Older blog entries for Bram (starting at number 114)

Somebody trying to steal my domains

I just got the following mail. It looks like somebody's trying to steal some of my domains. Anyone got any suggestions for what I should do? From the rough translation babelfish does, it looks like it's saying that if I do nothing then the transfer will go through by default, which sounds too fucked up to be true, but hey, this is the domain name system we're talking about. This is also right after I told a couple jackasses who made offers for bittorrent.com to take a hike.

From: support@ovh.com
To: bram[at]bitconjurer.org
Subject: [TRANSFERT] bittorrent.com

140, quai du Sartel
Tel : 0 899 701 761
Fax : 03 20 20 09 58
Email : support@ovh.com

Objet : Demande d'acception ou refus de transfert
Vos Ref : bittorrent.com

Roubaix, 2004-07-19 16:29:42

Madame, Monsieur,

Nous avons reçu une demande de changement de registrar pour le nom
de domaine bittorrent.com et vous êtes l'un des contacts actuels de ce

| DOMAINE: bittorrent.com
| CONTACT: bram[at]bitconjurer.org
| CONTACT: bram[at]bitconjurer.org

Si vous acceptez le transfert du domaine, l'enregistrement
sera suivant:

| DOMAINE: bittorrent.com
| CONTACT: LT795-OVH (tlsx@wanadoo.fr)
| CONTACT: LT795-OVH (tlsx@wanadoo.fr)
| CONTACT: LT795-OVH (tlsx@wanadoo.fr)

Pour accepter ou refuser ce transfert, merci de cliquer sur
ce lien:


Si vous NE repondez PAS dans les 5 jours ET si AU MOINS un autre contact
ACCEPTE le transfert, le transfert sera effectué.

Si aucun contact ne répond, le transfert échouera.

si vous n'êtes pas indiqué ci-dessus comme contact pour ce domaine, vous
pouvez perdre tout privilège sur ce domaine. N'acceptez ce transfert que
si vous êtes sûr d'en avoir compris les implications.

Ce transfert a déjà été payé.

Pour toute information complementaire, contactez-nous :
Merci de votre confiance.

Le Service Client OVH

I got another one for bittorrent.org as well.


Computer chess programs have gotten very good. Not only in terms of raw computational power, but the very best have improved their play a lot. Deep Junior in particular has a very aggressive attacking style. Consider for example this game, played as black, which involves seemingly almost accidentally sacrificing two pawns so that a knight can take a liesurely stroll across the board to the opponent's king. Unlike many of the celebrated human attacks, this is done against Diep, a no-slouch computer competitor which placed third in the tournament. Many famous chess games turn out to have had extremely bad play by the losing side, with the celebrated attack being unfortunately unsound. With Diep as the opponent, it's doubtful that any simple or even not so simple tactics were overlooked.

Hopefully this newfound attacking ability will make there be not so many goddamn draws in high-level chess. The long-feared draw death has been essentially happening at the highest levels of human competition, and the way computers were able to defend much better than they could attack probably exacerbated the problem a lot.

Board evaluation in hex

I came up with an interesting heuristic for playing hex the other day. We will compute a score for all remaining cells, and move on the cell with the highest score. To compute scores, we will conduct several million independant random trial runs, each of which increase the scores of some cells. On each trial run, randomly place the remaining pieces in unoccupied cells. If you win, throw the run out. If the opponent wins, find each cell owned by the opponent for which the winner would be changed if that single cell were converted to your side, and increase the score of each of those cells by one.

One bad aspect of this algorithm is that it will play randomly when one side or the other has a near lock, since all scores will be zero. That could be fixed by changing the number of remaining cells owned by the opponent, to produce 'optimistic' or 'pessimistic' play. Comparisons of optimistic vs. pessimistic evaluations of about equal positions would be interesting.

The stochastic nature of this algorithm might superficially seem quite ludicrous and computationally wasteful. Ironically, all that CPU-eating is quite refreshing. We have unbelievable quantities of CPU, but when one tries to throw it at hex the same way one throws it at chess, the cycles simply disappear down a black hole.

I'm optimistic that this heuristic will do well. Clearly experimentation is needed.

