet turns. One must imag­ine Sysad­mi­nus hap­py.

  • C, C#, Go, R, Rust: are these unsearch­able names tes­ti­mony to some feet of clay, or tes­ta­ments to mon­strous egos?

  • “Haskellers knows the type of every­thing & the value of noth­ing.” Unless they’ve turned on exten­sions, then nei­ther

  • C—fast and effi­cient and for when you don’t have enough mem­ory to remem­ber things like why you don’t want to use C.

  • “C pro­gram­mers stay sane by imag­in­ing that all the other inse­cure buggy pro­grams are thanks to avoid­able unrep­re­sen­ta­tive rea­sons.”

  • We have made an AI break­through! With Pro­log, we have cre­ated the intel­li­gence of a 2-year-old child: “No. No. No. No.”

  • “Cre­at­ing a new good pro­gram­ming lan­guage is so diffi­cult it tends to be only fools who try.”

  • “Just as winds pre­serve seas from stag­na­tion, so also cor­rup­tion in lan­guages is the result of pro­longed calm.”

  • “That code is ill-writ­ten of which one must repent; as long as the PHP bears no evil fruit, the fool thinks it sweet as hon­ey.”

  • Sin­gle-par­a­digm lan­guages are admirable for push­ing until it break­s—or does­n’t!; cf Ein­stein & atomic Brown­ian motion; Colum­bus & the “spher­i­cal Earth”.

Bitcoin

  • If you find your­self sur­prised by man or mar­ket, remem­ber you have learned as much about your own think­ing as them: rev­e­la­tion comes in twos.
  • Bit­coin involves no new prim­i­tives or fancy proofs; per­haps cryp­tog­ra­phers should all along have been study­ing soci­ol­o­gy, not math­e­mat­ics.
  • The hor­ror of Bit­coin: money really is a social con­struct! & not always by nice-s­melling well-groomed peo­ple in sharp suits.
  • A mod­est pro­pos­al: end tenure for com­puter secu­rity & cryp­tog­ra­phy researchers. Given the sta­tus quo, if they’re not already rich, they can’t be any good.
  • “What is the price of two satoshis—one cop­per coin? But not a sin­gle satoshi can fall to fees with­out your Eter­nal Blockchain know­ing it.”
  • The brav­ery of inno­va­tors: what sus­tained Satoshi Nakamoto dur­ing those lonely days in 2009 when no one cared enough even to attack Bit­coin?
  • One weird phi­los­o­phy trick for analy­sis! Imag­ine worlds where X failed: “Sure Bit­coin died: a defla­tion­ary cur­rency requir­ing ever more waste?”
  • Let us hope Karpe­les will not be the of Bit­coin.
  • The opti­mal num­ber of dou­ble-spends is not zero.
  • was a genius until he was a fool & a knave; Satoshi was a genius until…?
  • Once you get locked into a seri­ous cryp­tocur­rency invest­ment, the ten­dency is to push it as far as pos­si­ble & col­lect all the alts. There is noth­ing in the world more help­less and irre­spon­si­ble and depraved than an early ether investor in the depths of an ICO binge.

Darknet Markets

  • “Ulbricht, Ulbricht, Ulbricht! A mil­lion coins were not enough for Ulbricht!”
  • We crit­i­cize Ulbricht for not know­ing when to quit & enjoy life; but what are we our­selves refus­ing to quit?
  • Every­thing looks per­ma­nent until its secret is known. For all too many things, the begin­ning of fear is the begin­ning of knowl­edge.
  • “A begin­ning is the time for tak­ing the most del­i­cate care; this every stu­dent of the OP SECurity knows.”
  • Mod­est pro­pos­al: let’s aban­don the term “OPSEC”. Instead, let us refer to the mis­takes of Ulbricht, Blake Ben­thall, , etc as “OOPSEC”.
  • Is cryp­to-lib­er­tar­i­an­ism self­-un­der­min­ing? Can we han­dle the sausage fac­tory of the dark­net mar­kets?
  • When a cen­tral­ized escrow mar­ket claims 1% com­mis­sions, it should append an aster­isk: “and a one-time 100% fee when (not if) hacked or shut down”.
  • vs cen­tral­ized escrow: proof that con­ve­nience is a drug faster & more addic­tive than the finest hero­in.
  • Decen­tral­ized Bit­coin mar­kets are the future of dark­net mar­ket­s—and I fear always will be.
  • Lust for lucre is the root of all evil? Alert the the­olo­gians: the num­ber of cen­tral­ized dark­net mar­kets is an index of Satanic activ­i­ty!

Economics

  • Good results fol­low good inten­tions even as the .
  • Con­no­ta­tions: ‘thrift’ is achiev­ing one’s goals as cost-effec­tively as pos­si­ble and max­i­miz­ing one’s bang-for-buck; ‘fru­gal­ity’ is choos­ing one’s goals to be as cost-effec­tive as pos­si­ble, and pick­ing a bang which min­i­mizes one’s buck. The for­mer is a virtue; the lat­ter, a vice.
  • If pigs were smart enough to wor­ry: “the humans nei­ther love nor hate you, but you are made of tasty bacon they can use for some­thing else.”
  • Opti­mists: believe every­thing has been selected to be as effi­cient as is pos­si­ble to achieve; pes­simists: every­thing is selected to be as ter­ri­ble as is pos­si­ble to get away with.
  • is the of eco­nom­ics”: it always seems to be employed by asses upon high.
  • A roll of the dice suc­ceeds 1 time in 6, and stage magi­cians can learn to con­trol rolls, while per­haps 1 in 10 star­tups will be truly a good use of time for the founders & employ­ees and suc­cess rates do not increase with expe­ri­ence; per­haps say­ing a startup is ‘as ran­dom as a coin-flip’ and ‘really rolling the dice’ should be taken not as an insult but as the high­est of com­pli­ments.
  • ’tis a fool­ish VC who invests only in things he’s sure of, and not adven­tures. If a VC does­n’t reg­u­larly look fool­ish, he’s a fool.
  • Mar­ket volatil­ity is igno­rance made vis­i­ble.
  • A mod­est pro­pos­al: replace col­lege with gym mem­ber­ships. Equal sig­nal­ing value for Conscientiousness/conformity/discounting; cheap­er; objec­tive; equally (un)­fair; health ben­e­fits; pro­gres­sive, not regres­sive; real RCT-verified trans­fer to IQ & cog­ni­tion, not hol­low gains; pos­i­tive exter­nal­i­ties for looks & tax­es; fit­ness knowl­edge increas­ingly use­ful in increas­ingly mod­ern­ized envi­ron­ments & diets, rather than less like tra­di­tional school­ing knowl­edge; more eas­ily researched & opti­mized.
  • Copy­right and tech­nol­ogy are odd­—when I was a kid, it was infi­nitely eas­ier for me to get my hands on 1000-year-old clas­si­cal Japan­ese poetry I wanted than on 1 (or 10) year-old Japan­ese ani­me; and now it’s the reverse.
  • No reward with­out risk? But
  • The past is a Third World coun­try.
  • Prob­lems that can be solved by mon­ey, should be.

