Skip to main content

Startup Ideas

Proposals for new technologies, businesses, startups, satirical or serious.

For serious ideas, see my project wishlist; and also my list of open research questions.

CYOA AI Games

AI-based text adventure games, but using the Choose-Your-Own-Adventure structure for much greater compute-efficiency & ratcheting up quality using player choices rather than running every game ab initio.

CO2 Coin

CO2 Coin: Decentralized Carbon Capture Blockchains: a hypothetical way to do carbon capture in a meaningfully-decentralized incentive-compatible way would be a slashing game around the creation of tokens about instances of mineralization. Mineralization can be easily checked, enabling a slashing economic game: (literal) miners report the creation of mineral deposits to claim CO2 tokens to sell to do-gooders while putting up a security deposit, and skeptics can challenge fraudulent creations to try to claim the security deposits, and because of the physical properties of mineralization can be easily audited by a (trusted?) third-party to adjudicate these disputes. In equilibrium, miners simply honestly report their mineralization progress, claiming a stream of tokens which are sold to fund further mineralization.

Lipreading Website

There are no good websites for learning how to lipread, despite widespread hearing impairment and the aging of the US population. Looking at some numbers, it seems like a potentially profitable niche?

[Now exists as Lipreading.org.]

As far as I can tell, there is no free resource for learning how to lip read, much less free online resources.

Learning to lip read is basically:

  1. watch a video with obscured audio

  2. guess what you think they said

  3. be corrected

  4. go to #1

This is eminently doable as a website: YouTube for hosting videos, Amazon Mechanical Turk or similar services for generating Free videos, and perhaps a SRS algorithm for scheduling periodic reviews videos of particular words or sentences.

Lip reading is useful to know. There are roughly 28 million people in the US with hearing issues and as the Baby Boomers age and lose hearing, many will want to learn; estimates of Baby Boomers who will have any degree of hearing loss range 20–60%. See:

For older Americans, the rate is 63%.

Nor is the loss limited to Baby Boomers; a report in the August 2010 issue of the American Medical Association estimated that over the 15 years 1995–2010, teen hearing loss rate increased 30%1, to a total of all teens with detectable hearing loss of 19.5% of 12-19 year olds (and a similar increase in the number with mild hearing loss).2

The small industry of lip reading and the international scattering of lip reading classes shows that people will pay hundreds of dollars and go places to learn it.

Costs

Optimistically, one-time >$129.69$1002015 for content, $25.94$202015/month then on, and a substantial time investment in putting together a site and a process for acquiring or creating video.

Technical

Hosting

Hosting cost are neglible. Assuming videos are hosted on YouTube or Amazon S3, the website would require extremely little bandwidth and accommodate >1000 users at ~20$ a month.

Assume a webpage requires 100KB to be loaded (very pessimistic), and that an user spends 1 hour a day using the website (30 hours a month), going to a new page every minute. That user will use 100KB × (30 × 60), or 180000KB, or 180MB of bandwidth. Linode’s cheapest offering at [$20]($2015)/month pays for 200GB of bandwidth; or ~1112 users. A domain name costs ~$12.97$102015 a year, or ~$1.3$12015 a month.

More reasonable would be assuming 10KB per pageload, and 10 hours a month, cutting the per-user bandwidth down to 6MB, and assuming fewer than 1000 users; then hosting could be even cheaper. DreamHost is known for screwing over its more-demanding customers, but should be reliable enough here; their hosting is $11.67$92015 a month.

Coding

Obviously a site custom-made for lip reading & very user-friendly doesn’t exist. I’d have to code one or reuse some framework, though offhand I don’t know of any really suited for the task. It’d be a big coding task - at least dozens of hours to learn the specific technologies and build a prototype. But then, I can’t really count my own time as a cost - I’d just spend the time reading elsewise.

Marketing

Unknown. These sorts of sites seem to do best with word of mouth marketing, so who knows? Maybe just time.

Lipreading Content

The content is the wildcard. There are a couple possible sources:

  • There’s a cottage industry of books and occasional CDs/DVDs, whose copyright obviously would be far too expensive to purchase.

  • Hiring professionals to make lip movements also is obviously right out. To make it worth their while and to get at least 10 hours of material would take thousands of dollars.

  • Online freelancing sites. I have a theory that one doesn’t want professionals because one intends to use lip reading in real life, to read the lips of the ‘amateurs’ one interacts with.

