Ray Bradbury said, ""

Woo!

I just installed Sabayon. Shiny.

All the power of Gentoo, with less need for sacrificing goats.

Posted by: Tom "The Pooklekufr" Treloar at July 13, 2008 06:20 EST | Permalink | comments |
geekery

Comments open but moderated. I reserve permission to kill spammers on sight.

Moar Zombie!

I just found an excellent zombie apocalypse, the Monster Island trilogy. Shiny.

Posted by: Tom "The Pooklekufr" Treloar at July 06, 2008 05:11 EST | Permalink | comments |

Comments open but moderated. I reserve permission to kill spammers on sight.

Shiny

I just brought a cheap new box. For $284 I got a 1.8 Ghz single-core, 160 gig HDD,CD/DVD-RW box. Not much, but then again the most computationally expensive thing I will use it for is compiling Gentoo packages and my own code.

Funny- think about how much this would have cost only 5 years ago. And it is still an order of magnitude less win than the next cheapest computer I could find.

Here's hoping FedEx doesn't lose it.

Update: Within 10 minutes of setting it up, I will have burnt .iso's of Ubuntu, Gentoo, Slackware, and BeOS. Within half an hour I will have ported all my Ubuntu programs and documents, and within an hour have all OS's up to speed. Approximately one minute after successfully compiling the last program, I will immediately cream my pants.

Posted by: Tom "The Pooklekufr" Treloar at June 28, 2008 14:09 EST | Permalink | comments |
geekery

Comments open but moderated. I reserve permission to kill spammers on sight.

Stumpwm

i installed Stumpwm, a tiled mouseless windows manager written in Common Lisp. Absolutely amazing.

I can change everything in it at will. Open a command line, enter code, eval, and instantly have a new functionality. I can tinker in its guts, tweak it, and make my computer completely unuseable for anyone but me.

I'm going to work on it in my free time, fixing it up a bit; also I plan on developing a little for Closure, a CL web browser.

Posted by: Tom "The Pooklekufr" Treloar at June 22, 2008 06:11 EST | Permalink | comments |
geekery, hackery

Comments open but moderated. I reserve permission to kill spammers on sight.

What's my mental bandwith?

Psychological Parameters and Learning: how many bits per second do you learn?

Posted by: Tom "The Pooklekufr" Treloar at June 19, 2008 23:59 EST | Permalink | comments |
geekery

Comments open but moderated. I reserve permission to kill spammers on sight.

The Parenthetical Way

I need to get more in tune with the Parenthetical Way. It's not enough to play around (although Lisp is by far the most fun language for playing with), I've got to put its awesome productivity and super-power-bestowing mindset to use. I need deadlines, challenges, and the ability to gloat over Java programmers who are 10 times slower than me.

I'm going to select some interesting projects to work on. Various open-source Lisp projects, Lisp operating systems, CL Gardeners, etc. Maybe I'll play around with the Netflix Prize.

Whatever I choose, I will use three criteria. A)It must be fun, B)it must be challenging, and C)it must be useful. Yes, this is all a roundabout way of achieving my goal for Project Vinge.

Update: I'm coding in CLisp (predominantly), CMUCL, and Scheme. BTW, I haven't figured out how to invoke SLIME with different Lisps, at least yet (and its Installation manual doesn't help with multiplatform loading. Anyone figure it out?

Posted by: Tom "The Pooklekufr" Treloar at June 15, 2008 04:09 EST | Permalink | comments |
geekery

Comments open but moderated. I reserve permission to kill spammers on sight.

Three Links Toward The Parenthetical Way

CLiki

CMU AI Repository

Source Code from Norvig's Paradigms of Artificial Intelligence

Posted by: Tom "The Pooklekufr" Treloar at June 14, 2008 18:41 EST | Permalink | comments |
geekery

Comments open but moderated. I reserve permission to kill spammers on sight.

North Korean Realism

North Korea. Professional. Thriving. Photoshopped.

Picture from Banking Services in DPR Korea- North Korea's official site.

Posted by: Tom "The Pooklekufr" Treloar at June 12, 2008 23:29 EST | Permalink | comments |
humor

Comments open but moderated. I reserve permission to kill spammers on sight.

Truly, very sorry

Sorry for mentioning it, but you just lost The Game.