Umeshisms

Umeshisms (1; see also GPT-3 Umeshisms):

  • If you never lose an eBay auc­tion, you’re bid­ding too much.

  • If you never return an item to the store, you’re buy­ing too few things.

  • If every­one in your class gets an “A”, you’re not teach­ing enough.

    If all fresh­men in a major earn their degree, too few are drop­ping out.

  • If all your exper­i­ments work, you’re just fool­ing your­self.

  • If you eat all your food before it goes bad, you’re eat­ing too much bad food.

  • If none of a city’s build­ings col­lapse in an earth­quake, con­struc­tion stan­dards are too cost­ly.

  • If your NASA rover pro­gram is seri­ous about “fast and cheap”, 0 is the wrong num­ber of rover fail­ures.

  • If your Proof-of-Work cryp­tocur­rency never has dou­ble-spends, the (min­er) rent is too damn high.

  • If there isn’t bla­tant waste in an econ­o­my, too many peo­ple are look­ing for waste.

  • If a race car sur­vives the race, it’s over-engi­neered.2

  • If your self­-driv­ing car fleet never runs peo­ple over, it’s being rolled out too slow­ly.

  • If you recall your cars for every defect, you’re recall­ing too many cars.

  • If your pack­ets are never cor­rupt­ed, you’re on error-cor­rec­tion & reli­a­bil­i­ty.

  • “The opti­mal num­ber of falling build­ings in an earth­quake is not zero.”

    • The opti­mal num­ber of mask short­ages…
  • If your new AI pro­gram is unde­feat­able by humans, you waited too long to run the tour­na­ment. (If your super­in­tel­li­gence isn’t rel­a­tively close to human lev­el, you’re hold­ing off too long on turn­ing the earth into com­pu­tro­n­i­um.)

    • If an AGI costs sub­-bil­lion dol­lars to train, you waited too long to scale.
  • Dark Lordisms: if you’re get­ting into pitched bat­tles with heroes or armies, you’re not using dark enough mag­ic.

  • If you keep going to the funer­als of your ene­mies, you have too many ene­mies.

  • The final Umeshism: If every­thing you do is easy, you’re under­achiev­ing.

Anti-Umeshisms

Anti-Umeshisms:

  • If you’ve never dropped any cours­es, it’s because edu­ca­tion is not about human cap­i­tal.

    Han­son­ism: one should never drop any cours­es, because edu­ca­tion is not about human cap­i­tal.

  • If you’ve never changed your favorite char­ity based on cost-ben­e­fit, it’s because char­ity is not about help­ing.

    Han­son­ism: …

  • If you’ve never dropped a bad book, it’s because read­ing isn’t about gain­ing knowl­edge.

  • If you’ve never changed your mind pub­licly, it’s because pol­i­tics isn’t about pol­i­cy.

Malthusianisms

38. “The world is not what any­one wished for, but it’s what every­one wished for.”

James Richard­son, Vec­tors: Apho­risms & Ten-Sec­ond Essays 2001

Malthu­sianisms:

  • Why do hairdry­ers blow fus­es? Because if they did­n’t, you would buy a hot­ter hairdry­er.
  • Why is it so hard to get com­pe­tent, hard­work­ing, vol­un­teers? Because if they’re any good, they already have too much work to do.
  • If books weren’t so long that they were tir­ing to read, authors would make them longer.
  • Why do returns dimin­ish? Because if they did­n’t, some­one would rein­vest in earn­ing returns in an expo­nen­tial spi­ral of wealth until they did dimin­ish.
  • Why are worth­while things often so hard to do? Because if things worth doing were easy, every­one would do the pleas­ant things and avoid the unpleas­ant things until the hard things were most worth doing.
  • Why are infec­tions so fatal? if the viruses & bac­te­ria did­n’t take risks & engage in overkill, they could­n’t get past our adap­tive immune sys­tem and would be selected against. Why can’t we have sim­ple cheap innate immune sys­tems? Because then we would be selected against by the cur­rent viruses & bac­te­ria…
  • Why are so many of the peo­ple you date cra­zier, uglier, mean­er, or poorer than you’d prefer? Because if they weren’t, they’d already be tak­en.
  • Rumeshism: Why is the rum is gone? Because if there were more rum, peo­ple would drink more rum.

Matthewsisms

:

  • Those star­tups who least need VC invest­ment & deliver low­est expected returns are those who receive the most VC offers of invest­ment.
  • Those who most want a girl­friend are the least suc­cess­ful in ask­ing girls out.
  • Those peo­ple with the most need for Con­sci­en­tious­ness & depres­sion treat­ment are those least able to fol­low ther­apy like .
  • Those schiz­o­phren­ics with the most need for med­ica­tion are those least likely to keep tak­ing it.
  • The most igno­rant who most need to fol­low advice & instruc­tions are least likely to under­stand their need.
  • Coun­tries which most need to fight cor­rup­tion are least able to do so.
  • …Mer­it-based admis­sions aid/scholarships; being rich and get­ting a loan; apply­ing for jobs…

Misc

Gen­er­al:

  • How would the world look any differ­ent than it does now if X were true? You say “because of X”, but where does X come from?