    • I mentioned Mechanical Turk, but that may not be appropriate; many Turkers do not have cameras or webcams, and it may not be doable to ask them to submit videos through Amazon, but Turkers could definitely be used to verify that the person in a clip is saying the things they are supposed to say. (This would cost ~10¢ per review, and usually one double-checks with multiple Turkers, so 20¢ a clip.)

    • Other freelancing sites like guru.com list video/photo people working for $25.94$202015$51.88$402015 an hour. I figure that means that amateurs in both department will be no more than half that, $12.97$102015/hour. 10 hours of content then would be ~$129.69$1002015.

Revenue

Ads, obviously. A competitor would be lipreading.com; so that’s a reasonable starting point. With zero effort at doing anything other than selling a DVD, some estimates of its ad revenue are 4¢ to $2.22 a day. At hosting costs of $27.24$212015 a month or <75¢ a day, the site could at least pay its on-going expenses.

Indoor Tree-Climbing Gym

Climbing walls, and gyms built around climbing walls, are one of the highest-growth sectors in American athletes since the first Seattle climbing gym in 1987, growing with the popular of rock climbing. Climbing walls are climate-controlled, available everywhere (particularly inside cities), let newbies walk in off the street to climb, generally safer than real rock climbing, and regularly change the arrangements on their walls to offer new challenges to climbers. Despite the popularity of tree climbing among children (it appeals to something primal in our monkey soul), and tree houses/tree sitting, there is not anything really similar either indoor or outdoors—a fullgrown adult human weighs too much for most branches, and there are not many dense forests within easy reach of population centers which would allow treetop traverses, and the guardians of surviving giants like cedars or conifers are jealous ones. (Ropes courses & rope pyramids, as the names indicate, offer a different kind of experience.)

Just as indoor climbing gyms offer a controlled wall to ascend, an “indoor tree-climbing gym” can offer a controlled tree—or better, forest—to climb & traverse. A climbing wall provides a primarily 2D challenge, but a tree or forest provides a 3D challenge. Climbers can be challenged by difficult limbs set far apart, or be challenged to reach waypoints in a certain order, or (crossing over with parkour) finish within a certain time. Other sports could probably be done but “in the trees”. (How about paintballing? It’s worth noting some of the most memorable martial arts film scenes take place in trees: ninja fiction like Naruto is full of tree-level travel or combat, there’s Ong Bak, and the most famous scene in Crouching Tiger, Hidden Dragon is surely the treetop duel.) Or how about book nooks for taking a break?

A series of pillars in fixed positions has the disadvantage that, like a climbing wall, they cannot be easily moved. Doodads can be affixed or moved, but the overall shape and position remains the same. That is part of why a climbing wall can be so boring. A forest could be more mobile: the ‘trunks’ could be attached to cables hung from the ceiling (perhaps supported primarily by the ceiling?), and locked into place on the ground, to be unlocked and moved around as necessary. Limbs can be inserted & removed from sockets inside the ‘trunks’.

Aside from the question of what sort of sport it ought to be and how to provide challenge (difficult to do in the absence of actual play—I’m sure Quidditch had to be modified based on experience to find a way through the forking paths to a success), safety is a big question here. In rock climbing gyms, bouldering is done low enough that no particular safety equipment beyond mats to fall on is felt necessary; for climbing up walls, one can be roped into a harness and caught, as the 2D wall does not obstruct the rope no matter how one climbs.

Is treeclimbing safe enough to need no equipment? Would regularly clipping/unclipping onto a trunk be too much friction? Can one be roped into, say, a gantry-like grid on the ceiling without losing too much freedom? Perhaps if the limbs are flexible enough one can just pull the rope behind one? Or perhaps some entirely new approach, similar to “auto belays”, which would give the rope a lot of slack during normal use but then rapidly tighten if the user is detected to be falling? (Indeed, with a sufficiently smart rope system, one could—ninjas again—leap from tree to tree, machine-assisted.) Or perhaps some sort of “sticky hands” system like magnets, so the climber can always be attached? Gurkenglas suggests inflatable trees, perhaps filled with some sort of non-Newtonian fluid, which could be actively deflated: if a climber falls, the rigid trees around them are all deflated so they fall into floppy tree limbs & a soft central mass deflating into the floor like a giant airbag.