Interestingly, this algorithm bears a strong resemblance to some trust metric ones I've suggested before. I think that's because hex sort of is like a trust metric - one side is trying to get two sides of the board to trust each other, and the other side is trying to keep that from happening.

60 cycle per second trailers

I noticed an interesting phenomenon in my bathroom the other night. The light on my toothbrush is very small, and cycle at 60 cycles per second (or maybe 120, I'm not sure), and, due to the way human eyes work, when the room is a completely dark leaves a very brief afterglow. As a result, when your eye scans over it a distinct trail appears for a split second. The distribution of the dots is interesting. Some acceleration is visible with the dots at the beginning being closer together, but the eye movement appears to stop pretty much instantaneously.

Financial Hijinx

For a while after I started work on BitTorrent I had no job and lived off savings. Eventually the savings ran out and I started living off of credit cards (don't worry, they're all paid off now).

Living off credit cards is an interesting experience. To do it for a while, you have to keep getting ever larger amounts of credit while keeping your interest rate low. Fortunately credit cards offer low introductory rates, frequently 0%, especially when you take out money on them as a balance transfer from another card. They particularly like doing this for people who have a history of paying off credit cards. For example, people who have been paying off cards completely by transferring to other cards when they get an introductory rate. Yes, this is completely gameable and insane, but it's also standard practice.

When undertaking to get credit cards, it's important to have as few inquiries in your credit as possible. Credit granting formulas are quite arbitrary and erratic, so to keep people from simply applying for credit until they get it, the credit agencies keep records of when you apply for credit and count that heavily against you later on. The types of inquiries aren't differentiated, so for example getting a cell phone counts against for getting a mortgage. Yes, this is also completely insane. To avoid having dings against your credit, it's a good idea to do as little applying for credit as possible. You can keep credit cards from seeing each others's applications by letting 'pre-approved' applications pile up for a while, then sending them all out on the same day.

If you have a very old collections for a very small amount of money and pay it off the payoff counts as 'recent collections activity', so if you need to get some credit in the near future you should hold off on paying it. This policy goes beyond insane into completely nuts, but it continues to be the case.

Once a very low rate on one card is about to run out you should, of course, transfer the balance completely over to another card. It's a good idea to do this later rather than sooner because there's frequently a fee for doing a balance transfer. It's always a good idea when given a new offer of a low rate to call and ask what they can offer you, because the people on the phone can frequently offer you a lower rate and waive balance transfer fees.

Some financial institutions let you transfer money into and out of your credit line via online banking without getting a cash advance charge. You should of course use this if you have it. Making a minimum payment this way is fairly amusing - you transfer money out of the credit card, transfer it back in, and presto, your minimum payment has been made.


Studies of interviews yield a very discouraging result. Interviews are practically worthless for screening candidates. In an interview you can tell if a person is a pleasant conversationalist, and you can give some technical questions to rule out the truly inept, but beyond that you might as well be rolling dice. This is counterintuitive in the extreme. Of all the different things you can say about a person, you'd expect there to be at least some correspondence with one of them. There's a whole long list of criteria, most of them completely illegal - pick candidates who are male, female, black, white, young, old, married, single, thin, fat, tall, short, blond haired, brown haired, etc. None of these have ever been shown to have any correlation with job performance, although I'd like to note that in an all-male work environment hiring an attractive female is probably good for morale.

If I ever have to hire someone from a few candidates, the first thing I'm going to do is call up their schools and employers listed to find out if they lied about degrees or dates of employment. Even though work history doesn't correlate with job performance, being a lying sack of shit almost certainly does. A surprising number of people lie on resumes, mostly because nobody ever checks them. One thing I most definitely will not do is ask former employers what they thought of the person. I've seen those calls get answered, and the quality of responses is dubious, to say the least.

Next I'm going to pick the person, if any, who I've already heard of, since they're likely to have done something worthwile. By 'heard of' I mean in a professional context. Being someone's cousin or loose acquantance doesn't count.