  • To learn to build sand­cas­tles on the beach is to learn to live and die an athe­ist. (“Here on the level sand…”)

  • “Every­one knows what would be best for them to do—­to­mor­row.”

  • “Neo, what if I told you… every­thing you knew was cor­rect?” (That would be the biggest dis­ap­point­ment ever.)

  • At times, tol­er­ance can be the most rad­i­cal of posi­tions to take; just watch when a weak group gains pow­er.

  • “Read­ing a flame war: and all you peo­ple must once have been lit­tle chil­dren, who smiled of a sum­mer day.”

  • Imag­ine a world where the just world and fun­da­men­tal attri­bu­tion bias were cor­rect, and karma exist­ed. Would­n’t that be unspeak­ably trag­ic?

  • War seeks to mold phys­i­cal con­di­tions as one wish­es; art seeks to mold minds as one wish­es. Both are based on decep­tion.

  • Art is often hatred: it shows the past, or future, or far-away but never the present moment. Anti-art: a TV & cam­era show­ing the view­er.

  • The “fal­lacy of gray” or the “fal­lacy of grey”? My sug­ges­tion “the fal­lacy of græy” was rejected with­out, I thought, proper con­sid­er­a­tion.

  • I have made progress in my med­i­ta­tions: yes­ter­day I told myself one truth, and only nine lies.

  • Black­mail is a on hypocrisy, either per­sonal or soci­etal.

  • Adult­hood is accept­ing no one will save you any­more; from this, all else fol­lows.

  • As the global pop­u­la­tion grows, the extremes become more extreme; they become not just weirder than we sup­pose, but weirder than we can sup­pose.

  • “But he was sin­cere”, we say of some­one, when finally no defen­si­bil­ity is pos­si­ble than the pos­si­bil­ity of defea­si­bil­i­ty.

  • There is no first-per­son pre­sen­t-tense verb for “to believe falsely”, Wittgen­stein notes; like­wise, there is no past tense for recall­ing bore­dom—un­like many emo­tions (hap­pi­ness, sad­ness), one can remem­ber that one was bored, but never the bore­dom itself.

    This for­get­ful­ness is a bless­ing when we remem­ber how much bore­dom there would be to remem­ber: from schools, if noth­ing else. Do not begrudge peo­ple their smart­phones and small moments of obliv­ion.

  • The Rule of Three: “If you find your­self doing some­thing 3 times, fix it.”

    If some­thing has come up 3 times already, then it’s prob­a­bly going to come up again. If you rant about or have to explain some­thing 3 times, write it up! If you for­get some­thing 3 times, make . If you make an error 3 times, . If you pro­gram some­thing 3 times, write a library or tool. If you do some­thing highly mechan­i­cal and tedious, auto­mate it. And so on.

Meta

  • An apho­rism is an algo­rithm, of we know not what input, we know not what out­put.
  • “It is eas­ier to write an incor­rect epi­gram than under­stand a cor­rect one.”
  • Voltaire’s Third Law: for every apho­rism, there is an equal and oppo­site anti-apho­rism
  • The epi­gram is a com­pressed, golfed, idea, with all the virtues—and sin­s—of golfed code.
  • If we mea­sure the entropy of epi­grams by how many peo­ple under­stand them, who is the gzip of epi­grams? The xz? (…the ZPAQ…?)
  • Remem­ber! Most strings are incom­press­ible, most reals uncom­putable, most the­o­rems unprov­able, most pro­grams unde­cid­able.
  • Should lan­guages sup­port the writer’s con­ve­nience or the read­er’s under­stand­ing? Frame it as a sta­tus debate, and all becomes clear.
  • Fear not known but unknown pro­pa­gan­da; I have the utmost respect for Pravda or research papers report­ing p = 0.04—how else will I know what to not believe?
  • Writ­ing down your ideas takes 90% of the time; debug­ging code takes another 90% of the time; rewrit­ing takes the third 90% of the time…
  • A day with noth­ing unusual is an unusual day. Fic­tion writ­ers & GANs must include the improb­a­ble to be prob­a­ble. Spend sur­prisal shrewd­ly.
  • Tran­shu­man­is­m’s vic­tory can be seen in how few now bother with the word.
  • The meta “M” hier­ar­chy: method­olo­gies over mod­els over mate­ri­als (raw data) over mem­o­ries (anec­dotes)
  • Remem­ber for later the point of read­ing a man­u­al: it is not to remem­ber every­thing that is in it for later but to later remem­ber that some­thing is in it.

See Also

Startup ideas

Poten­tially profitable late-stage cap­i­tal­ism per­for­mance art pieces:

  • sim­u­la­tor. Email the user once a week with n vari­ants or actions to pick, like directed evo­lu­tion games. Graphical/plant back­end can be pow­ered by CNN , paint­ing engines (eg ), out­sourced human artists in Chi­na, etc. To increase rev­enue, kill bon­sai at ran­dom with low prob­a­bil­ity & present memo­r­ial videos of how it evolved over the years, while offer­ing to make a micro­pay­ment to res­ur­rect a ver­sion of it or start grow­ing another from its ran­dom seed.

  • cus­tom man­u­fac­tur­ing of extremely dense metal (lead//) objects such as credit cards/keys/passcards/etc for elite lux­ury brands: unex­pect­edly heavy objects sub­con­sciously sig­nal Qual­i­ty. This is part of why SUVs are so heavy and why devices such as Beats head­phones have metal weights inserted in them.

  • NN GANs for pho­to­re­al­is­tic-but-le­gal child porn; trans­fer learn­ing from West­ern & Japan­ese porn and fine­tune on Dark Web & loli­con cor­pus­es. Startup slo­gan: “we put the AI in waifu and the ANN in husbanndo using our patented hentai”. You know what they say in SV: every good startup breaks at least a few laws…

  • Snapchat/Instagram but for audio, with such nov­elty fil­ters as talk­ing in or with heli­um, and audio “style trans­fer” from an approved bank of adver­tis­ing-spon­sored sound & voice sam­ples.