Posted by: Tom "The Pooklekufr" Treloar at June 12, 2008 18:30 EST | Permalink | comments |
geekery

Comments open but moderated. I reserve permission to kill spammers on sight.

Juxtapositions

Garkov: Markov Chains and Garfield FTW

A brief description of Algorithmic Information Theory

Posted by: Tom "The Pooklekufr" Treloar at June 12, 2008 13:39 EST | Permalink | comments |
geekery

Comments open but moderated. I reserve permission to kill spammers on sight.

Holy Erdos!

I just saw a new level of beauty and elegance in algorithmic information theory.

*breathes deeply*

If number theory is the Queen of Mathematics, AIT is the ghost of Erdos coming back from the other side with The Book.

Posted by: Tom "The Pooklekufr" Treloar at June 11, 2008 18:31 EST | Permalink | comments |
geekery

Comments open but moderated. I reserve permission to kill spammers on sight.

Bayes Factor and Marginal Likelihood

Note: this is mostly just a public brainfart as I try to assimilate this.

Bayes Factor and Wikipedia's page

Suppose I find a buffalo nickel in my attic. Hypothesis One is that it is fair, Hypothesis 2 is that it is biased toward Heads 9 out of 10 times (suppose I know my house was once a gambling parlour). Given the former hypothesis, I ought to expect half the flips to be Heads; given the latter I ought to expect 9/10ths of the flips to be Heads. In sexier notation:

P(H|H1) = .5

P(H|H2) = .9


How should I intelligently accumulate evidence toward each hypothesis?

Let's compare the ratio K of prior probabilities for each hypothesis given flipping a Head and flipping a Tail.

K_H = P(H|H1) / P(H|H2) = .5 / .9 = .56
K_T = P(T|H1) / P(T|H2) = .5 / .1 = 5

What does this mean? Intuitively, it means that flipping a Tail ought to be weighed more heavily as evidence against Hypothesis 2, than as evidence for Hypothesis 1. Flipping a Head is less decisive than flipping a Tail. By how much?

Let's convert K to decibels:

K_H = 10 log_10 .56 = -2.51 dB
K_T = 10 log_10 5 = 6.98 dB

This tells us that a Tail is a hell of alot more decisive evidence than a Head. Flipping a head gives us only -2.52 dB of evidence for Hypothesis 1 and against Hypothesis 2, while flipping a tail gives us 6.98 dB of evidence for Hypothesis 1 and against Hypothesis 2.

Just for the hell of it, let's convert K to bits:

K_H = .083 bits
K_T = -2.3 bits

We should conclude that a Head is barely significant evidence for Hypothesis 1 or 2, while a Tail is substantial evidence against Hypothesis 2.

Ok, where am I wrong and fuzzy-headed?

Update: a good analysis of P-value to Bayes Factor. Hat tip: Black Belt Bayesian

Posted by: Tom "The Pooklekufr" Treloar at June 10, 2008 17:07 EST | Permalink | comments |
bayesianity, geekery

Comments open but moderated. I reserve permission to kill spammers on sight.

Lottery

Assume you want to play a lottery. Each number can be from 1 to 52, and you must choose the winning 10-digit sequence.

The probability of winning is:

(p(x))ⁿ where p(x) is the probability for each unit, n the number of iterations. Thus

p(winning) = (1/52)¹⁰ ~ .019¹⁰ ~ 6.13 x 10^-18

How many decibels of belief should you allocate to this outcome?

d(winning) = 10 log_10 p(winning) ~ -172 db.

Remember how decibels work.

Alternatively, you could ask how much information you'd have to possess to justifiably believe you will win:

b(winning) = - log_2 (p(winning)) ~ 57 bits. If you cannot answer 57 yes-or-no questions regarding why it is certain that you will win this specific ticket, you are acting on ignorance. If you are not prepared to give me 5.7 bits of information for each number you have chosen, then you are acting on ignorance.

One way to look at it, is to ask to what other outcomes should one allocate roughly -172 decibels of belief.

A pair of dice can have any integral value from 2 to 12. There is a 1/36 (.028) probability of rolling a 2. After a little calculation on your own, how many consecutive rolls of 2 are on the same order of magnitude of (im)probability as winning the lottery? Would you bet the price of a lotto ticket on this?

The cognitive bias that lets you view these identically improbable events as different, is incidentally the reason why the elderly gamble more than the young.

As Richard Dawkins pointed out, were you to live a couple million years, you would expect to see a royal flush or two, a few "once in a million year coincidences." Or as R.A. Fisher said, "The million, million, million ... to one chance happens once in a million, million, million ... times no matter how surprised we may be that it results in us."

So we should expect that the average octagenarian has seen some improbable things. We should expect that he will have more stories of strange coincidences than his twenty-something grandson. Having seen one "once in a lifetime" improbability and a few "once a decade" chances and many "not your everyday" events, this octagenarian will have allowed confirmation bias and selective recall to skew his sense of improbability. He doubtless remembers a lucky night in Vegas where he rolled 7's all night long- but he passes over the next night where he lost his house. His sense of probability has been squashed flat.

Where his 20-something grandson sees the lottery as improbable, on the order of rolling more than a dozen consecutive snake-eyes, this elderly gent sees the lottery as just as probable as that night he bumped into a long lost friend out of all the millions of people in New York. Hell, he probably thinks pointing out that he has, in fact, once rolled 15 consecutive snake-eyes is evidence of how likely he is to win the lottery.

I wonder. Maybe this same cognitive bias skewing the sense of improbability, also accounts for the inordinate presence of the elderly in the polls. Despite what we think about the elderly being cynical and cantankerous, perhaps they really don't understand how vanishingly improbable it is that any politician will actually improve things.

Update: Read Eliezer Yudkowsky's Anchoring and Adjustment. It turns out we are rather poorly equipped to compare likelihoods. For each of us, there is probably a limit to how finely-tuned our sense of improbability is, and anything under this limit we will unthinkingly lump together. If you knew roughly where that limit is, and understood on a gut level what it means to allow that amount of sloppiness in your estimates, you would likely be much more well equipped to avoid various cognitive biases.

Posted by: Tom "The Pooklekufr" Treloar at June 10, 2008 01:52 EST | Permalink | comments (1) |
geekery, bayesianity

Comments open but moderated. I reserve permission to kill spammers on sight.

I shall call it...

I woke from a siesta today with a stunning book idea. A probabilistic analysis of Charles Dickens coincidences, embedded in an information-theoretic detective story, set in a New Jersey neuroprosthesis clinic. Oh, the main character is an AI simulation of Franz Kafka (choosing secretarial AI's has become a subtle form of humor).

The interesting thing about the book, is that its plot will have such a low Kolmogorov complexity that this blog post has just described it in the fewest number of words necessary.

Take that, Hemingway.

Posted by: Tom "The Pooklekufr" Treloar at June 09, 2008 22:56 EST | Permalink | comments |
geekery

Comments open but moderated. I reserve permission to kill spammers on sight.

This has nothing to do with Richard Pryor.

Let's play the Aumann Game.

The rules are at the linky, but the gist is simple: for each statement, without using google-fu, tell me your prior gut feeling regarding the probability it is true. For each answer p(x), your score is tabulated by converting to belief decibels (10 log_10 p(x)), then summing over each answer (according to logarithmic summation rules). Alternatively, you could convert each answer to bits (- log_2 p(x)) and sum to arrive at how much information about the world you think you are giving.

Note: assigning zero probability, absolute impossibility, to any statement drags your entire score to (10 log_10 1/infinity) ~ negative infinity. This is not good, you just expressed perfect, psychotic, infinite skepticism and, at the same time, godlike omniscience. Certainty, in contrast, raises your score toward (10 log_10 1) ~ 0 decibels.

Alternatively using bits, assigning absolute improbability to any statement raises your entire score to infinity: you are attempting to describe all of reality. In contrast, assigning certainty drags your score by one bit.

One of the morals of the game: absolute certainty means you are saying nothing, and absolute doubt means you are saying everything. Between this Schylla and Charybdis, so tempting to philosophers and politicians, the sane man must walk. When in doubt, doubt a little less.

Statistics

1. The population of Transylvania is less than the population of New York City.

2. India has more enclaves (enclosed regions) in Bangladeshi borders than vice versa.

3. Tennessee is wealthier than Saudi Arabia.

4. More people speak Klingon than Lojban.

5. San Francisco is more population dense than New York City.

History

6. In the 1950's Fidel Castro's brother Raoul crashed in a flying automobile.

7. Moresnet almost voted to make Esperanto its national language prior to WWII.

8. President McKinley was governor of Ohio.

9. No people died as a result of the Three Mile Island accident.

Future

10. Before 2020, there will be mass adoption of exabyte hard drives.

11. An artificial intelligence will not be widely acknowledged to have passed a Turing Test until the year 2050.

12. Some country will, in the next decade, using currently available biotechnology or nuclear technology, destroy over 35% of another country's population.

