Hacker News new | past | comments | ask | show | jobs | submit login
Amazon’s $23M book about flies (2011) (michaeleisen.org)
168 points by ArneVogel 5 days ago | hide | past | favorite | 81 comments

This reminds me of a story going around at the chemical department here (probably it is false). Graduate student was told by prof to synthesize a substance. Process is well known, but it is tedious, he would have to spend the next two month in the lab. Since this guy was resourceful he called up privately one of their chemical suppliers and ask if he can order this substance from them. They send him a quote and he ordered it.

A few days later his prof comes in, and said: "can you synthesize a bit more, the company we sometimes take assignment from has a customer who also needs it"....

This is true, kind of. The original source is Peter Norvig's speech to Berkeley's graduating CS students in 2006.

> Before I answer that, let me return for the moment to Dewey Decimal 663: beverage technology. It turns out that in my very first professional job, as an intern thirty years ago, I had a colleague, a chemist, who had worked in beverage technology. The summer before he had been an intern and was given the task of figuring out if there was a particular chemical that gave whisky the distinctive taste of being aged in an oak cask. The company figured if they could isolate the chemical maybe they could just mix it in, and skip five or ten years of aging. So my friend went into the lab and isolated what he thought might be the right compound. He ordered a small vial from a chemical supply company (remember, this was before online shopping and email, so they had to actually write words on paper) and mixed up a batch. It tasted pretty good. (Why can't we computer scientists get research projects that involve consuming alcohol?) My friend was duly congratulated, and he wrote to the chemical supply company and asked for a 55 gallon drum of the stuff. They wrote back, saying "we regret that we can not fill your order because we are currently low on stock and, as I'm sure you know, to produce this chemical we need to age it in oak casks for five years."


Thanks! Story is slightly different, and since it was told to me in 1997 there might have been a different original source.

If true that's very funny, but also would either mean the student is very rich or their time is sold at an abysmal wage.

Almost certainly the latter, considering grad students are paid practically slave wages.

Off topic but it reminds of the book by Jack Stratton called "How I made $290,000 selling books" priced at $290,000.

Couldn’t find the original listing.

The author is a member of Vulfpeck, pretty fun band.


Vulpeck also made some noise with their album Sleepify, which was an album of silent tracks they used to game Spotify's pay for plays model. It made them ~$20,000 before Spotify took it down.



I love malicious compliance in the face of dictatorial companies.

All this ends up doing is decreasing the amount of money going to the artists that people are actually listening to.

That's a funny way of saying fraud. Do you also love companies that send boxes of sawdust or online store customers ?

It's not fraud unless they said that the tracks are not silence.

If I buy sawdust from someone and they send me sawdust, and it's structured so that Amazon pays them some money for it, that's gaming the system, not fraud. Amazon can stop allowing them to do it (just like spotify did).

> If I buy sawdust from someone and they send me sawdust

The validity of your point depends on whether you can demonstrate that people do pay spotify to listen to silence.

My understanding is that people would intentionally put these tracks in their playlist to get periods of silence scheduled ahead of time. If that's not the case, then my original comment was wrong.

> It's not fraud unless they said that the tracks are not silence.

It is fraud, and although IANAL you'd lose in court.

I'd lean copyright infringement? Probably has no audible difference from a John Cage song

Just so you know, copyright infringement does allow independent development to be a valid excuse.

Copyright infringement requires intentional copying without permission, and so if you independently produce the same thing without copying, that thing now has two distinct copyrights.

Patents may be what you're thinking of, as they require zero knowledge nor intent to infringe upon, but patents do not apply to artistic performances.

Then get a computer, get a piano-playing-software, hit 2-3 random notes VERY softly in the beginning & middle & end of each song, and call it a day. No silence. No infringment.

Maybe go learn Esperanto so you can bring joy to the lives of others.

What precedent can you cite to justify your claim?

I think this is more in the vein of Cards Against Humanity shenanigans.

It's a lot more like the fraud attempted in Office Space (... and Superman 3). Spotify artist payments are not based on play counts directly. They are actually based on Spotify's total revenue and the proportion of global Spotify streams that the artist received.

