Bram Cohen

Here's an interesting search query challenge for relevancy ranking: xkcd metallica. Currently google doesn't, but yahoo does, give the right answer.

Let this be a lesson to you. If you want your web site to show up in search queries, and have words encoded in audio, video, or images, include a transcript!

The technique for building tension on American Idol is an interesting one. Rather than tell you who came out on top each week, they instead tell you who came out on the bottom, giving the impression that it's an active competition which anyone can win. The truth is quite a bit more boring. If you look at the previous season eliminations, in four of the six seasons the eventual winner was never in the bottom group, and the earliest the eventual winner was in the bottom group was in the sixth show. So the show could instead knock out a third of the top 12 on the very first show, and probably not have changed the eventual result in any of the contests thus far. Chances are that in most seasons the eventual winner starts getting the most votes early and stays that way every episode.

Does this mean that American Idol would do a better job of picking the winner if it had stayed with the format of the first few seasons, where votes were used to advance competitors through to the finals? Probably, but there's a distinct lack of any criterion with which to measure 'better'. Yes, votes are more representative of later votes than the judge's judgment, and if the pool of people is bigger there's a greatly reduced chance that there won't be any decent competitors in it (as happened in season six, yeesh), but the only measure available for determining how accurate voting was is later album sales, and those have some very wacky things going on.

Looking at the American Idol top sellers one thing stands out: in most of the seasons, exactly one finalist sold over a million albums, with a big gap below that. The exceptions are season 2, which had two, and season 6, which had none (see aforementioned observation about complete lack of talent in season 6). Apparently the show's producers pick exactly one singer to put some real marketing muscle behind each season, and let the others sell however many they happen to sell. The other thing which stands out is that with the exception of Kelly Clarkson, the first season winner, not a single top seller has sold more than half as many copies of their second album as their first, strongly implying that almost all of their sales are due to Idol-related publicity, with complete failure to build a fanbase on their singing merits alone.

Clarkson was the huge exception to that rule. She sold more than twice as many copies of her second album as her first. My guess is that had American Idol not lucked out with Clarkson in the first season, it would be viewed as having no legitimacy as a source of new talent whatsoever.

The web browser is the new operating system. My own computer is at this point little more than a glorified web browser, with a text editor, command prompt, python interpreter and svn thrown in for the occasional color.

Since the web browser is the new OS, it should really, really, have the equivalent of a process list. I almost always have a whole bunch of tabs open, and firefox is most of the time using a nontrivial amount of CPU doing not much of anything. I have to guess which tab is causing the problem when the CPU gets pegged, and sometimes it seems that even shutting down all tabs doesn't completely fix the problem. Could somebody please implement metrics on how much CPU each tab/window is using, and get the process separation right so that whenever a tab/window is shut down all remnants of it are completely toast?

I had an idea the other day. A see-saw has two seats, each of which is offset from the other one by 180 degrees. A more complicated linkage could create a three-saw which had three seats each of which was offset from the next by 120 degrees.

Why? Because.

A horse recently died after finishing second in the Kentucky Derby. This kind of thing happens a lot. How is it that horses are so fragile that they die just from stumbling while running? They're overbred for racing, making them fast but not really very healthy.

I have a simple solution to the horse overbreeding problem: Have races consisting of all genetically identical horses.

Don't laugh. It's gonna happen.

Credit scores in the United States are used as the end all and be all of one's credit score, nonsensically used for approving credit cards, mortgages, and rentals, as if those all had the same risk profile. Moody's ratings are being criticized now for being taken as gospel, as if 'AAA rated' was the only useful thing which could be said about an investment. And my cursory research indicates that the same thing is starting to happen with Zillow estimates. It appears to be that as soon as any official rating appears, americans will cease to apply any judgement of their own and simply accept the official number as gospel.

It occurred to me today what's going on here. It's the religion of the Grade Point Average. Americans are so completely indoctrinated from a young age that a single official number is the sum total of all that can be said of a person's moral worth that they're psychologically incapable of evaluating people or investments in any other way, even when it's their job to do so. Hooray for US education, teaching those important lessons about life.

An article on chess variants with fewer draws. The most obvious modification of eliminating stalemate is proposed by several people, including me. Beyond that, by far the best suggestion is to allow a king to capture the opposing king if it's a knight's move away. Someone should do an analysis of some basic endgames with that rule, for example knight versus bishop.

Fixing the problems in Single Transferable Vote is a subject of much research. Meek's Method does good job of fixing the most obvious problem, and I believe the approach is basically sound, but it still has some bad artifacts. Specifically, Meek's is used for electing only a single candidate, it will behave like instant runoff instead of picking the condorcet winner, and in an election with many candidates to get elected a candidate can appear second on almost every ballot and still be eliminated first.

Obviously those are very busted behavior. I have a simple, and I believe new, suggestion as to how to fix them.

The only part of the algorithm which is changed is the one for picking which candidate to put in the 'excluded' category. Meek's picks the candidate with the lowest weight, which is a simple but highly flawed approach. Here is an alternative algorithm: For every pair of candidates in the 'hopeful' category, calculate which of them would get more weight if all candidates except those two and the ones already elected were crossed off the ballot. Then move the condorcet loser based on the results of those two-way races into the 'eliminated' category.

This technique straightforwardly gets rid of the two artifacts I mentioned and also fixes the strategic voting which real political parties do in practice, an interesting technique which probably warrants its own separate post for explanation. I believe it is in all ways a clear improvement and should be adopted by anyone actually using Meek's (or STV for that matter) in practice.

The remaining artifacts which this technique leaves are quite deep and generally both unimportant and intractable. They are:

If a voter disagrees with the choices of the other voters who have the same first-place candidate, they can move their first-place candidate farther down the ballot and thus make their own vote count for more. This is a fundamental issue in proportional representation voting, and not a big problem in practice.

When there's no single condorcet loser, all the usual issues apply. These are fundamental issues for any voting system, using any algorithm and voting method, and the usual techniques have their usual pluses and minuses, and the choice of which one to use doesn't matter all that much because such situations rarely occur in practice, and when they do simply rolling a die to determine which member of the smith set to pick will work about as well as anything else.

There are some strange edge cases where the monotonic approach of my proposed algorithm is fundamentally limiting. In practice these are unlikely to occur all that much, and when they do the monotonic approach will do something suboptimal but entirely reasonable. CPO-STV basically fixes this but at a cost of much greater cognitive and computational complexity. I for one have a hard time intuiting what CPO-STV will do under some circumstances.

Those are all the artifacts which are left. Other than that, it just plain works.

I'd like to thank Philip Neustrom for reminding me of this subject in general and Meek's Method in particular.

There are a bunch of wiki features which are nonstandard, or not available at all, which I would like.

Pages should have an 'annotate' view, where you can see who wrote each line of a page.

When modifying a page, if the page gets modified by someone else between the time you start editing and commit, it should attempt to do a three-way merge of your changes before rejecting them outright. This would make a big difference for large pages which many people modify.

It should be possible to see all recent edits by a particular user, and to search for a keyword in edits done by a particular user.

On Sunday there was a radio interview with me on NPR. Probably the first time the church-turing thesis has ever been mentioned on NPR, and surprisingly the quote didn't even get mangled in the editing.

Since I posted a bit about driving, here's some good suggestions of how to make traffic run more smoothly which they don't teach you in driving school.