If I'm stuck in the end with several candidates none of whom I've heard of I'm going to pick based on the only criterion I've come up with which seems likely to have some correlation with job performance. I'm going to pick the one with the shortest commute time. Employees who can get to work quickly are going to have less incentive to switch to another job.

Bohemia Redux

For a while now I've been attempting to find an interesting two-player game of no chance which is based on the very special square board topology of viewing sets of four spots which form the vertices of an orthogonal square as being somehow connected. I think I've figured out how to do that now.

Players alternate placing pieces of their color in unoccupied squares. Every time a player places pieces of their color in all four corners of an orthogonal square they get a point. In the end, the player with the most points wins. The game can be made fair with a swap rule, and draws can be eliminated by giving either the first or second player to move the win in case of tie score.

This is essentially a territorial game at core, sort of similar to go in that you have to evaluate which parts of the board are most in dispute. The topology of the board is quite mind-bending, even calculating the score is a chore without a computer. But that mind-bendingness might be quite interesting.

27 May 2004 (updated 27 May 2004 at 01:55 UTC) »

I accidentally hit 'post' instead of 'preview'. Move along, nothing to see here.

Cayley Graphs, Permutation Puzzles, and Archimedean Solids

For many years I had an idea for an approach to solving rubik's cube. Just recently I found a trivial expository example of my approach. Jaap of course took my idea and ran with it and now has a gallery of quite remarkable correspondences between simple permutation puzzles and archimedean solids. (My contribution was the cuboctahedron.)

Gambling pools with odds

Racetrack betting has a very curious feature: When placing your bet, you don't know what the actual odds are. The current odds are readily available, but those in principle could completely change before the close of betting. An interesting variant would be to allow people to place bets on a horse which were contingent on the odds on that horse being at least a certain value. I haven't worked out whether it would always be possible to set canonical odds for a given set of bets, but I suspect it is. This would be an interesting (and, of interest to casinos, potentially quite popular) variant on normal betting pools.

Memory Allocation

graydon has some commentary on memory allocation. Specifically, he's in favor of explicit 'might be null' status for variable, and strictly tree-structured memory allocation. These are good ideas, but there are many thorny edge cases. For example, with non-nullable variables there's the case of an object's constructor passing 'this' to something which reads a variable which hasn't been set yet, even though it's guaranteed to be set once the constructor finishes. For memory allocation, usually that approach works fine, but there are occasional data structures which require making an exception to work properly, such as a circularly linked list. No single allocation scheme can capture the full generality of practical behavior, so the options are to either make a system which handles the vast majority of cases well (which strict tree-based doesn't quite) or have a system for violating the general rule in special cases.

I've spent quite a bit of time thinking about how to make a type-safe language with the declarative power of C++ (and more) but with the transparency of C. A friend of mine told me that one of my ideas amounts to 'jump tables but no vtables' - If you subclass an object, rather than making the class data structure point to a superclass object, it simply inlines a copy of references to all the methods it inherits from the parent, with overriden methods substituted. This vastly reduces the complexity of method lookups at runtime, both in terms of cycles taken and difficulty of proving correctness. The tradeoff is that in theory there could be an exponential blowup in the size of class objects in memory, but in practice that won't happen. (Well, maybe it will happen for those truly obsessed with OO wankery, but they get what they deserve.)

Advogato Feedback

I get very little feedback from this weblog. I know people read it because every once in a while I post an entry which generates lots of response (for example, the first one on solar thermal energy). Maybe this is because my entries are so obtuse that people aren't sure what to make of them. Or perhaps it's because I've made my email address so difficult to find. I had to do that, unfortunately, due to a deluge of not only spam but non-spam BitTorrent tech support requests. I simply couldn't keep up with them, and now make the assumption that anyone who I really want to hear from will make the effort to find my email address. I hope I don't lose too much important mail this way.

Spam Filtering

Speaking of spam filtering, I've now basically gotten my problem under control with the following filters - anything with 'dsl' or 'cable' in a Received: line but not in the From: goes to /dev/null. Anything with 'unknown' in the Received: line gets chucked. Anything with a locally tacked-on Message-Id: which isn't From: a locally hosted domain goes to /dev/null. And finally, anything which matches any of a very long list of bounce formats or bounce subject lines gets chucked.