As a result of these extra plays, Vulpeck's stream rate increases, but Spotify gets virtually no extra revenue (especially since a user would have to be a premium subscriber to get a fully quiet experience listening to the album). Vulpeck gets paid a bit more, and everyone else gets paid a minuscule amount less.

For reference, here's one article that mentions this model of how Spotify pays its artists: https://qz.com/1660465/the-way-spotify-and-apple-music-pays-...

Did they not also sell their 10th track as an ad on some album? (not the first with that idea though)

A pre-internet joke on the same topic. An ad in the classifieds advertised "How to make money with classified ads. Send $10 to xxxxxx". Sending $10 got you a letter that said: "Place a classified ad with the text `How to make money with classified ads. Send $10 to xxxxx`"

If you think that's great, you'd love this website:

[How Many People Paid 1 Dollar To See How Many People Paid 1 Dollar](http://www.howmanypeoplepaid1dollartoseehowmanypeoplepaid1do...)

There was the $1k "I Am Rich" iPhone app that just put up an image of a glowing gem. The price was set by the author as a work of art so it's not directly related to the amazon book prices. 8 people bought it according to the Wikipedia page


While working on my side project [0], I regularly find some pretty outlandish prices on Amazon. This post makes me want to add further analysis on 3rd party prices and how they relate to one another.

[0] https://unitprice.org

I once saw a regular cd from Yngwie Malmsteen for several thousand dollars. It wasn't even rare. Left me wondering what the purpose of that was.

To drive up the "average price" shown to customers by the algorithm, of course.

Awesome side project. Amazon.ca next!

Thank you! One day I hope to expand beyond the US Amazon!

I noticed that after I bought some old less known books for < $50, few weeks later they would cost over $1000. I think to myself that now for the amount these books are worth I could buy myself a condo in the Caribbean. Joking of course. This type of bot pricing is very irritating. Sometimes when I want to buy a book for research I need to wait a lot for it to have a normal price if someone bought it before me.

If you're waiting for a book to reach a reasonable price, the site camelcamelcamel.com lets you set a price target for a book and informs you when the Amazon price falls to that target. It also provides historical price data so you can see if the current price is good or bad. (I have no connection to the site except I find it useful.)


I've used them for so many higher end items and got great prices (watches, tools, cameras, etc). You just need a bit of patience is all.

I know ebay now let’s you “pause” listings; but they didn’t used to. If you discontinued your listing, it was eventually lost forever; so the best way to pause your listing was to just set a ridiculous price.

I have a listing like this for some tiny screws that I seem to have lost, but might one day find.

Both Amazon and eBay have vacation modes that have been around for as long as I can remember.

Yeah but vacation still presents the listing and allows purchase.

"Vacation Settings" for Amazon sellers doesn't show the listing to buyers, except when it's Fulfilled by Amazon.


eBay seems to have a similar option with "Pause item sales":


This happens all the time on Amazon. Just search for a product, like shirt, and then change the sorting from "relevance" to "high to low". For this particular search at this time there is a nice women's $2,000 Nike tee shirt.

In general, sorting by anything but "relevance" is just a way to find bad data, like sorting TVs by size and finding a TV which reinterpreted its size from mm to inches or vice versa and now seems to be an absurdly large or small TV, on paper. It's not even spamminess in this case, just bad data.

Makes one appreciate the effort that's put into the "relevance" scoring. It's probably >100 engineers working in Amazon on just this core ranking problem.

Tuning relevance is extremely difficult, and it’s not just an information retrieval problem - at Amazon’s size (and even smaller companies) a lot of what you see for “relevance” is hand-tuned by humans, through machine learning on longer-tail queries, and some combination of both in the middle (with sponsored or high margin or velocity items weighting more heavily). If you search for something like “men’s shoes” there’s 0% chance you’re getting a search result, it’s going to be a curated result page. Lots of people work on this hand tuning, but they’re not engineers, they’re product and marketing people.