Now is the time in the US primary cycle when the US media starts to lie. It happens every time - just when it's becoming clear who's going to get the nomination, the media completely manufactures the story of a competitive race, because it sells newspapers.

The truth is, it's basically over. The chances of Clinton winning the Democratic nomination have become remote. Now that that's the case, I feel it's time to talk about how completely surreal her campaign has been.

The Clinton campaign rested on three core talking points. First, that she was the inevitable candidate, which now looks so ridiculous as to not be worth debunking (and should have at the beginning, to).

The second talking point was that Clinton has the most experience. This is an example of 'the big lie' - if you make a claim which has no basis in reality whatsoever, it's more likely than a claim which closely resembles reality but is slightly wrong. The Clinton senate biography page doesn't even mention any past history before being in the senate, and as first lady her main real involvement was in the Clinton health care plan, which was a disaster. The way she's run her campaign isn't a so great either. Basically, there's no record of experience, not even a made up one. Specific debunking aside, everybody knows the only reason Hillary has any national presence is because she's married to Bill, and he used to be the president. Any claims to the contrary should rightfully be greeting with raised eyebrows.

Clinton's third talking point was that she was the most 'electable' candidate. While it's true that she's widely liked, she's also widely disliked, with a large section of the population already being familiar with and hating her. She also has a tendency to do flagrant political maneuverings with no apparent awareness that people might see through them. For example, all the candidates agreed to basically cancel the Michigan primaries, including her, and everybody else removed their names from the ballot, except her, and then after the primaries were held she argued the delegates should be reinstated, presumably because she'd given the people the right to vote for her and now that should be respected. She also outright lied about doing 'dangerous' diplomatic work. Add to those and similar incidents questions as to where exactly all the of the Clintons's money has come from since Bill left the white house, and it's very clear that Hillary isn't exactly a compelling candidate. The polls (which, granted, have to be taken with a grain of salt) have backed that up as well - early polls indicated she would win the general election against leading republicans by the most narrow margin of any leading Democratic candidate, and polls now indicate the same thing.

Politics is politics I understand, but for the central themes of a presidential campaign to be claims which are obviously untrue to the general population is just plain bizarre.

I've been thinking about humidity in food preparation recently. It seems like, after temperature, humidity levels are the most important thing in cooking, with 'soggy' and 'dry' being failure modes just below 'raw' and 'burnt' in their prominence.

Some systemic humidity problems seem like they should have technical solutions. The first is freezer burn. Cold air holds onto less humidity than warm air, so every time you open your freezer some warm air holding on to more humidity than the air previously in the freezer is let in. That air is then cooled off, forcing it to let go of moisture, which then crystallizes on your food, and voila, freezer burn. When you wrap food in plastic you're limiting the amount of warm air its ever exposed to, thus hopefully reducing the amount of freezer burn to the amount of moisture that was trapped in the plastic initially. Wrapping in plastic works fairly well, but it would be nice to have a freezer which regulated moisture levels, to eliminate the need for plastic and so that foods which themselves let go of or took on moisture at different temperature levels wouldn't get damaged. I'm not sure what sort of mechanism would be good for regulating moisture in a freezer though, especially with the requirement that the thermal efficiency should be damaged as little as possible.

The second big moisture issue is in cooking. When you bake food in an oven, the air gets warmed up, causing it to be able to absorb more moisture, thus drying out your food. Adding more moisture to the air in the oven can fix that problem. It should be possible to make an oven which does this automatically using a mechanism similar to the one which puts oil in your car engine, but a simpler approach is to calculate how much water will be needed, preheat the oven to the desired cooking temperature, then add the right amount of water in a pot, and when the pot starts to boil add the food. If anyone cares to point out how to calculate the appropriate amount of water for a given size oven and cooking temperature I'd appreciate it (yeah, I know, this is basic high school physics, but it's easier to ask the lazyweb). I'd also appreciate it if anybody can point out a fish recipe on the web of the form 'season to taste, then cook at X degrees for Y minutes'. Online recipes seem to really not like being simple. Fish seems like the most obvious dish to try, because it's notorious for drying out easily. Suggestions of particularly fickle fish would be appreciated as well.

Update: I was basically wrong about freezer burn. The phenomenon I described is the cause of frost build-up in old fashioned freezers, but modern freezers which auto-defrost do so by drying out the air, and the dryness is what causes freezer burn. See comments for some discussion and links.

With some more practice driving (I recently got my driver's license) I've noticed a few more things about driving.

The main thing to do while driving is to pay attention to what's ahead of you. Not just immediately ahead, but the whole section you plan to go over in the next few seconds. This is because cars generally move forwards. Sometimes they move forwards very quickly.

What's odd thing is that the driver's test does hardly any checking for how well you're paying attention to what's ahead of you, and does tons of testing for how well you look away. Partially this is just because it's a lot easier to grade, but there are some frequent situations, such as turns and lane changes, which necessitate checking in a direction other than forwards. The effect winds up being so exaggerated that it's actually an advantage to take the driving test wearing sunglasses which cut off your peripheral vision, because that forces you to make the exaggerated head looks the instructor is looking for.

This produces an odd distraction from the core focus of driving, looking forwards, onto something quite secondary, which is looking everywhere else. A typical maneuver for a new driver is to look for a lane change, see that there's a car in that lane but just barely far back enough that it's okay to move over, then signal and start to move. Since the place where there isn't room is behind you, the natural tendency is to spend the merge time looking in the mirror. Humans naturally and correctly slow down when not looking forwards, which tends to accidentally make the person behind you catch up and crowd them off the road, resulting in a near accident and a bunch of honking. I've taken to glancing to make sure there's enough room, then looking forwards while merging, speeding up a bit to give some more room, and trusting the person behind me not to slam into me. I think almost everybody does basically the same thing.

Even worse is checking the blind spot on the left side (I'm in the US, those of you in backwards land can flip the handedness). To check the left side blind spot, I have to turn my head almost 180 degrees, try to make out what I see in the sliver of space between the B and C beams, and turn my head back and try to refocus on the road in the split second before starting to merge left. Whether this maneuver is a net positive for road safety is highly unclear, since it takes the better part of a second and one's eyes are on stuff not even vaguely useful for most of it. I've taken to either passing someone, checking that they're far back enough and then merging or checking that there's a gap behind the first car back in the other lane, letting them pass me, and then merging, basically using another car to sweep out my blind spot. Driving instructors don't tell you to do that maneuver, but it feels under control, while looking 'properly' feels hella dangerous.

All this leads to a very obvious question: Why is there no secondary mirror? You know, the small rounded kind that buses have. Sure, most people don't know how to use them, but after driving with one for a while you'd learn to recognize what you see in it after a few months without even making a conscious effort. I sure hope the answer isn't that it would deprive driving test administrators of a checklist of things to grade people on.

Dyson vacuums work on the principle of cyclonic separation, which has long been used to get sawdust out of the air, and was only with dysons applied to vacuums.

The limitation of cyclonic separation is that it can't get rid of particles less than some size from the air. A technique which could remove all particles would be to bubble the water sucked in by a vacuum cleaner through some water. That would cause all the particulates (and some noxious gases) to get dissolved in the water, which could easily be removed when you were done cleaning.