Streamer Memorabilia

One generation’s ephemeral pop culture junk is another generation’s beloved collectible—baseball cards, Pokemon/Magic cards, Hollywood memorabilia (particularly props), comic books, animation cels… As universal as they were, things like the Honus Wagner or Mickey Mantle baseball cards are so valuable in part because out of neglect and contempt or deliberate destruction of excess cards, few survived. There are countless examples of beloved artifacts like the Ruby slippers from Wizard of Oz or anime/animation cels which narrowly escaped the trashcan or furnace, sometimes because employees just ‘stole’ them, trading early on at trivial prices. For example, the ruby slippers survived because of an unprecedented auction, where they were but a handful of items among the collected contents of “seven soundstages” with “>350,000 costumes alone”—Debbie Reynolds purchased a tiny fraction of the items offered, and her collection sold for ~$35.06$252011m. Even as recently as the late 1980s, Planet Hollywood was able to find countless iconic Hollywood props for next to nothing (eg. PH found the ax Jack Nicholson wields in The Shining sitting in a garden shed, and traded the owner a replacement). If you had somehow known in advance how popular these things would get, you could have purchased or acquired a wide variety of items for ~$0, and sold them decades later for hundreds of thousands or sometimes millions of dollars. It is unlikely that all possible memorabilia & collectibles have been invented and we are at the end of history (of collecting), given that the psychology of collecting is insatiable and fundamentally requires turnover; so there are almost certainly collectible areas being born right now.

What is the future collectible memorabilia of today’s pop culture c. 2021? What is extremely popular among young people, formative, connected to their identities, but looked down on as ephemeral, unimportant, lowbrow, such that the idea of preserving or collecting regarded as absurd?

I suggest streaming: YouTube streaming, Twitch streaming, whatever. They are all enormously successful, some of the most consumed media on the planet, some with more recognition than Hollywood stars—but without their trappings. You can go to Christie’s to buy Tom Cruise Top Gun memorabilia, but where do you buy Logan Paul props? Nowhere.

So, a speculative investment strategy would be to simply start buying up and, as needs be, making physical memorabilia3. Caleb’s PS4 controller, which he’s used for thousands of hours, imbuing it with his manna, is priceless right now ie worth $0—but someday someone may pay $$$ for it. (Controllers seem particularly ideal as being small, sturdy, non-perishable, the main ‘tool’ of a gamer, and themselves often iconic; and they can be autographed.) Buy Logan Paul’s old smartphone, Belle Delphine’s bathwater, Ninja’s headband: the sky’s the limit. Right now, what do they do with all their old t-shirts or panties or costumes or headsets or microphones? They throw it out. What a waste!

None of this requires particularly expensive storage, so I expect the primary cost is the labor cost of negotiating with streamers on a case by case basis. The most famous streamers are probably too difficult to contact & would demand too much out of a general sense of exclusivity, but streaming pays infamously little for most streamers, and there is probably a sweet spot for streamers where they are successful enough that they may be truly famous some day, but still budget-constrained enough that being able to sell a few hundred dollars in otherwise-discarded objects will help make ends meet now. (Maybe somewhere in the hundreds of thousands to millions of subscriber regime?) Given good execution, mid-tier streamers might start coming to the collector: “my streamer friend told me that if I needed cash fast, you’d buy my stuff.” Once streamers are saturated, one can branch out into other social media stars and mediums, although I expect those to be less appealing as investments as their memorabilia would be more fairly priced.

A good followup to collecting objects is developing a sideline in verifying & grading objects. One has the established connections with streamers, one may have owned the objects in question, and a good grading or auditing system has always been critical in developing collector confidence to build a market—baseball cards would not exist without it, or look at the high “fine art” market (no matter how often art historians are fooled by forgers, their rubberstamps are necessary to create the aura of collectibility and help everyone keep the masquerade going). Indeed, in the classic “sell shovels to miners” business logic, this optional sideline might wind up being an even better business than the actual objects themselves (who’d’ve ever thought that a Pokemon card-grading company would ever sell for $500m?).