If you search for “men’s shoes with green laces” you’ll see the real challenge, with a top 10 search result that’s not even shoes (it’s a pack of laces).

Some of my favorite test queries for shopping sites (from back when I used to work on a comparison shopping engine:

[apples to apples] -> frequently returns random "Apple" products, because when you dedupe, stem and remove stopwords from that query, you're left with "apple", and Apple products are big ticket items.

[shirt dress] -> frequently returns dress shirts, because many sites essentially ignore word order or the notion of compound words.

[fish swim vest] -> frequently returns fishing vests, partially because of overstemming (many POS stemmers will stem "fishing" -> "fish", but in a product search setting, it's usually the wrong thing to do).

Here's one that's fun if your objects of interest include cities:


What's the [best city in europe]? Obviously "Best, Netherlands".

The other day, I found a household appliance (I don't remember what kind), that boasted 16 gigs of RAM.

Probably a refrigerator. A washing machine only takes 8 gig.

Refrigerator Repairing And Servicing, Memory Size: 2GB, 4GB, 8GB, 16GB, 32GB


Haha! That's crazy! Though, it does point out that many listings may have been created by copy&paste, leaving them with incorrect information.

If your discard outliers you get usable data.

That helps, but you also have a lot of bad data at the extremes that are still within the bounds of plausibility.

For instance, nine out of ten 122" TVs might actually be 12.2" TVs that got misinterpreted or typo'd to 122", because small TVs are so much more common than large TVs that there can be more malformed small TVs than well formed large TVs, but you can't necessarily discard all of the 122" TVs as outliers, because there are actually valid TVs in that range as well.

It makes sense. There is no much cost for such a mistake. Some team somewhere has a "correct overprice bug" low in the priority of their backlog of tasks.

Have you found many books that you wanted priced at $0?

>Have you found many books that you wanted priced at $0?


> gutenberg.org

You are right. :)

My point is that a bug that set book prices in Amazon to zero would be fixed quite fast.

> Why though would bordeebook want to make sure theirs is always more expensive? This would seem to guarantee they would get no sales. But maybe this isn’t right – they have a huge volume of positive feedback – far more than most others. And some buyers might choose to pay a few extra dollars for the level of confidence in the transaction.

Alternatively bordeebook may be focused on when the competing book sells out (these are used book listings on Amazon which are single units). Then they will be the only alternative, and at a 27% markup. Books also often get bought in bursts (group sales, class assignment, labs, book clubs), so this might be reasonable purchase pattern prediction.

Yes, this is how I used to price my books. There are two reasons to price higher than the lowest. One is to make a little more profit because your book will sell as soon as the cheapest book sells. The other is to avoid my pricing bot fighting with someone else's pricing bot down to $0.01. I would rather accidentally shoot the price really high than give the book away for a loss. Especially considering a lot of the books were $150+ used textbooks.

The steam marketplace also has algorithmic traders (despite being against TOS to automate marketplace interactions). I found myself in a position where there was only one other listing for a limited time discontinued item that I had bought a few of. I bought the other one so I could set the market price (gouging sure but there wasn't any demand so still risky). Another one was listed within 15 minutes. I bought that one too, steam only reveals to you who the seller is after a purchase (and it was the same seller).

I wasn't convinced it was a bot and not just someone paying close attention to their steam account until I realized they would under-list me every 5 minutes or so. Since there was no activity on that item, and no buy orders, I experimented by listing my item lower and lower, and they kept undercutting me, down to a floor. Their floor ended up being pretty low (around an 85% loss on what they originally bought at), and after I bought at the floor they then re-listed another for double that, before slowly declining again to under-cut me. For an item with no demand, I decided not to try to slowly buy out their entire inventory at the algo's floor.

I found other items that bot was also trading (after I bought them) and discovered that they also change their listing prices based on where people have buy orders set (a good indicator of some demand / market value), so their floor is higher on an item with any demand at all.

So could you drive down the price of something you want to buy by offering the same item (that you don’t actually have) at a low price?