  • Soy­len­tant: a restau­rant serv­ing tast­ing menus of all drinks & bars, with rec­om­men­da­tions from the “soylem­mier”: “Ah, the August 2016 Soy­lent bar… A fine vin­tage. Usu­al­ly.” Also sells a cook­book for hack­ers: From Zero To Yum.

  • TVTropes+fMRI for ran­dom­ized movies/VR/novels which max­i­mize surprise/engagement by being some­what but not too unpre­dictable, fork­ing paths as nec­es­sary to main­tain inter­est at a high lev­el. Aes­thet­i­cal­ly, you want “famil­iar but not too famil­iar”, so there’s a sweet spot in ran­dom­iz­ing trope/subversion (cf Schmid­hu­ber’s the­ory of cre­ativ­ity). There’s prob­a­bly a way to infer surprise/arousal from (nec­es­sary for VR head­sets to do among other things), so this could be a use­ful tool for VR espe­cially in cal­i­brat­ing pac­ing.

  • —in VR.

  • dri­ver’s ed—in VR.

  • motion-track­ing play­ing chase with cat; inverse rein­force­ment learn­ing for devel­op­ing a cat mod­el; sell as cyber-com­pan­ion—in VR. VC fund­ing from Elon Meowsk. If the con­sumer mar­ket for VR cats is not there yet, piv­ot­ing to upload­ing cats, sell­ing the result­ing AIs to the DoD or Chi­na, and destroy­ing human­i­ty.

  • Uber—­for VR, using Uber. (How often are you using your head­set any­way?)

  • WiFi+blockchain++fa­cial recog­ni­tion CNNs+social net­work­s+­self-driv­ing cars+ride-shar­ing. (A sim­pler ver­sion would just be legally requir­ing or pay­ing car man­u­fac­tur­ers like Tesla to run in the back­ground on facial embed­dings, and phon­ing home to the police/security agen­cies. This would sell well in China & the UK.) Pri­or­ity orders would be exe­cuted by drones. The Sam­sung Android app will be named “Death Note”.

  • account/name reser­va­tion: reg­is­ters your favorite user­name on the top 10,000 web­sites so no one can steal it from you.

  • auto­mate plan­ning of panda bear sex & breed­ing by sequenc­ing their genomes & pre­serv­ing max­i­mum genetic diver­si­ty; land a $1b con­tract with Chi­na’s National Zoo (as pan­das are worth far more than their weight in gold to & zoo rev­enues)

  • cura­tion of & , with util­ity weights from med­ical-e­co­nomic research/surveys. -as-a-ser­vice.

  • a startup where the dress code is not hacker slacks or busi­ness suit casu­al, but suits, to reduce office pol­i­tics. Bonus: health pre­mi­ums sav­ings. And if employ­ees resist, sim­ply tell them it’s part of a diversity/anti-racism pro­gram to make it harder to dis­crim­i­nate based on skin color or appear­ance.

  • dig­i­tize motor­ized into PWAAS (Prayer-Wheel­s-As-A)-Ser­vice: highly par­al­lel head­less cloud servers ani­mate GIF prayer-wheels. This SASS (spir­i­tu­al­i­ty-as-ser­vice star­tup) can of course accept bit­coins: “Satoshis for Satoris”.

  • not-use­less spelling bee tour­na­ments: tran­scribe & pro­nounce strings. Be care­ful not to go bank­rupt fly­ing in !Kung judges for the national finals.

  • revive , build it on all the cryp­to­graphic ser­vices now avail­able (some of which go far beyond HTTPS in com­plex­i­ty), launch cut-rate cloud host­ing.

  • small drones launched from bazookas to reduce bat­tery con­sump­tion against other drones (con­sumer-level pric­ing: a rock)

  • An SDK for hack­ing smart­phone speak­ers into emit­ting , bypass­ing any OS lim­i­ta­tions; can sell to for use on dis­si­dents^W­crim­i­nal-ter­ror­ists.

  • fash­ion-as-ser­vice, NN smart­phone app to tell you “no, that’s hideous!” (eg + etc)

    Some­one sug­gested guide dogs for assist­ing their own­ers. Obvi­ously that would not work—ev­ery­one knows that dogs are par­tially col­or­blind.

  • Mon­e­tiz­ing pet ill health into rev­enue streams: exploit the tem­po­ral incon­sis­tency of peo­ple in hav­ing high in pet pur­chase costs3 but extreme price insensitivity/inelasticity in pur­chas­ing pet med­ical care by giv­ing away sick pets (espe­cially ), & get­ting vet kick­backs or direct profit-shar­ing. “CarePet­s™: The Pet You Can Care For… And Care For… And Care For…”

    Peo­ple won’t pay upfront for health­ier pets since no emo­tional bonds; but will pay almost indefi­nitely once the pets get sick. This is despite often know­ing, either first hand or watch­ing other pet own­ers spend stag­ger­ing sums on vet bills, what will hap­pen. (Inelas­tic­ity is why despite total free mar­ket in pet health, costs have actu­ally increased faster than human med­i­cine.) This asym­me­try makes them easy to mon­ey-pump, espe­cially since the fixed sup­ply of vet­eri­nar­i­ans (like human doc­tors) means it should be fea­si­ble to buy out or con­trol entire geo­graphic regions and cap­ture the increased rev­enue.