Finance-wise, it is highly risky to acquire objects for sale in potentially several decades due to the compounding opportunity cost & risk. Still, given how spectacularly high prices of collectibles can rise, it is feasible. For example, if an object is sold in 30 years, then to compete with baseline stock market returns of 7% annually, it need only increase 7.6× (1.0730) to break-even; it is not implausible to imagine, say, a streamer’s smartphone costing $100 increasing to $760 as a collectible to their grownup fans. The real profit would come from the long tail of objects skyrocketing in value to hundreds of thousands of dollars; this is difficult to quantify. But perhaps the real value of such an investment strategy is diversification—one of the main drivers of investment interest in “alternative investment” (weird niches like art or wine or stamps) is simply that returns tend to be uncorrelated, reducing risk, and many investors (particularly institutions) will pay for that. And it’s hard to imagine investments less correlated with existing sectors than betting on collectibles that don’t yet exist!

Mouse Longevity Perpetual Swaps

Aging is the most destructive of all diseases, and finding drugs that could slow aging would be incalculable public goods. Unfortunately, being multi-causal, slow, and subtle, aging is even harder than most diseases to design good interventions to try. (For aging, as with many simpler diseases, cures may precede understanding; we may only know what aging is after we have found a way to slow it, and can study that instead.) In such a situation, the hypothesis-free brute-force approach of “screen as many random drugs as possible to look for anti-aging effects” offers a chance for progress; ideally, we’d run an adaptive trial for “best arm identification”, which would start with many drugs, and as mice die and their survival curve on drugs can be inferred, gradually drop the worst-performing drugs to allocate new mice to the best-performing drugs—homing in on the best possible drug while spending as few mice as possible on mediocre or useless drugs (whose true effect is of no value to know). This has been done to a limited extent in mice already, but it has its own problem: it trades theory for data, and, well, data = money. As it’s so hard to get any funding to study aging compared to sexier diseases like Alzheimer’s (regardless of how badly research in those have failed), the mouse screens for anti-aging drugs have been grossly inadequate in sample size and drug selection, and one will not be surprised to hear that we have not found a cure for aging.

Where can we get more information on life-extension drugs? One idea is prediction markets: turn each drug into a prediction market, and traders can buy or sell shares on specific drugs—perhaps something hard endpoints like FDA approval for anti-aging by year 20XX, the results of meta-analyses on human clinical trials, or increases in population life-expectancy conditional on a drug being widely used. This is in fact one of the first proposed applications by Robin Hanson for prediction markets: use prediction markets to track various scientific hypotheses, as “ideas futures”. In practice, various attempts to prediction-market scientific hypotheses, whether on Intrade long ago or Metaculus now, have suffered from severe shortages of activity and liquidity; most science-related markets are totally moribund, and will take decades to settle with a final result. (Directly subsidizing long-term prediction markets hasn’t worked out too well either, the challenges are too great for some small subsidies to incentivize major trader investments of capital & research.) Sports gambling markets can reach wager totals of billions of dollars on single sporting matches, and bookie odds typically become extremely accurate; science markets typically won’t see even a million in transactions total, and while I’m not aware of any specific evaluations of their accuracy, I assume the worst.

To speed things up, one can use more short-term metrics: perhaps take specific sample sizes of mice, administer n drugs, and define markets which pay out based on which drug has the last surviving mouse (a dead pool), over/under on median lifespan in days, or ‘horse race’ shares that pay out everything to the drug that ‘wins’ the most mouse-days in a given sample. These metrics are only proxies for what we want (human results), and noisy ones too (given the small effect sizes we expect from random drugs—if any ordinary drug were a fountain of youth which halted aging, someone would have noticed by now!). But at least you’ll get results within a year or 3, and prediction markets do work on that timescale (eg. US presidential election markets). It’s a short enough duration that opportunity cost/interest rates aren’t too bad, traders can remain interested, gamblers who want entertainment get price swings & resolutions reasonably quickly. But even if it works to some degree, that does not mean it works well. The markets are still small, highly fragmented over each cohort and drug, with little particular incentive to trade, and not much gambling fun. Can we go even faster?

“Jesse, what the f— are you talking about?”

One inspiration comes from the monstrous success of the BitMEX cryptocurrency exchange. Where decentralized prediction markets like Augur continue to languish and centralized crypto prediction markets like Polymarket remain relatively small players akin to Intrade of yore, BitMEX became the most active cryptocurrency exchange in the world, reportedly hitting $1 trillion in notional derivative trades by 2019, largely on the strength of a weird kind of future called a “perpetual swap”. Perpetual swaps (Shiller1993, Gehr1988, “What are Perpetual Swaps?”) try to solve a similar problem that futures markets have: the straightforward way to do futures is to define a whole bunch of separate futures, like 1-year/5-year/10-year/30-year etc, and let each one trade as it will. This may be OK for things like US Treasuries or major currencies, but it doesn’t work too well for things in longer-term, illiquid, or small markets, such as housing. (Every house or piece of real estate is slightly different, after all, because they necessarily differ in location.) Thus Shiller’s proposal for enabling derivatives on such things.