On steam you have to actually HAVE the item in order to list it, but if you buy one (and wait for a 7-day re-listing period), you can start an undercutting war with some bots yeah if the item has low or no demand.

I talked once to a dev who does this sort of thing. Its common. It's also interesting to consider what options folks have for quick, easy short-circuit algorithms. Any ideas for clever hacks to prevent the totally absurd with minimal state retention needed?

When I started talking about moving averages and trend related metrics, the person I was talking to commented that they where doing this in Excel or an Access database and wouldn't know how to handle that with say 250,000 books they might know or care about... then again maybe the downside of having one random crazy price is not even worth a cheap hack.

I've worked in this space before (though ~5 years ago), and I would say that most of the solutions are so unsophisticated that any measures to prevent this, like the ones you propose, would likely be completely overkill.

AFAIK, there have also been some safeguards added into the MWS API itself, which prevent too drastic price changes to be uploaded, which prevent those bugs from surfacing nowadays.

There's a whole ecosystem of SaaS doing this fairly priced for smaller sellers, along with unfairly priced SaaS for large sellers (some take a percentage or more of revenue).

Anyway for repricing just have a hard max of 1k or so, and lower prices if it doesn't sell after six months and again after a year, two years, etc.

What makes the SaaS pricing manager unfairly priced?

Having used one briefly (Feedvisor) several years ago, they attach a very heavy handed enterprise sales operation to a mediocre platform. A single digit percentage of sales is insane.

I got calls every few months for years afterwards trying to get me to try them again. They raised $30M+ and built out a strong sales team.

That sounds like it was priced higher than you cared to pay.

I don’t find Ferraris unfairly priced; I just don’t buy them because they are too high priced for my preferences.

If you define fairness so that literally everything is fairly priced, sure. My definition is different.

Some outlandish prices are for money laundering etc. Not this one maybe. Seller is known cohort of buyer. For the Amazon sales fee, they get to transfer funds and get a great item at the same time.

I wonder could you exploit this if you're the company who has the book.

Buy it as a normal user from the other company for $23 million. They buy your copy for $22 million. Wait until the end of the returns window (assuming Amazon enforce one) and return it for $23 million. They try to return it to you, but it's too late.

Probably the bookshop notices something is wrong on the $23 million item, but maybe you could make a few thousand this way.

It is a more general problem on Amazon, and in other internet resellers. Products are proposed at insane price, and stay there during month or years, waiting for a dumb buyer. This is annoying, because when you browse Amazon looking for something interesting, you are polluted with all theses dumb offers.

What should happen is that another supplier steps in to disrupt the equilibrium towards the 'correct' price. Which is the case today -- instead of two sellers there are 15.

What's the correct price?

Gonna go out on a limb and say somewhere closer to the marginal cost of production and less close to 23 mil.

Theres a lot of predatory pricing now with these algos (not in the legal/official sense of the phrase which is to undercut competitors). For instance I noticed that I was being charged 6x compared to others for something that I frequently bought on an app.

I have experienced this myself on a smaller scale. I was looking for an old book but all the sellers wanted $80 or more. One day the prices started to fall with two suppliers, always a few cents every few hours. In the end I got the book for about $10.

This is fairly normal, Edition 5 is going out of stock when Edition 6 is released. Some famous youtuber puts a link to Edition 5 and drives demand/traffic on the item, and boom $50 manual becomes $5,000. Edition 6 is still $50.

I have noticed such pricing tactics with Udemy. I typically buy courses for $10-12 range, but they frequently show me courses around $94-$136 range. (especially when I am logged in my account).

I had hoped this would be about a book about flies that was genuinely worth $23 million.

How often may the price change? That might make a nice covert channel.

I bought a copy of this. Would be willing to sell it for $19M.

I assure you that book is only worth about $21,000,000

Next time someone asks how markets work (art, housing, stocks etc) ill point them here. The most mind boggling is that no one is to blame. Our system has gone nuts. We can hope someone will lower these prices out of greed but that is optimistically assuming they even have the books. It seems more like each system has a finite life span that we barely understand until we see it expire.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact