I said a couple of weeks ago that I'd write up one of the contradictions which believing in Egan-esk consciousness. (Just to recap, that means that you don't think there is anything magic about physical brains which precludes them running, in simulation, on a computer. Thus you're happy with uploads and all that other good stuff.)
Let me be running a discrete time simulation of your brain on a computer. At some time t you are in state s. I simulate you by running an update function which maps s to s+1, based on some input (vision etc) and produces some side channel outputs (your movement etc).
Now, it's a computer. I can simulate two people and each will run at half the speed. It doesn't matter that the process which is simulating your brain gets time sliced out for a few hundred milliseconds - it doesn't affect the computation at all.
So, in the middle of calculating s+1 I can go off and do other work so long as I get there in the end. It appears to you that the rest of the world is speeding up (because you are slowing down).
So I can just treat your state vector as a huge number and increment it with overflow. There is a lot of other computation which will occur before we hit s+1, but we will hit it. If we keep going we will reach s+2 and so on.
However, every single being with a state vector less than, or equal, to yours in length will live every possible life while we do it.
That's certainly a little odd.
As you'll see from my STM talk (see last post), or any other time you ight have seen me talk. I try now to put works on slides. When I talk you have to listen to me, I might talk about code examples or graphs on slides, but I don't try to bullet point everything I'm going to be saying. I've never liked it because I don't like to listen to talks which have it (I usually feel that it stunts the presenter as they fall back on sticking too closly to the slides).
But it seems that I'm right: “Research points the finger at PowerPoint” (via The Register)
You can see me giving a talk on STM (specifically the GHC implementation) at Google. I wasn't actually very happy with this talk - I didn't feel like the audience really got it and that's my fault since they were plenty smart enough. I guess I should have taken more time at the beginning to get everything up to speed before diving into the examples. It's very tough, once you know something to remember the aspects that you took to understanding. And, even if you chart it, that doesn't mean that the same steps will work for anyone else.
So Aaron writes to defend the Chinese Room Argument (go read that link if you don't already know the argument - you should).
I am absolutely a child who grew up reading Egan (who writes awesome short story books - the full lengths ones are not so great) so the idea of uploaded minds etc is as normal to me as trans-pacific shipping (I happen to be watching the ships in San Francisco bay at the moment).
Take a neuron from my brain. It's certainly complex: it has many inputs, both dendrites and chemical, and its actions are poorly understood at the moment, but it's nothing magical. It's matter and fields and probably some quantum physics in there. If you think there is something magic about them - you can stop reading now but you have a lot of proof to provide.
But if it's not magic, we could replace a neuron in my brain with a functional clone and I'm the same person. I'm not suggesting that we could do that tomorrow, but that we could do it at all. Repeat many times, and you have a concious person with a non-natural brain.
Unless you think that the result isn't concious. Did that feature fade out as more neurons were replaced? If so, and since our artificial neurons are assumed to be perfect functional clones, you do believe that there's something magical about them it seems.
On the other hand, if I have a concious person with my crazy brain, why can't it run in simulation? My artificial neurons can be implemented using beer bottles and bits of string. It really doesn't matter.
You say that informational processes have to be interpreted to mean anything. But conciousness is a process reflecting upon itself. That happens whatever the hardware is. Yes it leads to some crazy-sounding conclusions and it's certainly not good for one's sense of self importance, but I'm lead here by the above reasoning which seems sound to me and so I accept it.
Maybe I'll write up one of those crazy conclusions later.
By way of chump I ran across eCache [Tor hidden service via public HTTPS proxy] which is another crypto based gold backed currency.
I have a fondness for these things, I'm that way inclined and I recently took advantage for my daily communte time to listen to Rothbard's What Has Government Done to Our Money? (everyone should take the time to read it). It seems that there's quite the industry behind eCache (by which I mean "not very much", but it's still more than I expected).
eGold has existed for a long time and claims to have 35K 24-hour active accounts, which is tiny compared to VISA, but pretty shockly high otherwise considered. Then there's unlinQ, which converts eGold/eCache to virtual-credit cards (although "only available in fixed denominations, can't be reloaded or refunded" suggests that you loose a lot in the granularity since you generally can't pay for things with multiple cards).
eCache claims to have 310 grams of gold, which is about $7500 worth of backing. Again, hardly going to set the world on fire, but it's more than some guy in a basement somewhere. I have no hope that it will go anywhere, but it's nice cyberpunk porn.