13. England will restrict immigration policies by 2010.

Counterfactual

14. In a randomly selected Everett Branch splitting from ours on(IARSEBSFOO) January 20th, 2000, Al Gore won the presidency of the United States of America.

15. IARSEBSFOO 4 billion years ago, a meteor collision rendered the earth sterile.

16. IARSEBSFOO 13 billion years ago, an almost insignificant assymetry caused the universe to evolve with a charge distribution opposite our own (into predominately antimatter).

17. IARSEBSFOO this moment, your double will experience the plot of your favorite fictional novel.

Internet

18. "Vegeta" gets more google hits than "Vegetable".

19. "Paris Hilton" gets more google hits than "Paris France".

20. Google can accurately correct you if you mispell the proper name "Janowycz".

21. More bits of pornography are, at any one moment, being downloaded than the entire information traffic of Canada.

Update: Feel free to suggest alternate scoring techniques. Before doing so, read a little about Information Content and Bayesian Inference. In any case, read Jaynes' eerily available textbook on Inference.

Posted by: Tom "The Pooklekufr" Treloar at June 08, 2008 17:48 EST | Permalink | comments |
geekery, bayesianity

Comments open but moderated. I reserve permission to kill spammers on sight.

Evolution Is a Perl Hacker

Evolutionary biology/psychology indicates that our most complex functions are assembled out of creative arrangements of pre-existing designs. Evolution does not create novel designs, it creates novel arrangements of previous designs using a minimum amount of work necessary.

Consider the Panda's Thumb argument. Evolutionary biologists thrive in the search for suboptimal solutions and kludges. They relish in discovering that the panda's thumb is literally hacked together out of metacarpals, that our retinae are arranged backwards, that our genome contains vast redundancies of the ALU sequence repeated millions of times. Evolution shows itself to be a lazy copypasta programmer, in all her forms.

We should expect this. Our ordered complexity has been added piece-meal because it was not possible to do so otherwise; reducing the entropy of organisms is inherently an incremental process. Entropy and evolution?

Consider Richard Dawkins' idea of an evolutionary metric in probability space. Roughly, each genetic sequence has an empircally determined probability p of mutating at any specific location. Each successive n incremental mutations have a likelihood of pⁿ. For example, a genetic string of AATAAGCAATAGC... has a p probability of changing to AACAAGCAATGC..., and a p² probability of differing from the original by two units. Knowing the base rate of such elementary mutations, we can therefore determine the evolutionary distance between organisms probabilistically: we can determine how many mutations must have occurred in the speciation of organisms. Since the base rate of mutation is so phenonomenally small, we should expect to find it breath-takingly unlikely that any short-term one-step mutation would yield a viable new organism or organ. This is what Haldane realized when he refuted saltationism, the idea that evolution proceeds by large-scale mutations: it is so extremely improbable that any such mutation would be a coherent organism, that it is certain speciation occurs through extremely long processes of tiny, somewhat improbable changes.

Put simply, considering the difference between an organism without eyes, and an organism with eyes, the probabilistic difference between the two is so astronomically huge that only a lengthy incremental process of tiny alterations could explain it.

All this is simply information theory in disguise.

We should therefore expect evolution to be an obessesively conservative design process, greedily retaining previous patterns to be recycled. And indeed this is what we do see: enormous redundancies and kludges and legacy code and all the other hallmarks of an inefficient design process. The only thing to recommend evolution is that given geological time to work in, it could achieve any advanced complexity at all.

Now for the interesting stuff.

Your brain works algorithmically. Emotions, cognition, sensory processing... what makes you you is a vast conglomeration of algorithms hacked together over billions of years using a minimum of work and no intelligence whatsoever. The algorithms necessary for you to determine meaning in these black squiggles on the screen, although vastly more complicated than anything we can currently program, are nonetheless algorithms capable of being analysed for their speed and efficiency.

And improved in Lamarckian time rather than evolutionary time.