My suspicion is that this technique would require a lot of power to force the air under water, but that it would be quite effective. It might also work well as a standing system in a house, by having an air purifier which is constantly sucking up air from around a house and bubbling it through a central water tank, and periodically flushing the tank.

I wondered the other day if there are any interesting puzzle ring designs which involve small bands confined to the woven section, so I came up with the following:

            |  |
            |  |
          .-|----.
         /  |  |  \
        /   |  |   \
       /  .----|-.  \
      /  /  |  |  \  \
      \ /   |  |   \ /
       /    |  |    \
      / \   |  |   / \
      \  '-----|--'  /
       \    |  |    /
        \   |  |   /
         '--|-----'
            |  | 
            |  |

For this design to work well the small bands should actually be heavier than the large ones, so the puzzle gets pulled apart by centrifugal effect rather than held together by it.

Restaurants which generally sell out have an interesting dilemma. In principle they could make more money with higher prices, but then they'd risk not selling out, and empty seats would quickly wipe out any revenue gains from raised prices, not to mention harming that elusive 'buzz'. In practice such restaurants generally wind up leaving some money on the table, no pun intended, and take the stability of always selling out over the potential of higher revenues.

I've come up with a variant on dutch auctions which solves this problem quite beautifully. The restaurant continues to charge the same amounts it does currently, with the same menu, but there's a 'seating fee' for sitting down which might be charged if the restaurant sells out in advance. The amount of the seating fee is determined by when the restaurant becomes completely booked, with the fee going down the later the selling out happens, possibly going down to zero at the end. By making a reservation when the potential seating fee is a certain amount, a customer is declaring that they're willing to pay the seating fee for that time period if it is necessary, but they aren't penalized for making an early reservation unless it would have been necessary to do an early reservation to get a seat. By waiting to make a reservation until later, a customer is declaring that they are unwilling to pay a higher price, but also allowing for the possibility that the restaurant will become fully booked and they won't get a seat. One of the nice features of this system is that the reservation system is essentially unchanged, allowing for trivial support of reserving particular time slots and tables.

This system also works for concerts and other events which have the potential to sell out.

A decent dot com business model would be to make a web site which performs this service for restaurants, and does fulfillment of seating fees and keeps a fraction of the seating fees for itself as a way of getting paid by restaurants. Restaurants would probably be quite agreeable to that, since seating fees are essentially found money for them, and they'd be happy to let someone else have a fraction of it.

The coverage of the current US primaries is mindbogglingly wrongheaded. Recent coverage has focused on who would 'win' New Hampshire among the democrats, and Huckabee's 'lead' among republicans. The actual numbers can be found here. New Hampshire is not a winner-take-all state for democrats, and both Clinton and Obama got exactly nine delegates from there, making the declaration of a 'winner' extremely misleading, if not outright revealing of the declarer having dubious mental capacity. Among republicans, Mitt Romney now has the most delegates, with Huckabee in second, and the media is currently speculating that Romney will drop out because he's so far 'behind'.

Seriously, what is wrong with journalists? Are they not able to do basic arithmetic? Ideally I'd like to have meta-coverage discussing why some states are winner take all and others aren't, and what on earth 'super-delegates' are, but I'd settle for even an accurate portrayal of what's happening in the race as it unfolds.

But thank you CNN for putting up a nice site which gives accurate up-to-date information. Please expand it in the future with more explanation of what 'super delegates' are, and what happens to a candidate's delegates if they drop out of the race.

Times up on my puzzle ring pattern challenge. Nobody gets the cookie.

To refresh, the question was what the pattern/generalization of this design is:

___   ___   ___   ___   ___   ___   ______
   \ /   \ /   \ /   \ /   \ /   \ /
    \     \     \     /     /     /
___/ \   / \   / \   / \   / \   / \   ___
      \ /   \ /   \ /   \ /   \ /   \ /
       \     \     \     /     /     /
___   / \   / \   / \   / \   / \   / \___
   \ /   \ /   \ /   \ /   \ /   \ /
    /     /     /     \     \     \
___/ \   / \   / \   / \   / \   / \   ___
      \ /   \ /   \ /   \ /   \ /   \ /
       /     /     /     \     \     \
___   / \   / \   / \   / \   / \   / \___
   \ /   \ /   \ /   \ /   \ /   \ /
    /     /     /     \     \     \
___/ \___/ \___/ \___/ \___/ \___/ \______

And the answer is that if you chop the pattern up into four quadrants, and the crossings all go the same way within each quadrant, like so:

___   ___   ___   ___|  ___   ___   ______
   \ /   \ /   \ /   \ /   \ /   \ /
    \     \     \    |/     /     /
___/ \   / \   / \   / \   / \   / \   ___
      \ /   \ /   \ /|  \ /   \ /   \ /
       \     \     \ |   /     /     /
___   / \   / \   / \|  / \   / \   / \___
---\-/---\-/---\-/---\-/---\-/---\-/------
    /     /     /    |\     \     \
___/ \   / \   / \   / \   / \   / \   ___
      \ /   \ /   \ /|  \ /   \ /   \ /
       /     /     / |   \     \     \
___   / \   / \   / \|  / \   / \   / \___
   \ /   \ /   \ /   \ /   \ /   \ /
    /     /     /    |\     \     \
___/ \___/ \___/ \___/ \___/ \___/ \______

This can be scaled up or down to any number of bands, for example the classic design is four bands

___   ___   ___   ___   ______
   \ /   \ /   \ /   \ /
    \     \     /     /
___/ \   / \   / \   / \   ___
      \ /   \ /   \ /   \ /
       /     /     \     \
___   / \   / \   / \   / \___
   \ /   \ /   \ /   \ /
    /     /     \     \
___/ \___/ \___/ \___/ \______

And it can be chopped into quadrants as well

___   ___   ___|  ___   ______
   \ /   \ /   \ /   \ /
    \     \    |/     /
___/ \   / \   / \   / \   ___
------\-/---\-/+--\-/---\-/---
       /     / |   \     \
___   / \   / \|  / \   / \___
   \ /   \ /   \ /   \ /
    /     /    |\     \
___/ \___/ \___/ \___/ \______

I also gave the three banded version in a previous post. The five-banded design works as well, but I don't like that as much because it doesn't have a strong canonical position which the bands want to go into like they do with four or six.

I'm very happy with my new CEO, although the executive search firm we used did some things which I personally find quite humiliating, so I'd like to let people know a few things.

Just because you were approached about being BitTorrent's CEO doesn't necessarily mean that I'd ever heard of you. If I had ever heard of you, it doesn't necessarily mean that I thought you had the necessary experience for being BitTorrent's CEO. Even if I did think you had the necessary experience, it doesn't mean I wouldn't have gotten fuming mad at your name being suggested for any of a number of other reasons, including in some cases widely known lack of competence and lack of morals.

Normally I'd keep quiet about this for the time being, but at least one person is talking about how he was contacted by the executive search firm, and I happen to have completely blown my stack after I heard that he'd been approached, so I wanted to set the record straight.

The executive search firm's name? Heidrick & Struggles.

Unrelatedly, if there was any candidate who literally stood me up five times and then removed themselves from the search as if the job was theirs for the taking, I'd like to let them know that I'd already knocked them off my internal list of potential candidates after stand-up #3, because that was a strong enough hint to make me not care how godlike some people view them. Really, the nerve of some people.

My last 4-banded puzzle ring design gave me an idea, so I came up with this:

___   ___   ___   ___   ___   ___   ______
   \ /   \ /   \ /   \ /   \ /   \ /
    \     \     \     /     /     /
___/ \   / \   / \   / \   / \   / \   ___
      \ /   \ /   \ /   \ /   \ /   \ /
       \     \     \     /     /     /
___   / \   / \   / \   / \   / \   / \___
   \ /   \ /   \ /   \ /   \ /   \ /
    /     /     /     \     \     \
___/ \   / \   / \   / \   / \   / \   ___
      \ /   \ /   \ /   \ /   \ /   \ /
       /     /     /     \     \     \
___   / \   / \   / \   / \   / \   / \___
   \ /   \ /   \ /   \ /   \ /   \ /
    /     /     /     \     \     \
___/ \___/ \___/ \___/ \___/ \___/ \______

Aside from being very aesthetically pleasing and conceptually straightforward, this design has the remarkable property that if you remove any one of the three pairs of opposite bands the remaining two pairs will form the classic puzzle ring design.

The way I derived this one was actually quite laborious and involved reasoning out the relationships between the bands. I only saw the clear pattern when the design was finished. First person to point out the obvious generalization gets a cookie.

Here's a 4-banded puzzle ring design which I missed previously, because I was systematically going over all the ways which the rings can relate to each other and skipped the one which corresponds to the traditional design. This one is, unsurprisingly, similar in movement and asymmetry than the traditional design, but is probably slightly more difficult.

_______________   ___   ___   ___
               \ /   \ /   \ /
                \     \     /
______   ___   / \   / \   / \___
      \ /   \ /   \ /   \ /
       \     /     /     \
___   / \   / \   / \___/ \______
   \ /   \ /   \ /
    /     /     \
___/ \___/ \___/ \_______________

This is a variant of the above. In this one the two outer bands are linked to each other, which might make it a bit easier, but it's very pretty.

_________   ___   ___   ___   ___
         \ /   \ /   \ /   \ /
          \     /     \     /
______   / \   / \   / \   / \___
      \ /   \ /   \ /   \ /
       \     \     \     \
___   / \   / \   / \   / \______
   \ /   \ /   \ /   \ /
    /     /     /     \
___/ \___/ \___/ \___/ \_________

And here's a variant of the single-chain 3-loop design, modified a bit to make it more interesting.

___   ___   ______
   \ /   \ /
    /     \
___/ \   / \   ___
      \ /   \ /
       /     /
______/ \___/ \___

After the last 5-banded puzzle ring got successfully prototyped, I came up with the following criteria for maximum difficulty of a ring: Adjacent bands shouldn't be linked, each pair of adjacent bands should have one band connected to the one of the left, one on the right, and one both but going through them on opposite sides, and the final positions of the bands connected to either side should be on the wrong sides of where they start.

So I came up with the following design:

______   ___   _________   _________
      \ /   \ /         \ /
       \     \           /
___   / \   / \   ___   / \   ______
   \ /   \ /   \ /   \ /   \ /
    /     /     \     /     \
___/ \   / \   / \   / \   / \   ___
      \ /   \ /   \ /   \ /   \ /
       \     \     \     /     /
______/ \   / \___/ \   / \   / \___
         \ /         \ /   \ /
          /           \     \
_________/ \_________/ \___/ \______

This is a 5-band design, where the bands form a 'ring' - each one is linked to exactly two others. It's almost symmetric, sort of. The small asymmetry
in this drawing could easily be removed, but then it would be very hard to take this one apart. There's actually a quite deep asymmetry here though. If you play with this ring, it will if properly made rapidly jangle itself into a ring of five bands. That's a big 'if' by the way, it requires the bands next to the center one be able to easily pass through the center one, which shouldn't be too hard given how 'punched out' the center band is with it being above everything else in the middle, although the oval bangle shape fights against that phenomenon. In any case, let's assume that it's properly made and that it jangles out properly and easily. The funny thing is, there are two different positions for the bands when jangled out into a pentagon of five bands, depending on which band passed through the center band, and both of those are distinctly asymmetric (they're mirror images of each other).

So whether the above design is a symmetric is a matter of some interpretation, and there's some trickiness in making it work well in practice (although the usage of bands passing through each other is quite novel and interesting). Since my friend who made a prototype of my last design specifically wanted maximum difficulty and asymmetry, I also came up with the following variant, which is most definitely asymmetric, and has even more of the bands passing through each other effect, but as a result it's trickier to make sure the passing through works properly in an actual built version, so it's higher risk from a construction standpoint:

______   ___   _________   _________
      \ /   \ /         \ /
       \     /           /
___   / \   / \   ___   / \   ______
   \ /   \ /   \ /   \ /   \ /
    /     /     \     /     \
___/ \   / \   / \   / \   / \   ___
      \ /   \ /   \ /   \ /   \ /
       \     \     \     /     \
______/ \   / \___/ \   / \   / \___
         \ /         \ /   \ /
          /           \     \
_________/ \_________/ \___/ \______

I rather like how both of these designs are natively 5-banded, in that if you remove a single band they aren't puzzles at all any more, and they have the very novel bands passing through each other effect as well as being very difficult just based on classical techniques.

Those of you who are into rare mechanical puzzles might be interested in that a puzzle of mine is going to be available in limited edition in the next few weeks. It's the 'unscrambled' which I did in collaboration with Oskar, and can only be made by 3d printing.

Say that you're at the end of a presidential primary debate, and the very last question is a softball one, hand-picked especially for you, which noone else gets to answer. Could you make a good impression with your answer?

Hillary Clinton found herself in exactly this situation, with the question "Which do you prefer, diamonds or pearls?" The correct answer to this question is "I like looking pretty just as much as the next woman, but I'm not fundamentally a materialistic person." Kind of an easy one.

Instead, the answer she gave was "I know I’m sometimes accused of not being able to make a choice, [but] I want both."

Way to look like a materialistic bitch, Hillary!

I've been mentioned on xkcd.

Apparently it isn't obvious that there is no such person as 'Elaine Roberts', because people keep asking me about 'her'.

There are some puzzle ring variants where a single 'band' goes around the finger multiple times. These have not to my knowledge been studied thus far. The best way to actually make them would be for the band which goes around multiple times to be made as a chain, with universal joints connecting the segments.

Here's a simple but pleasing example, with a single chain which loops around three times:

______   ___   ___
      \ /   \ /
       /     \
___   / \   / \___
   \ /   \ /
    \     /
___/ \___/ \______

Here's a slightly trickier example with two traditional bands and one chain. The principle behind this one is very interesting.

______   ___   ______
      \ /   \ /
       \     \
___   / \   / \   ___
   \ /   \ /   \ /
    \     /     \
___/ \   / \   / \___
      \ /   \ /
       \     \
______/ \___/ \______

The following puzzle consists of two chains which both go around twice, and is a bit of a departure because they can be completely separated when taken off. The two chains form a loop knot around each other, sort of, and the puzzle as a whole has a subtle asymmetry. It might be very tricky.

_________   ___   ___
         \ /   \ /
          /     /
______   / \   / \___
      \ /   \ /
       /     \
___   / \   / \______
   \ /   \ /
    \     \
___/ \___/ \_________

Most of the renewable power sources used for electricity come indirectly from solar, for example wind and hydroelectric. There are three other distinct original sources of renewable energy (maybe four, but the last one isn't really distinct from one of the others and is currently theoretical anyway). My basic physics question for you is, what are they?

Fundamental laws of physics dictate that all energy sources eventually run out. For the purposes of discussion I'll count anything which can maintain current output levels until the sun goes nova as 'renewable'.

I'm pleased to announce that we've hired a new CEO of BitTorrent Inc, Doug Walker. We spent a long time on this search, and believe Doug (who was the CEO at Alias Systems, maker of Maya) has the right software industry experience, skills, and energy to drive the continued growth of BitTorrent Inc.

My new title, since I'm no longer CEO, is chief scientist. What that really means is that I'm focusing once more on technical work, which is my core strength and how I got BitTorrent started in the first place. As such, I'm working with an extraordinary team of engineers and R&D team dedicated to developing the most reliable, well-behaved, and fastest P2P technology possible. I'm excited with this transition, both because of the person we've hired and because I get to go back to my roots as a technologist and focus on P2P innovation.

Many people don't know this, but the ferry building in San Francisco is situated in front of a currently functioning ferry terminal. The ferry is quite pleasant to ride on, and drops everyone off right in downtown San Francisco, without adding to the traffic.

A while ago there was an 'experiment' to add some security to the Larkspur ferry, which was a little odd, because there's no hint of there being a need for any security whatsoever. This was announced well in advance, and on the day of the 'experiment' there were about a dozen people giving full-on airport style security, and being quite self-important and obnoxious about it too. Several people appeared to have no role other than to hand out flyers. Regular ferry staff informed me that they were all making overtime.

There hasn't been any overt mention of it since then, and the only policy change has been that the gates are kept locked until a few minutes before boarding time, which I guess helps prevent fare jumpers, but mostly just makes people stand outside if they show up early.

Actually, there's an exception to that - every once in a while, there's a huge hubbub with DHS officers and local police cordoning off a wide berth from the ferry building in San Francisco, as if there's about to be a big explosion, and it causes everyone to be late. After a while they announce that it was all a false alarm and everyone can go about their business, albeit late. My guess is that somebody left their luggage inside when they went outside to watch the view (this is after all a tourist thing) then security found it and freaked out about their being luggage left alone until the person comes back in to reclaim it.

In any case, my assumption has been that everybody else found this 'experiment' just as nauseating as I did and told the DHS that they could shove it. Apparently not. There's some renovations happening at the Larkspur terminal now, and I just walked over to look at it, and there's a sign in front saying that it's a new 'security and inspection station' paid for by the department of homeland security. Great. Hopefully it's got bike racks which are kept secure, but my money says that the DHS will expect to have the right to do full pat-downs and luggage inspections on everyone boarding the ferry in exchange for doing us the favor of building yet another useless piece of infrastructure at the Larkspur ferry building.

Oh yeah, about that - there's already a boondoggle of wasted money at the ferry building, in the form of a giant triangular steel and glass structure to block rain and sun which is remarkably ineffective at blocking either. I've heard various stories about how that came to be, but the available evidence makes clear that it was something ridiculous - it obviously exists, it obviously cost a fair amount of money, and it obviously was a giant waste. But at least that last boondoggle didn't further enable the department of homeland harassment.

Most web sites do image uploading with the file selection dialog. This works okay, but for images it generally involves, saving to a file from some application, opening the file via a file selector, and then deleting the file when done. That's way too many steps, and requires using the OSX file selector, which I hate.

It would be vastly simpler if one could just post to file selector dialogs from the clipboard. Then I could copy from whatever application I made the image in, post in the web browser, and not touch the file system at all.

Could somebody add this functionality to firefox? Please?

How do you switch a tune from a major to a minor key? You turn it 'upside down' - pick some note to be the center, and then for every note which appears in the original tune, count the number of half-steps it is up from the center, and replace it with one which is the same number of half-steps down (or count down and replace with up, depending on which side it is). It's important that you count the actual number of half-steps, counting the gap between adjacent white keys with no between them as one half-step, same as the gap between a white key and the black key next to it. After transforming a turn you can move different notes up and down by octaves to make it sound a bit more normal, for example notes which started out just above the tonic can be moved up an octave to be back above it, and the bass line can be moved down a few octaves to be back into the bass range.

This trick is very, very different from having a filter which flips frequencies upside-down. That would change the timbre of notes in a very unnatural manner, and result in something which was consonant but quite alien-sounding.

Major and minor tunes are about equally consonant, the aesthetic difference is that in major keys most of the matching overtones are at the same place, so minor tunes sound ambiguous.

There recently was a man vs. machine poker match. Result: draw (ignore the claim that the humans won, the truth is that the sum across all days was close enough to zero that it should count as a draw).

From commentary on the match, it's clear that at least on some days the computer was playing a straight-up nash equilibrium approximation, which means that it was assuming a perfect opponent and only taking advantage of flat-out opponent errors. With just a little bit of intelligence on opponent modeling, it would be vastly better. My prediction: next match, the humans are toast. Then the interesting game will move from limit to pot limit, where top humans will still have a big advantage for a while.

Dear Mortgage Companies,

I understand that there are financial reasons to re-sell my mortgage every other month, and can even accept that it would be some real work to have an entity which I always send my check to whose name and address wouldn't change every time this happens. But why on earth does my old mortgage company not pay me the courtesy of telling me about my new mortgage company, and what exactly would be so difficult about the new company getting my mailbox address from the old company so I don't have to reset it every single time? Seriously, I'm sick of places I've never heard of calling me and demanding that I give them my social security number to authenticate myself for a very important message about my mortgage, and getting mail sent to a secure mailbox is just common sense.

Seriously, it's only a matter of time before we start seeing mortgage phishing scams.

Rola bolas are fun, but not sturdy enough to make a proper extreme sport. The roller tends to get away if you catch any air, and the board isn't strong enough to withstand doing ollies.

In the interests of extremitude, the roller should be mounted in the air on a fixed axle. The roller should also be given a small diameter, so that if you catch air you don't lose too much momentum by fighting against the roller on your way down. Also, the board should be made of something much stronger than it typically is, possibly even metal. You could then have foot bindings so you're attached to the board to enable catching more air.

The basic trick is to jump higher and higher in the air, using the roller to deflect yourself back up again, sort of like two people on a teeter totter but with one person doing both sides. This raises a question for the physics inclined: How well does that trick work? Anybody have any ideas?

A street park for these would consist of a bunch of rollers mounted in different places for people to jump between, and props in the middle to do grinds from.

Driving around, there are a few things which bother me about typical car UI.

When you're at an angle and stopped it can take quite a bit of gas to start going, then once you start going it requires a lot less force, as a result of which there's a tendency to push down the gas until you suddenly lurch, often backwards, because this happens when you're getting out of a parking space. It's possible to push down the gas delicately at a steadily increasing rate until the car just barely starts to move, but that's very tricky, especially when you're trying to not squash pedestrians in a parking lot. Modern electronics could easily detect what angle and speed the car is at and respond intuitively in such situations. This matters a lot in the San Francisco area.

There's no way to tell if the steering wheel is straight or has done a complete rotation to one side or the other when you first get into a car and turn it on. You can try moving and see which way it goes, but that's quite hazardous in tight parking situations. There should be some kind of indicator.

A much less trivial feature to implement, but one which other cars have and mine doesn't, is headlights which can change angle. They're used to point in the direction you're about to be when going around a turn. This serves two purposes: one, so you aren't going blind into turns in the middle of the night, and two (and more importantly) to accurately let other drivers know where you're about to be. Truth be known, most people can have their eyes adjust to the dark and see where they're going at night no problem, but it's very important that cars broadcast where they're about to be by highlighting it, so everybody else knows to stay clear.

IRC is a very baroque an strange protocol with a lot of history, which happens to enable some interesting hacks, one of which I'll now describe, and haven't seen implemented anywhere (although it could be in some client I'm unaware of, there's a whole lot of odd functionality buried in those things).

Fundamentally IRC protocol sends commands which are delimited by returns whose arguments are delimited by spaces, so it's impossible for arguments to contain spaces, with one exception - the final argument can contain spaces, and this is specified by preceding the last argument with a colon, so the parser knows that it's the last argument, and include everything from that point until the return as part of one argument (no argument other than the last one is allowed to have a colon as its first character).

IRC clients tend to not use this feature properly - they generally have lots of raw calls to write on the socket, rather than a dispatch layer, as a result of which some functionality including spaces in an argument is disabled by accident. In particular, if you do a who command for someone's whois info, it will find them even if they've changed their nick. This requires that they have unusual whois info (usually the case) and that the client use the colon trick in case the whois info contains spaces (which clients usually don't and whois info usually does).

Doing that properly would allow a nifty extension of the notify feature. Do a whois on the person who has the nick currently, then periodically do a who on that whois information to follow the person around across nick changes. Last time I tried this (which was, uh, ten years ago) this trick worked quite well, and didn't appear to be made available via any common client.

There's some controversy going on about the role of 'proofs' in cryptography. Basically, the argument goes: the mathematician says that cryptographic proofs aren't real proofs, since they require from a huge number of axioms with dubious correspondence to the system they claim to model, and even then are merely a reduction rather than a proof. The cryptographers retort: They're better than nothing [true!], so why are you complaining?

To which I would like to reply: Because protocol designers frequently design around what the provers can prove, resulting in far less secure protocols than if we stuck with ones people could actually understand.

There is remarkably little you should actually use in cryptography. Diffie-Hellman key exchanges are the beginning and almost the end of how you should set up an encrypted channel. (RSA has a performance advantage over DH which justifies it in practice, but just barely. It makes anyone who knows about it queasy.) The only block cipher mode you ever need to use is counter mode (the dunderheaded section on modes in 'Applied Cryptography' notwithstanding). For digital signatures there are a few standards to use, but with bandwidth getting cheaper it will soon become much more reasonable to use ones based on secure hashes than public key techniques.

Why then is there so much complexity in crypto protocols? Basically, because the people designing them think they're smart, and have an unjustified faith in proofs of security. If one uses common sense approaches to protocol negotiation (hardly ever done) and keeps the list of features under control (never happens) and views simplicity as an important design feature (hah!) then even cryptographic protocols are reasonably straightforward. And I don't mention simplicity as a feature just because I'm one of those lazy implementers. Historically, good humans eyeballing protocols and judging what they were comfortable with has been far more reliable than proofs of security, and putting on layer upon layer of crap so that the human can't analyze the protocol but the prover gives it a clean bill of health is optimizing for the wrong thing. In practice cryptographic protocol designers pour on tons of such rubbish, and convince themselves that their own vast intelligence enables them to do a reasonable eyeballing of the ridiculously byzantine results.

My own claim to fame in cryptography is relatively small. I designed a protocol which has done cryptographic processing of more bytes than the protocols done by all cryptographers put together. Cryptographically this isn't very notable, because the analysis of it is completely trivial. But that's the point.

In this article on the re-insurance industry there's a slightly tangential point I'd like to draw attention to.

In effect, the insurers weren’t insuring against disaster; they were only pretending to take the risk, without actually doing so, and billing their customers retroactively for whatever losses they incurred.

Insurance companies do the same thing with investment returns. When the stock market crashes, insurance premiums go up for a while as insurance companies replenish their reserve capital. Isn't it great how the market always sets prices perfectly efficiently?

As for rebuilding the rest of New Orleans, I don't think it's going to happen. The insurance industry has historically been disinterested in accurate pricing, and they have few incentives to fix it - any one employee is unlikely to hit a real disaster in their entire career, so they have no reason to hedge. And to the extent that anyone is actually doing the math of the newly available reasonable re-insurance, I suspect that the economics of rebuilding anything on the floodplain don't even vaguely work out, so rather than quoting offensively high prices the insurance companies are simply not offering plans.

With the current (apparent) crisis in the mortgage market in the united states, I'd like to point out a central fact of that system which is completely berserk:

Credit scores are the primary mechanism for determining creditworthiness for getting a mortgage.

This is more than a little bit nuts. Credit scores are designed to tell how much returns a credit card lender can get by giving someone a line of credit. Mortgage default risk is something else entirely. A person who has a huge rotating debt on their credit card and pays off the minimum every month is a great credit card risk because they'll have payed off several times what they were loaned in the first place by the time they go under, but a terrible mortgage risk because they're likely to go under, while a person who never has any rotating debt is a terrible credit card risk because they never rack up any interest but a great mortgage risk because they're unlikely to get into financial distress.

Why then, are credit scores used? First I have to explain what kind of risk mortgage lenders are taking on. Historically, mortgages were so absurdly overcollateralized that the risk of default isn't that the money doesn't get paid back, but that it gets paid back too early, requiring that some new sucker be found to loan the money to to get the expected returns. It's obviously a bit of a stretch to call this 'risk', but that's the way things have been.

So about using credit scores - since there's very little risk being taken on, mostly it's about who can be beat up. Whatever sociopolitically justifiable games banks can play to increase interest rates they do, and as it happens people with lower credit scores are viewed as Bad People and collusion against them is a lot simpler.

What, then, would be a reasonable thing to do? Obviously drop credit scores, but replace it with what? Well, the current situation can tell us a lot. What's happened recently is that housing prices have actually fallen, so a number of mortgages are underwater, so there's a risk not only of the money being paid back early, but of it not being paid back at all. The way to gauge risk of this is something which is already used in mortgage assessment, but not heavily enough, which is percentage down. By definition, if a houses's value drops by ten percent then any mortgage with less than ten percent down will be underwater, but mortgages with more than that will still be just fine. The current mortgage mess is in part difficult to unwind because when mortgages got packaged up and resold they were clumped by credit rating and not percentage down, so the effects of real estate prices going down on them are quite difficult to infer.

As for aftershocks, I don't think this is such a big deal. The people holding on to mortgages right now are mostly hedge funds, which by definition only have accredited investors as shareholders, and accredited investors by definition have enough money that they can afford to suck it up if their investments go bad. Hopefully that's what will happen.

In the future, I think mortgages will continue to be much as they have been right before this crisis. This historical business of mortgages only coming in one risk class (damn near risk-free) and getting returns as if they were taking on real risk was a bad thing, and I anticipate that if there isn't any bailout this time then the mortgage industry will get together and start handing out interest rates which actually make sense - very high rates for anything interest-only, and rates only a hair above fed funds for anything with 50% down, regardless of the borrower's credit-worthiness.

I've been reading the book Nature's Building Blocks, which I highly recommend, and a few things have stood out thus far. One of the more notable ones is the discussion of nuclear physics. Related to that, I have a few questions for anyone who might know the answer.

How come with all the journalist fearmongering about 'dirty bombs' they never mention simply dropping off radium-beryllium neutron sources in populated areas? For those who don't know, radium releases alpha particles, which when they hit beryllium causes it to release a neutron. Some shielding could easily prevent such objects from giving off conventional radiation, and sources of large quantities of nearly undetectable nasty radiation would obviously be a Bad Thing. It could be that neutrons so produced move too slow to get very far, especially since the beryllium itself slows them down, although some tradeoffs could be made there by having specks of radium in small quantities of beryllium instead of a block of radium in a block of beryllium. The materials might also be prohibitively expensive, although that appears to not be the case, or the yield in terms of number of neutrons might be low. If anyone has any real information I'd be curious to hear it.

Relatedly, would a parallelogram-shaped lens of beryllium be effective at getting neutrons from a source at the focus to all go in the same direction, or do neutrons get bounced in random directions whenever they hit a beryllium atom?

A bit more seriously (I never ask any question about why the nearly nonexistent terrorist menace doesn't do something seriously) what's wrong with using radium-beryllium (or polonium-beryllium) neutron sources as a way of getting lithium deuteride to give off lots of heat? If you embedded lots of little neutron sources in a block of lithium deuteride, there would be chained reactions giving off neutrons, without criticality because each reaction absorbs one neutron and gives off another, but if it's a big enough block those chains get long and the amount of heat could be substantial because it's all fusion. Such a block would make a great heat source to be used in the same way as the heat sources in standard nuclear reactors are used.

Speaking of which, can someone explain to me why thermoelectic generators (also described in Nature's Building Blocks) aren't standard? They have obvious reliability advantages, and their simplicity is appealing. I assume that either their efficiency or cost of materials is bad?

I don't like how little suits factor into poker play in hold 'em. They only really create flushes, and those happen rarely. Here's an idea for how to make the suits more important which I think enriches play: Instead of deciding equal hands based on the kicker, decide it based on who has more of a single suit. If two players have the same amount of a single suit, the one which has a higher bridge rank wins.

For example, if one player has 2H 2D 7H 9D 4H and another has 2C 2S 8S 6C JH, then the first player wins, because they have three hearts while the most the other player has of any suit is two. But if the first player had the 4S instead of the 4H, then they both would have two of a single suit and the second player would win because two clubs beats two of anything else.

If there's still a tie the second 'kicker' is the next most common (or weaker of the same amount) suit, followed by ordinary kickers.

Here's an interesting bit of trivia I learned the other day. Computer clocks will typically drift over time, at a rate determined by the ambient temperature. GPS units need extraordinarily accurate clocks, so rather than spend 5 cents on an oscillator like is done in laptops, they spend a few more cents, and put it next to a thermometer. At the factory they then measure how much that specific unit drifts at each specific temperature, and program in a little look-up table customized for that unit to correct for drift based on temperature.

Laptops actually have a thermometer as well, sitting right next to the oscillator. Why don't they correct for drift based on temperature as well? Well gee, I dunno, that would probably be a useful thing to write (although obviously it couldn't correct for drift while the computer was powered down).

Checkers has been proven to be a draw. You can play through an opening book here. Interestingly, if you go through moves marked 'd' it comes up with a single 'main line' for checkers, which I will now transcribe so you don't have to.

(from here)

1. 09-13 22-17
2. 13-22 25-18
3. 11-15 18-11
4. 08-15 21-17
5. 04-08 23-19
6. 05-09 17-13
7. 09-14 25-29
8. 08-11 27-23
9. 15-18 19-15
10. 18-27 15-08
11. 12-16 32-23
12. 03-12 24-20
13. 10-15 20-11
14. 07-16 25-22
15. 16-19

In an effort to find a puzzle ring design which is somewhat more difficult than the ones I gave previously, I came up with the following design, which is basically the first design of that earlier list with another band added:

____________   ___   ___   ___   ___   ___
            \ /   \ /   \ /   \ /   \ /
             \     /     \     \     /
_________   / \   / \   / \   / \   / \___
         \ /   \ /   \ /   \ /   \ /
          \     /     /     \     /
______   / \   / \   / \   / \   / \______
      \ /   \ /   \ /   \ /   \ /
       /     \     /     /     \
___   / \   / \   / \   / \   / \_________
   \ /   \ /   \ /   \ /   \ /
    /     \     \     /     \
___/ \___/ \___/ \___/ \___/ \____________

The way this one works out is quite beautiful. Usually puzzle ring designs with this amount of crossing and linking get hopelessly braided and can't come apart, or are completely inconsistent about which way the bands want to move, while this one comes apart quite easily and the natural position for all the bands in the fiddling state is quite clear. The difficulty level of this one is completely absurd.

The show 'America's Got Talent' has a top prize of a million dollars, so if you win it then you get a million bucks and never have to worry about money again, right? Well, not exactly. The prize is a million dollar annuity, paid out over the course of forty years, or the winner has the option to get the current cash value of that annuity immediately. It isn't clear how that cash value is calculated - it depends on the assumed creditworthiness of the annuity issuer, and few financial instruments go out that far, but it's definitely less than half a million dollars. After paying income taxes on that, you'll be left with an amount which makes for a perfectly reasonable down payment on an entry-level condo in San Francisco. My advice to anyone of not much means who hits a payday like that is to shove the money into a retirement account, then live your life exactly as you would otherwise only without having to worry about having the money to retire when you get too old to work. (This applies to winning the lottery as well, which uses the same sort of accounting for the announced prize values.)

Anybody care to guess how that discounted million dollars compares to how much the judges are paid for exercising their, er, 'judgement'? The dubious selection criteria aren't as bad as you might think - these sorts of shows tend to contractually lock people who do well in to management agreements with fairly dubious terms, so the best thing which can happen to an entrant is to get some airtime then get kicked off.

There are few technologies from the last hundred years which really make a difference in everyday quality of life. A modern mattress is good, as are a modern washer and dryer, a decent vacuum cleaner, a net connection with a web browser, and a phone which gets better audio quality than a hundred years ago. Well, all except that last one, because IT DOESN'T EXIST. Despite all the hoopla about the iPhone, 90% of its functionality is still voice calls, and it it still has the same crappy ones as everything else. You can't even change its ring tone to be an arbitrary mp3. It may have some cute gizmos, but the iPhone is still fundamentally a mediocre consumer electronics device.

Would some carrier, any carrier, please offer the feature that when calls are made within the carrier you get better audio quality than is available over the plain old telephone system? Is it too much to ask that background music sound like background music, rather than bizarre noise?

Rendering web pages is safari is cute though. It upgrades the phone-based web browsing experience from 'useful in case of emergencies' to 'useful when you don't have a real monitor around'.

Why does the fork have four tines? The Evolution of Useful Things has an explanation of how the fork came to be, but not a coherent explanation as to why the number stopped at four. More tines would clearly be better, because it would require less force to skewer food and hold it in place better. It appears to be an engineering limitation - tines are very thin, and you don't want them to bend excessively.