I'll be presenting this paper at SPIE in San Jose next Wednesday.
Google Books: Making the public domain universally accessible
Here's something I knocked up: it's both a literate Haskell program and a HTML file describing the JPEG file format
This week I ended up with one of these: 
.
It's a Seagate SATA disk in an external SATA enclosure. I didn't have an external SATA port on anything useful, so I decided to take the hard drive out and put it in a USB enclosure. This shouldn't have been hard.
Sadly, Seagate decided to use Torx Plus Security screws on the enclosure. They are like torx screws, but with only five points and a pin in the middle. Even that took a bit of finding out, and you can't get them in sets of "Security bits". You can get them on Amazon for $80. Yes, $80 for a set of 11 screw bits.
So I drilled the buggers out and even going slowly didn't stop two of the screws welding themselves to the drill.
Why on Earth do Seagate make that kind of crap? Don't ever buy one.
Documentation and source code.
So, yesterday, I utterly forgot a password which I use several times a day. I'd typed it twice in the morning and, about mid afternoon, just stopped. This password was so instinctive I just couldn't believe that I couldn't type it, but no amount of frustration over the next 30 minutes changed that.
But I remembered that I was logged in using it in a Firefox session which was running. So I managed to extract the password from the memory of the firefox process (more on that later) and, although I now had the password, it didn't make any sense. There was no “Oh! That's it!” moment, and no muscle memory when I tried to type it.
So I was safe, but a little bit freaked out. A couple of hours later (after a beer, which should have been the obvious solution all along) I sat down at the computer and unlocked the screensaver without even thinking about it. (The screensaver needs the same password). I took several minutes before I realised and now ever having forgotten the password seems as impossible as ever having known it did a couple of hours ago.
I guess it's only going to be a few short years until I start walking into rooms having forgotten why I went there in the first place.
But anyway, I wrote up a very hacky program for extracting passwords from the memory of a running Firefox process - see the comments at the top for limitations.
This is great. He extracts timestamps from the TCP ISNs and uses that to measure skew of the CPU clock and thus the temperature. We can then tell how hard the CPU is working by measuring the temperature increase.
People already know that time is a side channel and that you shouldn't leak information in the speed in which you process queries. Now you have to heat the CPU uniformly too. Giving out such good timestamps via the TCP ISN is probably a bad design and should be stopped. But there are many ways to get a timestamp from modern systems and noise just means that you need more samples.
Say that we wanted to implement message-passing concurrency in Haskell. Haskell is a nice language which is ruled by its type system like a Mafia don. (It's easy to design yourself into a corner from which there's nowhere to go if you're not careful.)
From first glance, things look promising. GHC has lightweight threads and message channels already. But, unlike Erlang, the message channels are typed. So what's the type of a message? Well, we could implement a ping/pong example with:
data Message = Ping | Pong
That works fine, but it means that for every different type of message in the system we have to have an entry in that union type.
So let's try using the class system to define an interface which an actor can implement:
data Message b = Xon Int | Xoff Int | Reparent Int b class DataSource a where xon :: a -> Int -> IO () xoff :: a -> Int -> IO () reparent :: DataSink b => a -> Int -> b -> IO ()
The message sending operations are all implemented as IO () functions. Note that we have also used the type system to enforce that any actor passed to reparent has to implement the DataSink interface. Sadly, Haskell can't cope with cycles in the module graph, although cycles in the reference graph are just fine. So if DataSource named DataSink they would both have to be in the same module.
newtype MySinkThread = MySinkThread (Chan DataSink.Message) instance DataSink.DataSink MySinkThread where write (MySinkThread chan) fd dat = writeChan chan $ DataSink.Write fd dat closed (MySinkThread chan) fd = writeChan chan $ DataSink.Closed fd
There's the instance of a DataSink class, which has two messages: write and closed. The methods just add a message to the channel object. If the actor had several interfaces to implement we would need to create a union type of the Message types of each of the interfaces. This means that having default values for the class methods isn't going to work because we don't know how to wrap the values which get pushed on the channel. I expect we can use templates to generate all the instance decls.