So, what if we tried to borrow the logic of perpetual swaps, but on mouse lives? A research organization can set up a mouse longevity project powered by an associated crypto prediction market; the prediction market analyzes and funds the sample creation of mice, providing a cut of transaction fees, and incentivized by charitable goals, profit-seeking traders, and sheer entertainment value.

A prediction market on individual mouse-days would resemble an actively-traded tontine: perhaps each day a fraction of the starting value gets paid out between owners of each drug, until prices rebalance? Each day is a gamble, which will win or lose. Owners of successful drug A can look forward to a stream of ‘dividends’ paid by the bettors on failing drugs B and C, or they can speculate on D being underpriced by random fluctuations and catching up. As mice die and the sample falls, traders could purchase new contracts, whose face-value is equivalent to the total cost of maintaining 1 mouse for its predicted-by-the-adaptive-algorithm lifespan: if the algorithm is wrong, the purchase will be +EV; if it’s right, well, the trader takes an opportunity cost loss and has increased n by 1. These new mice could also be allocated to a new drug as well. (Other model organisms like cats can be used, there’s nothing mouse-specific here; one could layer on futures for external validity.) Traders are thus incentivized to find new drug candidates to purchase mice samples of, exploiting the a priori mispricing. With functioning markets, it’s easy for a philanthropist or enthusiasts to donate to the cause: simply buying a set of samples proportionally will not move the prices, but increase the sample size.

Alternately, if the logic of contract owners paying each other doesn’t quite work out, it can be between the research organization and contract owners: contracts pay up front the cost of a mouse and a drug, and the research organization pays a small dividend each day the mouse remains alive, pegged to be break-even at mean baseline mouse lifespan. (I admit the details of the mechanism design here are fuzzy and need working out by a better trader than I am to figure out a system which is incentive-compatible and incentivized to find the best arm efficiently.)

In the best-case scenario, wagering on mice & drugs becomes like sports gambling, with traders cheering on favorite mice and wagers, with public memorial services when Mouse #516 dies after beating the spread by 3 standard deviations, memeification of drugs (“Scumbag Rapamycin / Extends Lifespan 10% / Whole Batch Dies of Mouse-Monia”), drawing in millions or hundreds of millions of dollars to screen hundreds or thousands of prospective candidates, providing definitive evidence at large n of a dozen or so, which can be moved immediately into human clinical trials and with a battery of independent interventions on lifespan, power understanding of the causal pathways of aging.

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 Hall et al 2002 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?

Pet Lemon Market

And that’s the reason why we pay up willingly. There is no price on love. Lenny racked up a bill of just more than $6,365 for his errant corn on the cob, but I’d have paid twice, three times, that to ensure he was well again. In fact, I’d have paid every penny I have, which, thinking about it, is not a bad business model for a veterinary practice.

Simon Kelner

If we judge love by the majority of its results, it resembles hatred more than friendship.

François de La Rochefoucauld, Maximes 72

Many people claim that means are more important than ends, intentions more important than outcomes, that “death gives meaning to life”; it would be tragic if people were deprived of the opportunity to show how much they cared, and so there may be a modest business in ensuring them the opportunity to be virtuous. Monetizing pet ill health into revenue streams: exploit the temporal inconsistency of people in having high elasticity in pet purchase costs4 but extreme price insensitivity/inelasticity in purchasing pet medical care by giving away sick pets (especially bulldogs), & 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 then; but will pay almost indefinitely once the pets get sick. (Prevention is priceless, which is a synonym for “worthless”.) 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 a nearly totally-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 means it should be feasible to buy out or control entire geographic regions and capture the increased revenue (like the human doctor cartel).

To extract the most money, the disease should be treatable (untreatable 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 animal models of diseases 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; and 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 $5,000 (approximately, (20 × 12) / log(1.05)).