    To extract the most mon­ey, the dis­ease should be treat­able (risks own­ers opt­ing for euthanasia, and pal­lia­tive care is not profitable), dra­matic but not too dis­gust­ing (if it’s repul­sive the owner will find it eas­ier to break the emo­tional bond­s), appear within a year or two of adop­tion (to start gen­er­at­ing rev­enue as soon as the emo­tional fet­ters are solidly in place), chronic (to max­i­mize time before need­ing to sup­ply a replace­men­t), while being con­ve­niently treat­able (as peo­ples’ time can be more scarce than their mon­ey, par­tic­u­larly rich ones) in small steps (to break down the billing into small $10 or $20/month costs and avoid the own­ers real­iz­ing that the Total Cost of Own­er­ship will run into the thou­sands of dol­lars, sim­i­lar to sub­scrip­tion billing); and highly her­i­ta­ble, at least in humans (to per­mit easy devel­op­ment of a breed with extremely high risk, using well-s­tud­ied human genet­ics as a proxy esti­mate, and pro­vid­ing cover for the breed­ing pro­gram as a sci­en­tific pro­gram into devel­op­ing s for impor­tant human dis­eases).

    Thus, some­thing like late-life ter­mi­nal can­cer treat­able with a sin­gle expen­sive surgery would not work well because it will be many years before there is any rev­enue and many own­ers will sim­ply opt for euthana­sia. The ideal dis­eases in dogs & cats might be epilep­tic seizures and dia­betes: the epilep­tic seizures are dra­mat­ic, fright­en­ing, but not dis­gust­ing, both dis­eases are life­long chronic dis­eases treat­able by med­i­cine, tend to show up while young but not imme­di­ate­ly, and are highly her­i­ta­ble in humans. A good med­i­cine cost would be ~$20/month, which would yield a NPV of eas­ily $5000 (ap­prox­i­mate­ly, ).

    This strat­egy exploits: , , sub­ad­di­tiv­i­ty, mar­ket fail­ures, and . would approve great­ly.

  • con­sumer cat genomics: ini­tial hook, ances­try & bas­ket of cat psy­choac­tives (cat­nip+­va­le­ri­an+sil­vervine+honey­suck­le+thyme+Buck­bean+…) for response GWAS & pre­dict alter­na­tives to cat­nip. Then one can develop larger dataset of health/domestication/stress for breed­ing. (Boot­strap­ping from ances­try to sim­ple to com­plex traits has been a thus far suc­cess­ful pat­tern for 23andMe and con­sumer dog genomics com­pa­nies like Embark, and there are cur­rently zero com­pa­nies in the cat space despite cats being the sec­ond most pop­u­lar pet in the USA.)

    Like >95% of cat own­ers have no idea there is any­thing except cat­nip for cats, and the remain­der don’t know where to get the alter­na­tives: valer­ian is not adver­tised much for cats, I have to grow my own thyme, and I have no idea where to get the more obscure ones like Zinz­i­ba. So a bun­dle of cat psy­choac­tives is a valu­able way to amor­tize acqui­si­tion & research costs, and feed­back would quickly indi­cate which ones actu­ally work too. Com­bined with genomes for GCTA/GWAS, could quickly indi­cate what can be predicted/bred for.

    A suite of genomic tests for the known cat Mendelian dis­eases and also com­plex traits could be valu­able for breed­ers, and also feed into human med­ical research, as cats are fairly com­mon ani­mal mod­els.

  • dynamic cat toys: the behav­ioral study finds cats get bored with stan­dard cat toys because they don’t change after a ‘hunt’, appar­ently the sim­ple heuris­tic cats use for decid­ing whether hunt­ing the toy is fun; they also found that toys which fell apart were most pop­u­lar of all. (I real­ized after read­ing that that the cat toys my own cat found the most inter­est­ing were the puz­zle-treat toys, which do indeed ‘change’ in the sense of releas­ing a sin­gle bit of dry cat food.) Can we fix sta­tic cat toys to be more sat­is­fy­ing? To sim­u­late suc­cess­ful hunt con­clu­sion & avoid habituation/boredom—detachable parts, shape-mem­ory alloy/plastic, LEDs?

  • (TvTropes) mobile game but for . -ize the NFL/NBA/MLB/soccer and become a bil­lion­aire!

    Fran­chises like ///Girls Front­line demon­strate that the odd­est things can be suc­cess­fully anthro­po­mor­phized, but their appeal is inher­ently lim­ited to mil­i­tary otaku, when there are things which are far more pop­u­lar—pro­fes­sional sports. There’s no rea­son to not do moe anthro­po­mor­phiza­tion of sports­ball teams. They write them­selves—­like the Yan­kees are the oujo-sama char­ac­ter… (Given the gun/ship moe con­ven­tion of bust size=­cal­iber, I sug­gest that var­i­ous clubs/teams annual bud­gets be used for the rank­ings.)

  • web­site that lists startup ideas.

  • Cut­Fit: a cut­ting-edge new fit­ness fad, exploit­ing the fact that repair­ing bio­log­i­cal tis­sue after injury is ener­get­i­cally & meta­bol­i­cally expen­sive, requir­ing 4. Enough tis­sue dam­age could rep­re­sent a sub­stan­tial caloric sink, and if the scar­ring is vis­i­ble, serve as a costly cred­i­ble sig­nal of com­mit­ment to thin­ness & fit­ness by indi­cat­ing reg­u­lar weight loss “cuts”, dri­ving out infe­rior meth­ods of weight main­te­nance. This need not nec­es­sar­ily involve sur­gi­cal removal of tis­sue: the slow­ness of recov­ery from injuries & heavy exer­cise raises inter­est­ing ques­tions about how to opti­mally inflict dam­age on the body so as to trig­ger the exer­cise respon­se—how much exer­cise ben­e­fit is from the cost of repair? Are there more effec­tive muscle/cardiovascular dam­age meth­ods besides ‘run or lift weights’? (Care­fully tar­geted ultra­sounds? Hormetic lev­els of poi­sons like alco­hol? Some­thing elec­tri­cal? Unusual pos­tures? Phys­i­cal pres­sure? Heat/cold? Are there ways to raise basal metabolism/energy expen­di­ture which are safer than fevers, s, or )

  • ride-shar­ing improve­ment: find­ing one’s dri­ver or pas­sen­ger can be hard, espe­cially in cities or places where many peo­ple are being picked up. One sim­ple way to improve things would be to adapt the “flash­light” apps and let the ride-shar­ing app dis­play a solid block of color on the pas­sen­ger & dri­ver’s phones; then they can sim­ply hold the phones up to pro­vide a unique clear­ly-vis­i­ble iden­ti­fier vis­i­ble over long dis­tances. The dri­ver looks for a per­son with a smart­phone broad­cast­ing lime green, for exam­ple. If there are too many peo­ple in one place for the rain­bow col­ors to be unique, the color code can be two blocks of col­or, like green-pur­ple or yel­low-blue, and so on.