We can use a monad very like ReaderT to thread the channel through the body of the actor:
newtype ActorT t m a = T (t -> m a)
runActor :: t -> ActorT t m a -> m a
runActor thread (T func) = func thread
instance Monad m => Monad (ActorT t m) where
return a = T (\_ -> return a)
T func >>= k = T (\r -> do a <- func r
let T func2 = k a
func2 r)
instance MonadTrans (ActorT t) where
lift c = T (\_ -> c)
this :: ActorT t IO t
this = T (\r -> return r)
And then there's the question of how to write receive. Currently I don't have a better solution than so pass a series of values to a unary lambda with a case statement and to catch the exception if the match fails:
receive $ \x -> case x of
DataSink.Write fd d -> ....
The receive also needs a way to store the values which didn't match so that they can be tried with future receives. Probably the ActorT can carry a list of such values. Also, any pattern match failure in the receive body will cause the next message to be tried - I don't know any way round that.
In short: could work fairly well with some work
open_memstream and fmemopen (both link to the same manpage). These functions are for passing to libraries which expect a FILE * and you want to keep a buffer in memory. Certainly beats writing a temp file.
Did you wonder what that JBIG2 compressor was for? Now you know
“The final nail in WEP's coffin”
Headline: all the hardware works and seems to work well.
Warning: the first time I tried this Bootcamp chomped the OS X filesystem and make the system unbootable. Pay attention to the Bootcamp warnings and backup any data you care about first.
I'm working from a completly clean OS X install here so you may have already done some of these steps:
Select software update from the Apple menu and get 10.4.7 and all the fireware options offered (I only had one). After the reboot goto http://www.apple.com/support/downloads and search for “mini”. You want two fireware updates and the first (SMC) should have happened via Software Update. The other should now be listed. I don't know how stable the URLs are, but currently it's at:
http://www.apple.com/support/downloads/macminiearly2006firmwareupdate101.html
Follow all the instructions about installing the firmware and don't interrupt the process, lest you want to turn your mini into a brick.
Boot into OS X (you can hold the Option key at startup to select the boot volume - but it should be the default) and install rEFIt from refit.sf.net. Reboot and check that everything works. rEFIt is an EFI bootloader which will be booting lilo for us.
Get bootcamp from http://www.apple.com/bootcamp and install. It ends up in Applications/Utilities.
Run it, don't make a driver CD and partition the disk however you want. Click “Restart Mac OS X” and check that the system can still boot.
Insert a Gentoo 2006.0 minimal install CD and shutdown the system. Press the power button and hold 'c' to boot from the CD.
With the 2006.0 install kernel the Sky2 NIC will work, but the wireless doesn't (yet) so you'll need a cabled Internet connection.
You must use parted to partition the disk. The partition table is in GPT format and fdisk doesn't know about it. (Also, remember that when configuring the kernel).
After the partitioning and mkfs you can check that the system still boots. But when you run the install CD again, the kernel won't know about any of your partitions (because it's a GPT, which I'm guessing the install kernel doesn't know about). Run parted and delete and recreate the partitions with the exact same numbers again and the kernel will suddenly know about them. You don't need to mkfs again (unless you changed the numbers). (When you do this in future you need to set the boot flag, rerun lilo and rerun the rEFIt partition tool because parted seems to clear all that.).
Now, it's a standard Gentoo install. See http://www.gentoo.org/doc/en/gentoo-x86-quickinstall.xml
You can find the kernel config I used at http://www.imperialviolet.org/binary/macmini/kernel-config. It's probably not perfect for most people, but it will get you booting. The important points are:
Device Drivers -> SCSI -> low-level drivers -> SATA support -> Intel PIIX/ICH Device Drivers -> Network Device -> Ethernet (1000 Mbps) -> SysKonnect Yukon2 ... -> ... -> Wireless LAN -> Wireless LAN driversI'm using lilo. The /etc/lilo.conf config should look something like:
lba32 boot = /dev/sda map = /boot/.map prompt delay = 50 vga = normal image = /boot/bzImage-2.6.17.3 root = /dev/sda3 read-only label = 2.6.17.3Run lilo Run parted and do:
Reboot
At the rEFIt startup screen, select the partition tool and sync the MBR.
With a bit of luck, rEFIt will give you the option to boot from the Linux on the HD. You should now be booting into Linux.
Using the new, modular X, you want to set VIDEO_CARDS=i810 in your /etc/make.conf and you need Xorg 7.1 to drive the 945GM found in the mini mac. To get that you may need ACCEPT_KEYWORDS=~x86
I needed some tricks to get the correct resolution because my panel size (1680x1050) isn't in the native list of video modes. If you need this, see instructions a http://klepas.org/2006/04/09/ubuntu-on-the-inspiron-6400/
You can see my xorg.conf at http://www.imperialviolet.org/binary/macmini/xorg - note the odd ForceBIOS option for getting the resolution correct. The 915resoltion arguments I'm using are: 915resolution 5a 1680 1050 32 (yes, set 32-bit depth here and 24-bit depth in the xorg.conf).
The 2.6.17.3 kernel needs a patch to get the sound unmuted. Get it from here: http://hg-mirror.alsa-project.org/...
The madwifi.org drivers for the Atheros work perfectly. Follow the newbie HOWTO on that site.
Then just reinstall OS X and try again.
Insert the OS X install CD and power on. Hold down 'c' to boot from the CD.
The installer seems to make little or no effort to partition a disk so, if there are no volumes given as a target to install to, don't panic. Select Terminal from the menu and use diskutil to repartition the disk. The command will be something like diskutil partitionDisk disk0 1 "Journaled HFS+" OSX 1M.
I've gotten to go ahead to open source an JBIG2 encoder I wrote for work, look for that tomorrow or so.
Summer of Code is active. Go look at that project list (even if you aren't going to do anything under SoC it's a central TODO list for open source software: something we've not had before).
And this is a really cool paper about high speed concurrency in Haskell - with source.
Fantastic: after too much effort being put into Real-based stuff, radio 4 finally has downloadable MP3s of some shows. That's been going on for a while but the selection was very limited. Top of the new list is Newspod - a pick of the day of the BBC news. No the mention that the ever wonderful Now Show is also available.
I have a (fairly) well tested, fast onlines codes implementation which I would post here except that it's clearly not worth the pain when there's a crazy company which believes that it has a patent (baseless or otherwise).
At least that company does, in fact, make something I guess.
| / | Root |
| Alternate | The Weird and Wonderful |
| Backlinks | What are backlinks |
| John Gilmore | What's Wrong with Copy Protection |
| Archives | Blog Archives |
| One | Archive 1 |
| Two | Archive 2 |
| Three | Archive 3 |
| Four | Archive 4 |
| Five | Archive 5 |
| Six | Archive 6 |
| Seven | Archive 7 |
| Eight | Archive 8 |
| Nine | Archive 9 |
| Ten | Archive 10 |
| Eleven | Archive 11 |
| Twelve | Archive 12 |
| Thirteen | Archive 13 |
| Fourteen | Archive 14 |
| Fifteen | Archive 15 |
| Sixteen | Archive 16 |
| Seventeen | Archive 17 |
| Eighteen | Archive 18 |
| Nineteen | Archive 19 |
| Twenty | Archive 20 |
| Twenty One | Archive 21 |
| Twenty Two | Archive 22 |
| Twenty Three | Archive 23 |
| Twenty Four | Archive 24 |
| Twenty Five | Archive 25 |
| Twenty Six | Archive 26 |
| Twenty Seven | Archive 27 |
| Twenty Eight | Archive 28 |
| Twenty Nine | Archive 29 |
| Thirty | Archive 30 |
| Thirty One | Archive 31 |
| Photos | Poor People Caught on Film |
| Jack and the Beanstalk | Jack and the Beanstalk |
| RIP Scan | Results of a Stage Scan Fire |
| Yosemite | Yosemite National Park |
| Projects | Incomplete things from the lab |
| Seagull's Bane | Linux Automounter |
| bttrackd | BitTorrent Tracker |
| CAPTCHA | CAPTCHA CGI script |
| Conserv | Console Serving |
| Deerpark | Using Tor with Firefox/1.1 (Deerpark) |
| DNSFix | Fixing DNS |
| Xovers | XTA Crossover Control |
| IAFS | Archive Org Storage |
| JBIG2 | JBIG2 Encoder |
| Verify | PGP Key Verifier |
| MaxFlow | Maximal Flow in Python |
| PyBloom | Bloom Filters in Python |
| pyGnuTLS | Python wrapping of GnuTLS |
| Sxmap | Apache SuEXEC Map |
| Hellard | Union Server Notes |
| Recordings | Free recordings |
| ICSM Choir | St Paul's Church |
| School | Ancient School Stuff |
| Writings | Who knows |
| Cap Systems | Capability Systems |
| Intro | Introduction to me |
| Suprema | JMC2 Group Project |
| MP Letters | Letters I've written to my MP |
| Sound | Sound With Dramsoc |
| SyncThreading | The wonders of user-land threads |