I've taken this aggressive approach to spam filtering because even small amounts of spam were making me delete real mail thinking they were spam. Unfortunately people with even slightly dubious mail setups are likely to have their mail to me /dev/nulled now. And of course I never get bounces any more. I hope I don't lose too much important mail this way, but I undoubtedly lose some.

Gambling on Wall Street

The Springfield daily newspaper, the Daily Star, reported that their local baseball team was slumping, and had missed the spread five games in a row (which was true). Richard Smallbrain, CEO of the Company of America, wrote a scathing op-ed piece criticizing the paper for tainting its sports coverage with gambling. Smallbrain then also mentioned that his company is a a growing part of Springfield's economy and has beat wall street expectations three quarters in a row.

How to make money off of terrorism

The traditional spy novel scenario for terrorists profiting from their work is quite dubious. Usually the scenario is that the terrorist shorts the stock of a public company before attacking it. This is a bad strategy on a number of fronts. To begin with, shorting requires that you place stock or cash in reserve, thus limiting the amount you could possibly make to a small fraction of the amount you can front. Further, the potential upside of a short is capped at the amount you get up front, thus limiting how much you can make that way even more.

Those problems can be mitigated by using puts instead of shorts, but a much more fundamental problem remains. Attacks frequently fail, and even if they do succeed they often do a lot less damage than you expected. The rebound of a stock after it shoots down can leave it higher than it started. Plus, to reliably reduce the value of a public company requires doing a whole lot of damage, and some people have moral objections to doing that.

There is a much more reliable market-neutral way to make money off of terrorist attacks. First, go long and get puts on a company's stock, with strike price at the current value and maturation times for both on the same day as far in the future as possible. Now you will be out some money and have an interesting bet going where the more the company's stock moves, in either direction, the more money you make. (A nice benefit to this approach is that options with a strike price of the current value tend to be much more liquid than the more extreme ones). If you were to hold onto these options until maturity, this would be a very risky bet, but you aren't going to do that - if you unload them in just a day or two, the bulk of change in the value of your options is going to be in the volatility level of the stock, so even if the stock price is back where it started, if the volatility has gone up you'll make a substantial profit.

Next, set off a smoke bomb in the company's main plant. If that isn't your style you can just give a few ben franklins to your friend who's about to appear on CNBC to get him to say that there are rumors that the company is about to be acquired. In a day or two everyone will probably have figured out that whatever mischief you started was no big deal, so the stock will likely go back to where it started, but in the interim volatility will have gone through the roof, so you can unload the stock with a fat profit.

If you can't be bothered setting off smoke bombs, or happen to not want to do anything illegal, there's a perfectly legal technique based on this approach which makes money due to an artifact in the way that volatility is calculated. In general volatility is calculated by averaging it over the last N days. You can search for stocks which just so happen to have had a lot more volatility over the last N/2 days than the preceding N/2 days, then long and put them, wait N/2 days, and unload your positions. Since the first half of the range will now have passed, and the volatility level in the just passed N/2 days is likely to have been between the two values, the volatility measurement will now be higher, and you can probably unload your position at a profit.

The Recursive Universe

ANKOS follows a remarkably similar format to The Recursive Universe, a much older, and much better book. The Recursive Universe is much more rigorous, uses Conway's life, and doesn't have all that pompous bullshit which ANKOS is notable for.

A funny tangent - another book by the same author, William Poundstone, gives the formula for KFC's secret spices: salt, pepper, and MSG.

The automata I mentioned in my last entry is rule 54, described on pages 696 and 697 of ANKOS. The simple-looking diagrams on page 697 are deceptive - that apparent simlicity only happens on a large scale, there's a whole lot of complicated churning which goes on before things settle down.


Codeville has been progressing rapidly, and is self-hosting and able to talk to remote repositories.

The Axioms of Choice and Foundation

I have a question - if the axiom of choice is such a good idea, what's wrong with the negation of the axiom of foundation? Don't they both state the existence of absurd things? Aren't they both very far removed from practical mathematics? ZF is intuitively bothersome this way - it allows for some unreasonably powerful statements, and has axioms to undermine those statements enough to make it reasonable.