  • ceil­ing-mounted infrared beam emit­ter­s+guid­ance infrared cam­era, steered by seg­ment­ing CNN: it auto­mat­i­cally detects & warms up the bare skin seg­ments of female objects until ther­mal equi­lib­rium of 37.5°C is reached, ensur­ing ther­mal com­fort of women with­out requir­ing heavy cloth­ing—thereby resolv­ing the Great Office Ther­mo­stat Wars once and for all.

  • PGS screen­ing for sports recruit­ing: most sports depend on too many traits to mean­ing­fully pre­dict any­time soon from genet­ics, much less add valu­able incre­men­tal valid­ity to sim­ple phe­no­typic screen­ing meth­ods (ie “have them play a game and watch”). But there is one big-money sport which depends enor­mously on a sin­gle exag­ger­ated high­ly-her­i­ta­ble highly-PGS-predictable trait which can­not eas­ily be phe­no­typ­i­cally pre­dicted years in advance of ado­les­cence: height in bas­ket­ball. Recruiters will con­tact ran­dom youth sim­ply because they are very tall, try­ing to recruit them; there is an apoc­ryphal sta­tis­tic that some­thing like 10% of all men over 7 foot high in the USA will play in the NBA at some point, which is prob­a­bly not quite true but is touch­ing on a truth. So, a startup could

  • non-med­ical hear­ing aids but for peo­ple with —NNs for (eg “Sound of Pix­els”). Train them to delete sounds like peo­ple chew­ing, pos­si­bly cus­tomized for spe­cific miso­phon­ics, then they can be worn as nec­es­sary in social sit­u­a­tions. (These algo­rithms are expen­sive to run, which is why they are not used in stan­dard hear­ing aids despite hav­ing use­ful hard­ware like direc­tional micro­phones, but a spe­cialty device intended for spe­cific sit­u­a­tions can afford much shorter bat­tery life and/or larger bat­ter­ies.)

Tom Swifties

“It is not strength, but art, obtains the prize, / And to be swifty is less than to be wise.”

Alexan­der P.

Some­times, at night, SSC whis­pers to me, “go and make at every­one you ever loved. Pun it all down.” To exor­cise the demon, I make GPT-3 write them or I put them here instead:

  • “I hope Jared Kush­ner is able to improve the gov­ern­men­t’s Mid­dle East­ern pol­i­cy!”, Tom said sun­ni­ly.

  • “All these ran­dom fac­toids just about add up to a hill of beans”, Tom observed in a nor­mal tone.

  • “Eureka! I’ve fig­ured out how the brain works (again)!”, Tom hinted online.

  • “Help, I’ve lost the two sil­ver coins I keep in my pocket and now I’m broke as a Greek!” Tom said par­a­dig­mat­i­cal­ly.

  • “These end­less flag debates have made me cross and see­ing red”, Tom said vexed­ly.

  • “Would any­one like some lamb or omelettes for brunch?”, Tom mod­estly pro­posed.

  • “I’m now dead­-broke and unem­ployed”, Tom said in pub­lic dole­ful­ly.

  • “Darth Vader’s use of cor­po­ral pun­ish­ment is prob­a­bly bad for morale”, Tom said offhand­ed­ly.

  • “Impres­sive indeed, but there’s some­thing I ought to tell you—I am not left­-handed either”, Tom pointed out offhand­ed­ly.

    “I actu­ally am a skilled duelist and trick shooter too”, offend­edly shot out Tom offhand­edly

  • “This new male con­tra­cep­tive is a rev­o­lu­tion­ary break­through!” Tom point­lessly ejac­u­lat­ed.

  • “She has an unpleas­ant demeanor”, Tom said misog­y­nis­ti­cal­ly.

  • “I won­der if a cer­tain per­son who should not have ate those spe­cial cook­ies”, Tom insin­u­at­ed.

  • “I hardly have an appetite for this Soy­lent”, Tom said insipid­ly.

  • “No, thanks, but I pre­fer caffeinated bev­er­ages”, Tom said tee­to­tal­ing.

  • “They put too many birds into the zoo and it over­flew!” Tom avidly squawked.

  • “Some dog has unfor­tu­nately destroyed my flower gar­den, oh well”, said Tom lack­adaisi­cal­ly.

  • “I can’t think of any­thing to write”, Tom said blankly.

    “Why do you ask?”, Tom said queru­lous­ly.

    “Because I can think of lit­tle else than Tom Swifties”, Tom said auto­log­i­cal­ly.

  • “I refuse to dis­cuss what I donate for a liv­ing!”, Tom said spunki­ly.

  • “I got a great deal on my new Ray­bans”, said Tom shadi­ly.

  • “You were sup­posed to get my Chi­nese car­toons in August!” Tom belat­ed.

  • “I should have held onto my oars bet­ter”, Tom tholed.

  • “Our hand­made tra­di­tional arti­sanal cer­ti­fied organic baked goods have a secret vital ingre­di­ent”, Tom said gluti­nous­ly.

  • “I’m not sure I believe the offi­cial story about cur­rant events”, Tom doubted fruit­ful­ly.

  • “I’ve taken up weightlift­ing”, Tom grunt­ed.

    [LATER]: “I’m absolutely gut­ted today”, Tom belly­ached.

  • “We’re fresh out of wheat flour, but I think that has such a bland fla­vor, don’t you?” Tom said wry­ly.

  • “I am unable to pro­vide an exam­ple of that in my dis­cus­sion”, Tom ablat­ed.

  • “I’m dis­gusted watch­ing peo­ple on trains stroke their iPads with those strange ges­tures!”, Tom said spunki­ly.