This strategy exploits: dynamic & temporal inconsistency, hyperbolic discounting, subadditivity, market failures, baby/furbaby substitution, and virtue signaling. Swift & Thamiel would approve.

Miscellaneous

  • Bonsai 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 GANs, sumi-e painting engines (eg. Xie et al 2012), 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.

  • CNN for font kerning. By far most tedious, labor-intensive/time-money-expensive, rote part of high-quality font creation; but also most I-know-it-when-I-see-it, non-semantic, and easy to create millions of training samples by jittering random text set in existing kerned fonts (k > 50,000 is easily doable by downloading Internet fonts) and training to recover the human-expert-defined kernings. (A reader links to a tool released in July 2021, Kern On, which uses sets of user-specified example kernings as prototypes for suggesting new kerns, lauding it as enabling weeks of hand-kerning in half an hour; unclear what algorithm it uses.)

  • non-medical hearing aids but for people with misophonia—NNs for signal separation (eg. “Sound of Pixels”).

    Hearing aids (or cheaper equivalents like earpods+smartphones) currently are used occasionally for misophonics but in crude ways: settings which add loud white/brown noise to cover up bad sounds, dropping entire frequency ranges like cutting off high frequency ranges, or to turn off to block all sound (as a more convenient earplug). This falls short of what is already possible in audio processing.

    Train them to delete specific sounds like people chewing, possibly customized for specific misophonics, then they can be worn as necessary in social situations without loss of information. (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, and which doesn’t need to amplify at all, can afford much shorter battery life and/or larger batteries.)

  • 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 ~2.1m 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 screen kids’ SNPs and sell contact info to NBA recruiters seeking to build a relationship & loyalty years ahead of competitors.

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

    • low-tech version: insulated zentai suits—modern insulating fabrics are very insulating, so the major problem is thermal comfort of exposed areas like faces or hands. Solution: simply cover them with a thin layer.

  • 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 an 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.

    (Since I proposed this, something similar has apparently been done by Lyft as “Amp” LED light mounted in a car which displays a color synchronized with the rider’s app; not to be outdone, Uber has “Beacon”. Excellent!)

  • Moe anthropomorphism (TvTropes) mobile game but for fantasy sports. Kancolle-ize the NFL/NBA/MLB/soccer to exploit parasocial relationships and become a billionaire! Franchises like Strike Witches/Girls und Panzer/Hetalia: Axis Powers/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.)

  • fashion-as-service, NN smartphone app to tell you “no, that’s hideous!” (eg. “StreetStyle: Exploring world-wide clothing styles from millions of photos”+“A deep architecture for unified aesthetic prediction” etc). Someone suggested guide dogs for assisting their owners. Obviously that would not work—everyone knows that dogs are partially colorblind.

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

  • autonomous wedding photography: you receive a briefcase of a few dozen nanodrones; some follow bride 24/7, others roam evenly spread looking for action. You mail the drones back, video feeds reconstruct NeRF-equivalents, you pick every ‘photo’ your heart desires.

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

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

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

  • digitize motorized prayer wheels 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”.

  • a startup where the dress code is not hacker slacks or business suit casual, but zentai 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.

  • curation of polygenic scores & genetic correlations, with utility weights from medical-economic research/surveys. Embryo-selection-as-a-service.

  • 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 Chinese diplomacy & zoo revenues)

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

  • WiFi+blockchain+assassination market+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 Bloom filters 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”.

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

  • 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.

  • driver’s ed—in VR.

  • Zendo—in VR.

  • 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 eye tracking (necessary for VR headsets to do foveated rendering among other things), so this could be an useful tool for VR especially in calibrating pacing.

  • Soylentant: a restaurant serving tasting menus of all Soylent 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.

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

  • CutFit: a cutting-edge new fitness fad, exploiting the fact that repairing biological tissue after injury is energetically & metabolically expensive, requiring ~20kj/g5. 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, lipopolysaccharides, or 2,4-Dinitrophenol)

  • custom CNC manufacturing of extremely dense metal (lead/tungsten/osmium) 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.

  • VPN+speedtest service: Internet users use speedtest services like Speedtest.net to benchmark their ISP’s performance & see what they are actually getting (as ISP advertisements are so hedged & legally-unenforceable as to be lies); in response, ISPs appear to have often begin cheating by treating speedtest domains specially with top performance, while continuing to choke or traffic-shape regular connections. This renders the speedtest results less valuable.