Can modern materials do better? The strength of a tine is proportional to the fourth power of its width, so the strength of a tine on a fork with five tines instead of four needs to be about (5/4)^4 ~~ 2.5 times as strong as a regular tine. My rather sketchy research indicates that titanium ought to be able to handle that, and in fact the tines on a titanium fork I got are about the right width for five to fit. To get to six tines, they need to be about 5 times as strong as with four, which might be a bit much, although a material such as tungsten, corundum, or tungsten carbide might do the trick.

I decided, given the above calculations, to try other materials of cutlery. My other reason for trying other materials is that I find metallic tastes overwhelming, and wanted to see if any of them don't interfere with my enjoyment of food.

The two types of cutlery I tried are titanium and lexan. Both of these seem to be intended primarily for camping, and so are designed to emphasize their light weight than their other desirable properties.

Titanium I found unfortunately to have a slight metallic taste. Less than stainless steel, and much less than aluminum, but still there. Also, titanium cutlery being designed for light weight while camping tends to be rather small. It does have the nice small tines, which could be used to give it five tines but isn't. I think if there were titanium cutlery which was made to the size and appearance of regular table cutlery, but with five tines instead of four, it would be quite nice.

Lexan I'm rather fond of. After having it in my kitchen for a while I find I always instinctively reach for a spoon or fork made of lexan, particularly the spoons, because of the amount of contact between one's tongue and the spoon when eating soup or cereal. In fact, I've noticed that when using metal cutlery I always hold my tongue back while biting in a weird and awkward way to try to avoid the taste of metal, but I don't do that using lexan, which makes the food eating process much more enjoyable. It also has very low thermal conductivity, which is very nice when eating soup or ice cream, although it is possible to heat lexan enough to melt it, so you can't use it like a spatula. The lexan knife is somewhat pointless though - it isn't very strong, and knives don't generally come in contact with your mouth anyway, so there isn't much of a taste issue. There's unfortunately an aesthetic limitation of lexan cutlery as well. It looks plastic and disposable, which it most certainly is not. I don't know if that's a result of the strength and production engineering needs of making lexan cutlery, or simply poor design.

With currently used materials, I think what I'd really like is a cutlery where the forks and spoons had lexan ends, for taste and low thermal conductivity, and metal handles, for strength, and the knife was an ordinary metal knife.

Possible new materials, and I'm completely speculating here, are tungsten, corundum, and tungsten carbide. Tungsten would probably be a lot like titanium only more so, and rather heavy. Corundum and tungsten carbide would probably score very well in terms of both flavor and strength, and it would be amusing to have cutlery made of sapphire, although cost might be an issue. Or not. The cost of both of those materials has plummeted recently, and cutlery used to be something which held a significant fraction of peoples's savings, while these days a set is considered expensive if it costs much more than dinner at a fancy restaurant. For items used every day for decades, a budget of several thousand dollars seems not unreasonable. People certainly are willing to spend bizarrely large amounts of money on wedding and engagement rings.

When I bring up the subject of cutlery and my observations it has an odd effect on people. They seem to get a desperate need to find some other subject to talk about, or go feed their cat. Very strange considering how fascinating the subject matter is.

The axiom of choice is presented as 'obviously' true, an axiom which can be simply thrown on the pile in the interests of convenience, without raising any notable philosophical hackles.

Bullshit. The standard statement of the axiom of choice is heavily spun to be obviously true of finite quantities, which is the only thing most people have intuition for. Digging into the details turns up the seedy underbelly...

Let's say that there are a countably infinite number of people (that's plain old infinite to those of you who haven't studied such things) and that each of them has a hat on their heads, either red or blue. Each of them can see the color of the hats on everybody else's heads, but not the color of their own hat. Each of them must write down a guess as to the color of their own hat, with no ability whatsoever to communicate with anybody else. Can you devise a strategy for hat color guessing such that there's guaranteed to only be a finite number of people who guess their own hat color wrong?

Obviously no, but I wouldn't have asked the question if the answer wasn't yes. I'll start with the reason why not.

What is the expected number of hats which are guessed wrong? Well, each person has a 1/2 chance of getting their own hat color wrong, so the expected total number is 1/2 + 1/2 + 1/2 + ... which happens to sum to infinity, which can't be finite. What's wrong with this reasoning? Not anything really, I'll get back to that in a moment.

So what is the strategy? Well, first let's define two assignments of all hats as being 'almost' the same if they only differ in a finite number of hat assignments. Note that if assignments A and B are 'almost' the same, and assignments B and C are 'almost' the same, then assignments A and C must be 'almost' the same. Likewise, if A and B are 'almost' the same, and B and C aren't 'almost' the same, then A and C can't be 'almost' the same. So 'almostness' forms an equivalence class of assignments. By the axiom of choice we can pick a canonical representative of each set of 'almost' identical assigments.

Each person then simply observes what assignment is on everyone else's hats, uses that to figure out what element of the equivalence class they're in (which they can by using either hat assignment for themselves, because it changes at most one hat) then guesses the color on their own hat by picking the one in the canonical representative. By construction, the number of wrong guesses must be finite, so we have our strategy.

What was wrong with the argument that the expected number of wrong guesses is infinite? Nothing whatsoever. The number of wrong guesses is a 'random finite number'. The probability that it's greater than any specific finite number is one, and its expected value is infinite. Does this mean that the statement that there isn't a 'finite' number which isn't in the set 1, 2, 3, etc. conflicts with the axiom of choice? I think it does, but would appreciate it if someone more well versed than I in the foundations of mathematics chimes in.

In answer to the question of the definition of 'finite', a finite set is one for which it's impossible to form a one-to-one mapping between a proper subset of its elements and all of its elements.

Thanks to Lance Fortnow for linking me to this craziness.

Let's say that some alien (aka trusted third party) came down and presented a very important secret to a bunch of humans. But instead of simply giving this secret to the humans, the alien gave a bunch of strings which when xored together form the secret one each to each of the humans, and gave the secure hashes of all the strings which need to be xored together to all the humans, so their integrity can be verified. The alien is now gone and no trusted third party is available. The humans all want to find out the secret, but don't want any other human to find out the secret unless they get it as well.

How to do the reveal? If there are a lot of people, there's a neat trick which makes it so no single person can wait until everyone else has revealed their info and then simply refuse to reveal their own. Everybody arranges themselves in a circle, then everyone reveals their info just to the person on their right. Once that's done (and only once that's done) everyone reveals the info they just got to the person on their right. When that's done they reveal the info they just got to the person on their right again, and so on until everyone has seen anything. If a person ever refuses to hand over the info, there are plenty of other people to get the info from, and the protocol as a whole will simply stall out if someone refuses to cooperate.

Obviously there's a lot of detail to be filled in here. I haven't even specified what information is globally visible, much less what the pragmatic responses to noncooperation are. A thorough analysis appears to be quite difficult, but my trick does have some utility, and it's an interesting problem.

Update: I edited the explanation because nobody seemed to understand what I was talking about. Hopefully it's better now.

One of the things I would like to improve in poker is that the hands which really give it texture, flushes and straights, don't happen often enough. I have an idea to improve that situation: get rid of one of the suits, resulting in a 39 card deck. That makes flushes and straights more common, and pairs less common. I propose getting rid of clubs, since that's the least poetic suit and the lowest ranking one in bridge.

Interestingly, going to three suits happens to invert the relative likelihoods of flushes and straights, so straights should beat flushes in this variant.