A lot of the pro-AC propoganda people are taught in undergrad school involves things which can be straightforwardly constructed without AC, hence the seeming intuitiveness of the common statement of AC - it strongly intuitively implies a bunch of things about the sets involved which would make AC not necessary. Things which need AC tend to be along the lines of the Banache-Tarski, which is completely absurd.

Really Big Numbers

Scott Aaronson sent me mail a while ago, which promptly got lost in the disaster area known as my inbox. I just found it, so I'll comment on it now.

Who can name the biggest number? is, as I commented earlier, an interesting game. Large numbers loosely correlate with busy beaver machines which can use oracles corresponding to large infinite numbers. The problem with this is that we don't, and can't know which large cardinal axioms are consistent, especially when we're trying to be aggressive about naming the largest possible one.

The next step, at least to my mind, is to talk about 'the largest consistent large cardinal axiom which can be expressed with less than a million symbols'. This is analagous to, but more well-defined than, Aaronson's 'The biggest whole number nameable with 1,000 characters of English text', which suffers from a self-referential Russel-style paradox. Numbers named using this technique have an astounding level of inaccessibility, almost like we said 'The largest number which can be imagined by a human mind plus one'. We can contemplate the existence of such a number, but by definition we cannot imagine it.

I sure like picking on Wolfram

Scott also reviews ANKOS and points out a number of outright errors in the book. In particular Wolfram's comments about rule 110 being 'NP-complete' are unsubstantiated, since the construct used involves an exponential number of steps, and hence computing it isn't in P.

However, it is true that we now know that rule 110 isn't computable, once you deal with the encoding issues of computing starting at what state and what you want to know if the output finally does.

That said, my earlier comment about there being computable but nontrivial rulesets analogous to Presburger arithmetic wasn't flippant. I've spent a fair amount of time playing with Wolfram's rule sets and a definite category which he never mentions emerges. Many of the rules produce very messing looking output on random input, but any vaguely simple input results in obvious non-repeating but still simple patterns. In ANKOS Wolfram himself talks about one of the most complex-looking ones of these and says that he ran two million simple inputs and they all resulted in straightforward patterns. Somehow he completely avoids even speculating that this rule is less strong that fully general computation. Maybe he's in denial.

I'd very much like to see it proven that many of Wolfram's automata are computable. Partially because any such proof would be very deep and interesting, but also because I don't like Wolfram's attitude and would very much like to see his central thesis thoroughly debunked.

Public Key Cryptography

This paper gives a public key cryptosystem very similar to the one I came up with, and they manage to prove some solid results about the difficulty of their primitive, while I merely conjecture at the difficulty of mine.

I'm quite happy to have come up with a public key cryptosystem, despite it being comically impractical. Almost everyone can come up with a symmetric cipher which they themselves can't break, but designing any public key encryption algorithm at all which can't be broken trivially is quite difficult.

How to make money at gambling

Some gambling games, such as craps, are simply you the gambler versus the house. These games aren't really 'gambling' so much as they are simply handing money to the casino. Games such as racetrack betting are fundamentally different, in that you're competing against other gamblers with the house merely taking a cut, so in principle it's possible to win money by out-betting your competitors.

The most fundamental principle of serious sports betting is to always bet on the underdog. Intuition indicates the exact opposite, since you want to pick the likely winner, but the goal of sports betting isn't to reliably pick the winners, it's to pick the competitors whose winning changes are underestimated by the other gamblers.

There are several reasons for this. Most obvious is the human instinct to pick the single most likely winner. Also brand name creates bias, because in the absence of analysis people are much more likely to bet on a competitor they've heard of.

A much more insidious phenomenon biases even statistical methods towards the favorite. Given a field of equal competitors, after a few competitions some of them will do better by dumb luck. It is therefore necessary to attribute at least some success among the competitors with good reconds to good luck and some of the failure among competitors with bad records to bad luck, but this is difficult to do. Even worse, frequently competitors which do poorly in previous competitions don't even compete in the later ones, so the size of the pool of competitors which the lucky few remaining are culled from, and the importance of luck in their success, is much larger than it appears.

Unfortunately betting on the longest shot every time takes a long time to win money reliably. There are so few wins that luck plays a major factor. Fortunately there is a way of betting on many long shots at once and have the chances of them winning be somewhat independant. In the case of horseracing, we know that in a given race exactly one horse must win, so if any horse loses that makes it more likely that each other horse wins.

To win more reliably, do as follows: In any given horse race, sort the horses by ascending likelihood of winning, according to the odds. Set a high water mark such that according to the odds the chances of any horse above that point winning is 1/2. Throw out all the horses above the mark, then place money on each of the horses you haven't thrown out in such a way that your payout will be the same if any of them wins. For example if one horse pays out 10:1 and another pays out 5:1 you put twice as much money on the 5:1 as on the 10:1.

Remember to place bets as late as possible, so that the odds drift as little as possible between the time when you place your bets and when the race actually runs. If everyone were to follow this strategy it would cause wild odds swings during the last few minutes, but fortunately this betting strategy is so incredibly lame that the chances of it becoming that popular are practically zero.

While I'm sure this approach does reliably much better than random betting, and its payout is relatively consistent over time, practical horseraces may keep enough of the pot for the house that the long term trend is negative. Unlike on wall street, there are very few hidden fees in sports betting, but the commisions for the house can be exorbitantly high.

Mozilla Planning

Brendan Eich, May 15, 1998

4. XXX-1998 (let's say it's 1-Oct-1998): Declare the trunk stable, tag it with a blessed CVS label that anyone can use to derive product from (particularly, but certainly not exclusively, Netscape for Communicator 5.0), and then:

5. XXX plus 1: open the tree to next-gen layout, aggressive modularity, other changes. I should emphasize that the difference between 1-Sept and this date cannot be many months, or we're doomed.

Mozilla 1.0 shipped June 5, 2002.

Brendan Eich April 5, 2004

If we do only "the browser thing" [...] we are likely to be irrelevant in approximately five years. Not unused, but trailing-edge; not built upon as a platform, not invested in even as a hedge.

Draw your own conclusions.


Apparently GMail uses a ton of javascript and frames, and feels more like a client side application and a web page. What pisses me off about this isn't anything Google has done, in fact quite the opposite. Massive amounts of client-side javascript was my vision for the future of web applications as a young punk back in '96. A good example is this demo, which has worked since Netscape Navigator 2.0 and the concurrent version of IE. Of course, we all know know what has become of that vision - basically nothing, since a combination of javascript buginess and general fear of doing real software has relegated javascript to mostly ornamental use on web pages.

I hope that my demo at least helped inspire someone working on GMail. My own youthful enthusiasm for doing javascript got squelched by beurocratic mismanagement in the company I worked for at the time. Earthweb, later to become one of the most grotesgue dot come IPOs and flameouts of them all. I hate seeing hustlers get rich.

It's good to at least see that my original approach had some merit. Just kidding about the last eight years of web development, kids. The approach of ambitious but naive hackers which we had from the very beginning was the right one.

Puzzle Extension

Wriggle puzzles have a fascinating property - It's possible to extend a puzzle to make another puzzle which is strictly harder, because any solution to the harder puzzle also works as a solution to the easier puzzle. This applies to worm extension, addition, and fusing. Sokoban has a similar property under the addition of new walls. It's interesting to find places in the standard sokoban levels where a wall could be added without making the puzzle impossible.

A very interesting poorly behaved game can be derived from this property - two players alternately extend a shared puzzle until one of them challenges the other to find a solution, and the other player either comes up with a solution and wins or fails to do so and loses. Interesting tactics include making a particular region of the puzzle only solveable in a tricky way which you already know, so there is a solution but it's computationally difficult to find.

105 older entries...

New Advogato Features

New HTML Parser: The long-awaited libxml2 based HTML parser code is live. It needs further work but already handles most markup better than the original parser.

Keep up with the latest Advogato features by reading the Advogato status blog.

If you're a C programmer with some spare time, take a look at the mod_virgule project page and help us with one of the tasks on the ToDo list!