Nominative determinism

Noticed instances of in the wild:

Miscellaneous

  • A rid­dle5: “What activ­ity is this?”

    • "You can do this activ­ity at the begin­ning or end of a horse race but not the mid­dle.
    • This activ­ity is accept­able to do in pub­lic but unac­cept­able in your bed­room.
    • It is ille­gal to have sex with an ani­mal before this activ­i­ty, but legal after­wards.
    • Some feel eth­i­cally required to do it, while oth­ers feel required not to.
    • It is eth­i­cal to do it to an ani­mal, but not to your grand­par­ents.
    • It’s accept­able to do it for your­self or fam­i­ly, but not strangers; how­ev­er, it’s accept­able for strangers to do it for strangers/you….
    • You can legally do it reli­ably and pain­lessly for ani­mals, but only legally do it unre­li­ably and extremely painfully for human­s."6
  • Band names inspired by colds:

    “Angels & Aspi­ra­tions”; “The Breath Boys”; “Gesünd­heit”; “Infec­tious Clown Posse”; “Jaw Divi­sion”/“Joy Omis­sion”; “Mucus Machine”; “The Nasal Dilaters”; “Phlegm Phigh­ters”; “Snot­son and the Boogers”; “Trigem­i­nal Treach­ery”; “Tylenol Tears”; “WBC48”; “Flugees”; “Rage Against The Mucus”; “Lemon­tonix”; “Croak­w­erk”; “God Bless You Sick Emperor”; “This Is Dimetapp”; “And You Will Know Us By The Trail Of DM”; “Explo­sions in the Sinus”; “Roxy Mucus”; “Sneezy Top”; “Amy Benadryl”; “Hachoo-ne Miku”; “Megurine Puka”; “Sniffles of a Down”; “Link­in’ Puke”; “Lil Oozy”; “Germs & Blows”; “Hall­s-y”; “Miley Sinus”; “Justin Sneezer”; “Johnny Cough”; “Milky Chunks”; “Plan Slime From Outer Space”; “Cold­pray”; “The Long Dark Pee­time of the Soul”; “Puke Floyd”; “Honkie and the Blows­niff”

  • Things which sound like war crimes but aren’t: “”; “”; “Eng­lish break­fast”; “


  1. Rejected sta­tis­tics pro­pa­ganda slo­gans: “Togeth­er, we can do it! Keep­’em sam­pling!” · “Data is a weapon—­don’t waste it.” · “When you write alone, you write with bias! Join a data-shar­ing site today!” · “Ser­vice on the home front: bug reports; doc fix­es; Stack Over­flow answers. There’s a job for every Amer­i­can in these civil­ian efforts!” · “Unit­ed, our sta­tis­ti­cal power is strong. Unit­ed, we will win.” · “Men who know say no to sexy press releases” · “Tokio Kid say: ‘Much mea­sure­ment error make so-o-o-o hap­py! Sank you!’” · “It is far bet­ter to face the Type I errors than to be killed at home by Type IIs. Join a clin­i­cal trial at once.” · “Keep this hor­ror from your home: invest 10% in at once.” · “Old sta­tis­ti­cians never die, they just fade away and become non-sig­nifi­cant (p > 0.05).” · “I hate ran­dom error as only a sta­tis­ti­cian who has lived it can, only as one who has seen its bru­tal­i­ty, its stu­pid­i­ty.” · “It was close; but that’s the way it is in sequen­tial tri­als. You win or lose, live or die—and the differ­ence is just an eye­lash.” · “I can­not fore­cast to you the action of Ran­dom­iza­tion. It is a rid­dle wrapped in a mys­tery inside an enig­ma.” · “I have noth­ing to offer but Bayes, toil, tears and sweat.” · “Worker threads of the world, unite! You have noth­ing to lose but your Markov chains.” · “I will recur.”↩︎

  2. A vari­ant on a say­ing in the auto­mo­tive indus­try, along the lines of “The per­fect race car falls apart while cross­ing the fin­ish line” or “The per­fect rac­ing car crosses the fin­ish line first and sub­se­quently falls into its com­po­nent parts” (but there are innu­mer­able vari­ants). It has been attrib­uted to, among oth­ers, , an anony­mous news­pa­per pho­to­graph cap­tion, , and by way of (un­known source):

    …in ’59 I ran out of brakes four times—and I don’t mean they did­n’t work very well, I mean I had none. Like the main oil line had sheared. You know, so that oil, you know, when you put your foot on the floor, the oil just went squirt­ing out into the atmos­phere. I’d always believed that Colin was close to genius in his design abil­ity and every­thing, if he could just get over this fail­ing of his of mak­ing things too bloody light. I mean, Col­in’s idea of a Grand Prix car was it should win the race and, as it crossed the fin­ish­ing line, it should col­lapse in a heap of bits. If it did­n’t do that, it was built too strong­ly.

    and Peter Dron (2002):

    Chap­man prob­a­bly did die, aged only 54, of a heart attack. He had always had weight prob­lems (hence his nick­name), cru­elly iron­i­cal in one so fanat­i­cal about light­ness: his the­ory of the per­fect rac­ing car was that, sev­eral yards after tak­ing the che­quered flag for vic­to­ry, it should simul­ta­ne­ously run out of petrol and dis­in­te­grate. Those dri­vers who sur­vived when their Lotuses crashed rather ear­lier than this, due to com­po­nent fail­ure, some­times felt that he took this prin­ci­ple too far.

    ↩︎
  3. This asym­me­try might explain the apa­thy of pet own­ers to pay­ing for breed­ing or genetic engi­neer­ing, in con­trast to farm­ers who are will­ing to pay large sums for the devel­op­ment of supe­rior vari­eties. It is easy to imag­ine the com­mer­cial devel­op­ment of cat breeds devel­oped to be extremely healthy, long-lived, with good tem­pera­ments, cat­nip respon­se, no known genetic dis­eases (with tests reg­u­larly deployed as devel­ope­d), the kit­tens raised with opti­mal meth­ods such as large amounts of han­dling & expo­sure to stres­sors like other cats or dogs dur­ing the crit­i­cal early devel­op­men­tal win­dows of plas­tic­i­ty, and sold spayed/neutered to end con­sumers to pro­tect their IP. Such rig­or­ous pro­grams are com­mon in live­stock or plant breed­ing, and the prod­ucts often sold to end con­sumers such as hob­by­ist gar­den­ers. Yet, they are totally absent from pet own­er­ship.