Algorithmic analysis tells us that the difference between a poorly designed algorithm and a faster one, is absolutely astronomical. A slight improvement can mean the difference between spending centuries working on a problem, and spending a few seconds. Moore's Law of hardware improvements is virtually insignificant compared to algorithmic advances.

For the same reason that we need not fear giving birth to a mongoose, we should expect that applying our intelligence to our evolved algorithmic processes would yield unimaginable advancements.

[I'm working on a series of posts delineating this argument]

Posted by: Tom "The Pooklekufr" Treloar at June 08, 2008 15:21 EST | Permalink | comments (2) |
geekery

Comments open but moderated. I reserve permission to kill spammers on sight.

Vacation!

So what if my boss forgot to send up for vacation pay, leaving me with almost no disposable income after bills for possibly the next two weeks? Vacation!

I plan on doing little more than read, program, see my girlfriend before she journeys to London, read more, program more, and achieve Bayesian enlightenment by mastery of the Magical Turing Machine. Oh, and blog a bit more at length.

I have 192 hours of freedom. That's 11,520 minutes. That is 2,304 five-minute chains of thought available to me; over 230 times Mencken's Limit. That is at least an order of magnitude more time than necessary to realign one's priors according to the way reality actually works.

Hmm. Neurological research estimates that the human mind is capable of storing about 2 bits/second (will find linky later). At this rate, during one week of focus and abundant coffee it ought to be possible to learn a maximum of 1,382,400 bits. At about 4.5 bits per English word, that is 307,200 words, or about 614.4 paperbacks at 500 words/page. That's 25.6 books a day.

Next time you complain that you only have 24 hours in a day, think about what's possible.

Update: I spent rather too much my DI on Gregory Chaitin's Meta Math: The Search for Omega (could not afford his Algorithmic Information Theory), Lou Anders' Futureshocks, Gregory Bear's Cosm, and Greg Egan's Schild's Ladder. I saw Alonzo Church's Introduction to Mathematical Logic, but buying it would have meant mild starvation.

Posted by: Tom "The Pooklekufr" Treloar at June 06, 2008 15:49 EST | Permalink | comments (1) |
books, life, geekery

Comments open but moderated. I reserve permission to kill spammers on sight.

Morning Book Rush

Every day around this time of morning I must search frantically for books to carry me through the day.

I hunt through my books, trying to find something, anything, that looks capable of transporting me during the interstices of the day. Some bizarre algorithm makes me put back a copy of Vernor Vinge's Peace War and linger over a volume on linear algebra. Then I sieze upon Purves Augustine et. al's textbook on neuroscience. Now for another book, possibly cyberpunk pre-singularity without too much Gibson short stories, or even a Mark Twain so long as it is his slightly embittered era and not too long before or after...

Do non-bookworms have an analogous morning rush? What you put into your mp3 players, and such?

Update: Selected the neuroscience textbook, a John Varley anthology, and Wittgenstein's Tractatus

Posted by: Tom "The Pooklekufr" Treloar at June 04, 2008 04:58 EST | Permalink | comments |
books, life

Comments open but moderated. I reserve permission to kill spammers on sight.

You win some, you lose some

Ever have one of those days when you secretly relish the thought that under the Everett Interpretation of QM, there are a transfinite number of yous doing things you wish you'd rather be doing?

Exploring Ganymede, watching Sholay in a dusty 1970's Indian cinema redolent of curry and goat, seducing Ada Lovelace, writing the last LISP macro of a Bayesian superintelligence, herding towering masses of virally mutated immortality sheep, watching a Dyson Sphere erase the last glimmers of a sun under darkly radiating nano-engineered material, strangling your way through your Historical Black List in alphabetical order...

And then you realize that for each aleph-null you in a happy place, an aleph-null number of yous are waiting on line in a DMV.

Posted by: Tom "The Pooklekufr" Treloar at June 03, 2008 22:50 EST | Permalink | comments |
geekery, life

Comments open but moderated. I reserve permission to kill spammers on sight.

Wee Morning Link

John Baez's Information

Now I can go to sleep.

Posted by: Tom "The Pooklekufr" Treloar at June 03, 2008 02:47 EST | Permalink | comments |
geekery

Comments open but moderated. I reserve permission to kill spammers on sight.

This site is a member of WebRing. To browse visit Here.