    Thus, there is a natural complement or hedge here: if the speedtest services get preferential treatment, then users can simply piggyback on their special treatment using a VPN to the speedtest service. This preferential treatment makes the VPN much more useful than rival VPNs (which will presumably be choked like other regular uses).

    If ISPs respond to this by removing the preferential treatment and renders the VPN less valuable, then the speedtest becomes more valuable as its results become accurate again. Either way, it works. (And if ISPs try to have it both ways by analyzing traffic to choke speedtest-VPN traffic but not speedtest-actual-test traffic, they are engaged in a cat-and-mouse game requiring lot of extra work which would look bad if it ever came out publicly as it would be hard to justify.)

  • the Anti-Collectibles Museum: most museums can be summarized as simply containing now-expensive collectibles—preferably ones which have increased the most in price (if not value) over time. This means there is a logical opposite: a museum which contains formerly-expensive collectibles whose price (if not value) has decreased the most over time.

    The “anti-collectible museum” has several benefits:

    1. It is by definition cheap to acquire.

      This means that a large varied collection is feasible, and allows potentially amusing statistics like “the contents of the museum were once worth a collective $1 trillion” etc.

      Realistically, the major costs will be simply deciding what to acquire, as the most ‘valuable’ items will not typically enjoy regular appearances at auctions, glossy advertising, etc. No matter how dominant or expensive a collectible used to be, it will be difficult to even know its existence because there will be no one with vested interests in promoting or documenting it now, and the cheapest items will not trade for listed prices at all.

    2. Powered by Schadenfreude: if there is one thing people enjoy as much as seeing the risen, it is seeing the fallen. People will be interested to look at paintings which were once more famous than the Mona Lisa, etc.

    3. The exhibits will be inherently varied: things can decrease in price for reasons both good and bad.

      For every Beanie Baby or Death of Superman comics bubble whose price has crashed by orders of magnitude, there is an aluminum whose price crashed from literally “worth its weight in gold” to too cheap to pick up from the ground.

    4. Educational: helps undo the general bias towards only talking about currently expensive things.

      The idea that collectibles are a good investment may be spurred by news stories about paintings selling for half a billion dollars or some video game breaking the million-dollar mark at an auction—this is classic survivorship bias, where only successful collectibles & bubbles get news, but their stagnation or collapse is less remembered. (This appears equally true of attempts to financialize collectibles with instruments like hedge funds or “indexes”: they juice their returns by the usual gimmicks of dropping the losers & counting the winners.) Around every successful collectible like a Magic: The Gathering Black Lotus card lies a vast graveyard of cardboard & tinsel bric-a-brac.

      The anti-collectible museum documents & redirects attention to the graveyard.

    5. Self-funding: after a certain point, something which has become remarkably worthless can become valuable again due to that remarkableness (see interesting number paradox).

      For example, hyperinflation renders many currencies worth less than a penny, if treated as a currency, and ordinary transactions require wheelbarrows of bills—but those bills can then, outside the country or decades later, be worth entire dollars to collectors who would otherwise have no interest whatsoever in the currency of that country. The gift shop of an anti-collectible museum is the perfect place to sell such things.

  • Accessible D&D: accessibility is apparently a major concern of D&D in the 2020s, like “wheelchair-accessible dungeons”. A more interesting “accessible dungeon” would educate about accessibility instead.

    The accessible dungeon is highly welcoming & accessible when going downwards. For example, the dungeon would have wheelchair-accessible ramps… but only downwards, and with easily passed one-way doors or obstacles. All inscriptions or text would come with helpful Braille versions… which sometimes lie about critical details like colors. Spell scrolls would be written in multi-colored ink… and say different things to people with normal color vision vs colorblind people. Some traps or puzzles would have sound-based components, but with echoes and curved passages distorting the raw sound. Where there are not helpful ramps, progress might require leaping downwards a few feet over a gap.

    The main challenge of the descent is dealing with various cursed traps which, if not successfully circumvented or defeated, will inflict permanent wounds, like disabilities; party members may be forced to sacrifice eyes or ears and unable to heal them. (There would be no healing items, of course.)