    It is strik­ing that most selec­tive breed­ing of cats focuses on fur appear­ance, cross­ing domes­tic cats with a vari­ety of other felid species to pro­duce breeds with exotic appear­ances (com­pare sales of the ), or in the case of & Felix Pets, adver­tis­ing (ie sell­ing to those for whom a nor­mal cat would be excru­ci­at­ing & pos­si­bly dan­ger­ous to live with). This lack of inter­est in cat breeds opti­mized is despite the large, and ever esca­lat­ing, health­care costs for pets, which imply that pay­ing thou­sands of dol­lars for a par­tic­u­larly healthy & long-lived pet could pay off.↩︎

  4. Apell et al 2012:

    The liv­er, brain, heart and kid­neys make up for almost 70% of this basic meta­bolic rate W but only about 6% of the body mass M To cal­cu­late gross num­bers we can use a food intake equiv­a­lent to 6 MJ/day (Ma­han 2000). This cor­re­sponds to an aver­age power of 70 W or 1 mW/g = 1 pW/ng of body weight. A typ­i­cal cell is around 1 ng in weight giv­ing the pW scale for cell metab­o­lism. In phys­i­cal terms wounds need sub­stan­tial energy resources when pro­cess­ing the new mate­r­ial needed to fill the wound cav­i­ty. In the first instance it takes a lot of extra energy to boost the immune sys­tem to beat pos­si­ble inflam­ma­tions and clear up the debris. The energy needed for the tis­sue ingrowth can be decom­posed into two parts. The energy stored in the new mate­r­ial and struc­tures being built and the one used to main­tain old and new tis­sue. The first one dom­i­nates dur­ing the tis­sue growth process and depends nat­u­rally on tis­sue com­po­si­tion. How­ever if we know the frac­tions of fat and pro­teins one can use their heat of com­bus­tion to cal­cu­late a num­ber for the tis­sue con­cerned. This works out to a mean energy require­ment for growth of the order of 20 kJ per gram of tis­sue deposited or 60μJ for an aver­age cell (3–5 nanograms) (Ma­lina 2004 [Growth, mat­u­ra­tion, and phys­i­cal activ­i­ty, sec­ond edi­tion]). This cor­re­sponds to a power of 6 watts if con­verted to one hour. Being of the order of 10% of the total basal power we see that wound heal­ing has to take days if not to be a too heavy load on the sys­tem. Another way is to boost the local metab­o­lism where it is well-known that in peri­ods of ill­ness or injury we need to increase the energy intake with up to 40–50%. This means in phys­i­cal terms that wounds acts like sub­stan­tial energy sinks…­Tak­ing into account that fibrob­last cells are one of the major play­ers in mak­ing new tis­sue we find from Table I and the energy require­ment above of 20kJ/g to make new tis­sue that a time of the order of 10 days will have to elapse to pro­vide the nec­es­sary energy WT. This is defi­nitely in line with mea­sured heal­ing times. Notice how­ever that this time is basi­cally much longer than the typ­i­cal cel­l-dou­bling time or the time it takes a fibrob­last cell to move 10 times its own size (10h), which is typ­i­cally the dis­tance over which cells have no idea that a wound is present in their neigh­bour­hood.

    ↩︎
  5. Inspired by an exer­cise on pg5–6 of 2008, On Being Cer­tain: Believ­ing You Are Right Even When You’re Not, which is sim­i­lar to per­cep­tual illu­sions like :

    To begin our dis­cus­sion of the feel­ing of know­ing, read the fol­low­ing excerpt at nor­mal speed. Don’t skim, give up halfway through, or skip to the expla­na­tion. Because this expe­ri­ence can’t be dupli­cated once you know the expla­na­tion, take a moment to ask your­self how you feel about the para­graph. After read­ing the clar­i­fy­ing word, reread the para­graph. As you do so, please pay close atten­tion to the shifts in your men­tal state and your feel­ing about the para­graph.

    A news­pa­per is bet­ter than a mag­a­zine. A seashore is a bet­ter place than the street. At first, it is bet­ter to run than to walk. You may have to try sev­eral times. It takes some skill, but it is easy to learn. Even young chil­dren can enjoy it. Once suc­cess­ful, com­pli­ca­tions are min­i­mal. Birds sel­dom get too close. Rain, how­ev­er, soaks in very fast. Too many peo­ple doing the same thing can also cause prob­lems. One needs lots of room. If there are no com­pli­ca­tions, it can be very peace­ful. A rock will serve as an anchor. If things break loose from it, how­ev­er, you will not get a sec­ond chance.

    Is this para­graph com­pre­hen­si­ble or mean­ing­less? Feel your mind sort through poten­tial expla­na­tions. Now watch what hap­pens with the pre­sen­ta­tion of a sin­gle word: kite. As you reread the para­graph, feel the prior dis­com­fort of some­thing amiss shift­ing to a pleas­ing sense of right­ness. Every­thing fits; every sen­tence works and has mean­ing. Reread the para­graph again; it is impos­si­ble to regain the sense of not under­stand­ing. In an instant, with­out due con­scious delib­er­a­tion, the para­graph has been irre­versibly infused with a feel­ing of know­ing.

    Try to imag­ine other inter­pre­ta­tions for the para­graph. Sup­pose I tell you that this is a col­lab­o­ra­tive poem writ­ten by a third-grade class, or a col­lage of strung-to­gether for­tune cookie quotes. Your mind balks. The pres­ence of this feel­ing of know­ing makes con­tem­plat­ing alter­na­tives phys­i­cally diffi­cult.

    ↩︎
  6. The answer is “”. See , and WP on /.↩︎