    And once the party fights its way downwards to the grand prize, the dungeon is not over, because now they must return, with most of the members struggling with a handicap like a life drain, blindness, deafness, hobbled, or perhaps even muted. (To render mutism a real handicap for the players, a muted character’s player is no longer allowed to speak, either in game or in life, and must instead act out their desires like in Charades; except, because it would be too easy to guess standard actions if the other players could simply call out every reasonable choice and adventurers certainly would not have time to painstakingly figure out what their mute colleague meant, the non-mute players have 10s to silently watch the Charades and write down their best guess for the DM to execute.)

    The easier the descent along the primrose path of the accessible dungeon, the harder (and more educational) the anabasis.

  • Inflatable phone life vest: ‘waterproof’ smartphone cases do not save your phone when you drop it in deep or hard-to-see water; and ‘floating’ smartphone cases are extremely bulky, because they need enough volume to neutralize the smartphone density, so they are no good for accidents and are used mostly by snorkelers or scuba divers who want to take their phones with them.

    A similar problem has been solved with inflatable human life vests by using compressed gas CO2 cartridges: the cartridge is activated manually, or can be triggered by water eating through some water-soluble ‘pill’. These cartridges are small & cheap, and can last a decade before replacing them is a good idea.

    An inflatable smartphone life vest doesn’t need to hold up a drowning human nor be extremely reliable (it is just a smartphone, after all), and only needs to offset the smartphone just enough that it will eventually bob to the surface. So it’s possible that one could add a tiny CO2 gas cartridge, at relatively low pressure, into a smartphone’s case, at acceptable cost, reliability, and danger. (The longer/larger the cartridge, presumably the lower the pressure can be, and the less risky it is for it to leak or burst in one’s pocket; if a regular cartridge would be too bulky for comfort, it could potentially be flattened & wrapped around the smartphone. To reduce the danger of bursting even further, sacrificial vents/seams could be added at the corners where it would be mechanically weakest.)


  1. Cited in The Futurist November 2010↩︎

  2. “Childhood: Hearing Loss Grows Among Teenagers”, New York Times:

    The new study, published Wednesday in The Journal of the American Medical Association, analyzed data on about 1,771 youngsters aged 12–19 who participated in the National Health and Nutrition Examination Survey of 2005-6, and compared the prevalence of hearing loss with that of youngsters who took part in the survey in 1988–1994. The percentage with at least slight hearing loss increased by 30%, to 19.5% from 14.9% in the earlier study. For most the hearing loss is slight enough they may not even notice.

    ↩︎
  3. Why physical memorabilia, and not NFTs associated with the person, which were so hot in 2017/2021 and are so easy to buy/sell digitally?

    1. Speculation: NFTs, precisely because of the ease of global transactions, may be highly overpriced and any NFT wildly overpriced.

      • Cost: relatedly, because of low blockchain capacity & transaction fees, at the peak of 2021 Ethereum & NFT hype, it could cost more in gas/transaction fees alone to mint or auction a NFT than it would cost to just buy a physical object!

    2. Risk: in 30 years, Bitcoin and Ethereum will still be around—probably. Any given NFT platform… not so much. (Where are the original NFTs, ‘colored coins’ like Mastercoin, today?) Will platforms like BitClout (shades of Steemit) succeed or be superseded, perhaps successively, until reaching a final pinnacle? (We know that in technology startups, often it takes many tries before a good idea succeeds.) Physical objects will persist regardless; risks like fire or flood are well-understood and cheaply preventable.

    3. Option Value: in NFTs, an emerging strategy is tokenizing physical objects or even destroying them (‘proof of burn’) to create an associated NFT. But this is a one-way process: there is no interest in turning tokens into physical objects, regardless of burning.

    4. Availability: while easy to trade, ‘legitimate’ NFTs are in limited supply: people don’t understand them, regard them as a scam, object to cryptocurrencies on more general grounds like environmentalism, find them too technically challenging to do, don’t think they can make worthwhile NFTs, etc. For almost all streamers or social media celebrities, there are no associated NFTs. But on the other hand, everyone owns something which would make good memorabilia, generally accept & endorse the concept, and the mechanics are not arduous (“put the object in a box with an address, receive $200 to your bank account a few days later”).

    ↩︎
  4. 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 experimental felid hybrid breeds with exotic appearances (compare sales of the Russian Domesticated Red Fox), or in the case of Allerca & Felix Pets, advertising hypoallergenic cats (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.↩︎

  5. 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 (Mahan2000). 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) (Malina2004 [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.

    ↩︎

Similar Links

[Similar links by topic]