Slashdot Mirror
Slashback: Dataplay, XviD, PPC
Pins and needles, pawns and bishops. s20451 writes "It looked grim earlier in the week, but following a fifth game meltdown by Kramnik and a brilliant game 6 by Fritz, the computer has tied the match 3-3. Betting on the computer in game 6 would have brought you a 7-1 return! I'll be on the phone to Vegas."
The new, new, new economy has room for camels. SwiftOne writes "According to their website, The Perl Journal has gotten enough subscriptions to begin online release (the planning of which was previously covered, along with the concerns about not reaching their goal. The first (next) issue is expected in early November."
Maybe it was the 15th-mover disadvantage. melt writes "Dataplay, the Boulder-based manufacturer of quarter-sized recordable discs and drives, finally called it quits on Friday, October 11, 2002. The remaining 120 employees (who have been on furlough for the past few weeks) have been let go and the company has closed shop. They are looking for a buyer for the remaining pieces. Full story at the Rocky Mtn News web site."
Zoom in until you see little canyons ... Twirlip of the Mists writes "IBM's chief scientist for their iSeries family of servers (a.k.a. the AS/400 family) has an article on iseriesnetwork.com describing the somewhat confusing history of the POWER4 microprocessor. In light of recent speculation about a possible relationship between IBM and Apple, this article is of particular interest. It clears up-- at least partially-- some of the complex, incestuous relationships between the PowerPC architecture, the PowerPC processor family, and the POWER4 processor. As an added bonus, there's some talk about the upcoming POWER5 and POWER6 processors near the end. The key phrase (and disclaimer): 'expected to appear in 2004.'"
Shame on Sigma.
Gruturo writes "After almost 3 months the XviD project and website have reopened, though Sigma Designs has not complied yet with all their requests (they still carry their copyright on many modified sources). In these last 2 1/2 months the project still went underway, although unofficially:
B-frames are practically ready, motion estimation algorithms have been improved, work started for Qpel implementation."
Please stop teasing us. If you liked Cryptonomicon, you've probably been impatiently watching the announcements of when the next Stephenson book would appear. wka writes "Previous false starts notwithstanding, Amazon says Neal Stephenson's new novel Quicksilver will be published in January."
And next week, building box-girder bridges. scubacuda writes "Lawmeme has released Part III to their Law School in a Nutshell series (Part I and Part II were previously featured on /.)"
Dataplay, the Boulder-based manufacturer of quarter-sized recordable discs and drives, finally called it quits on Friday
Well, can't say I'm surprised. While there are a VERY few uses for drives this small, the demand is pretty dang small. For what I'm sure is much less, you can get much more storage at a resonable size. So while the technology is very, very cool... it realy doesn't have enough people to support it. (Unless it was made by a big manufacturer like IBM that also did many other things...)
As cool as quarter size media might be, I'm not entirely sure that is the way the market should go. 3-inch media seem to be a better bet (certainly as far as compatibility goes) and would be far less likely to get lost or get eaten by babies and pets. When I think of the ideal medium, I think of something that can fit into my shirt pocket, but not so small that it get lost in my hair.
Some interesting quotes:
Instead of using a sequence of instructions to perform a common function, the operating system will use a single instruction that causes the entire function to be performed by the POWER5 microprocessor hardware. Examples of these common functions include TCP/IP processing, communications message-passing operations, and virtual memory subsystem operations, to name a few. The interfaces to all of these silicon accelerators will be open so that other operating systems, for example Linux, can take advantage of them.
and
When POWER6 arrives in 2006, it is expected to extend the Fast Path idea to even higher-level software such as DB2 and WebSphere processing. Again, all of the silicon accelerator interfaces will be open, so other software developers wil be able to take advantage of the improved performance.
Bill Gates once said that when a given bit of functionality is sufficiently standardized, it should be part of the OS.
No we will make it part of the CPU.
The law is a weapon of the government, not a protection for the likes of you. Surely you understand that.
Try reading this Kuro5hin story and posted commenst about that exact thing:
2 69
http://www.kuro5hin.org/story/2002/7/11/62356/9
If you don't, a computer sucks at Go because of the exponentially larger solution sets involved.
Not quite right. Even *Deep* Blue couldn't map out an entire game, or even all possibilities in its 25(?) moves ahead. That would constitute perfect play, which is totally beyond the capability of any computer that we could currently conceive of being built. It would stop processing a line of possibility if it looked too bad (like "hmm, if I make this move I lose my queen and two rooks, so let's stop worrying about that one).
I don't understand what you mean by saying that the hard part is writing a random number generator. Random number generation it itself doesn't have anything to do with much. The question is the algorithms used to find the more likely moves.
As for Go needing 10 times as much storage, you are so far off that I worry that you don't know the meaning of the number 10! Each chess move gives about 35 legal options. A player in a Go move has about 200 possible moves on average (the number starts at 361 and mostly goes down from there). After five moves from each player there are about 1.8 billion possible positions in chess -- and 64 trillion for Go. That's a factor of 32,000 more positions, and that's only five moves in. Go games usually have more moves than chess games.
It is really laughable to even suggest that all the possible moves in Go will be stored in a computer within the next 500 years. Though that isn't necessary to beat a human shodan (as I mentioned, chess programs don't evaluate ALL of the possible positions). What's really necessary to beat a human master at Go is to be able to make some judgement on the relative value of different positions. Computers can't currently do that properly, so while a chess computer searches for that perfect move that forces checkmate, the computer playing Go has a hard time understanding what it's supposed to be searching for.
A good article I found and got some numbers from is http://www.anusha.com/times-go.htm.
I have seen the future, and it is inconvenient.
Actually, the exponentially larger sets involved are not the main issue. Sure, it makes finding a good move a longer task, but remember Moore's law: computer power grows exponentially too. So, if that was the only problem, we could guaranty that computers would eventually become go experts.
However, the main problem is deeper. It is at the root of the combinatorial game theory that sits behind Fritz' victories: the evaluation function. As mentioned in the article you link to:
Computers are just number crunching machines. If you can't feed them with numbers, them will do you no good. So that's what this evaluation function does: it takes a given board situation, and evaluates its value to the player. These values are computed for all the leaves of the explored moves tree, and are used to prune this tree down. Finally, you are left with the best next move. Therefore, 'best' depends not just on the size of the tree, but also on the quality of the evaluation function.
The bottom line is:
And that is exactly what the problem is. While it is pretty easy to say (compute!) who has an edge on a chess board, we still haven't found a good computational way to do judge the value of a go board situation. Of course experts go players DO have a feeling of a better situation, but that's exactly it: a matter of feelings, not of equations. Until we have put these go feelings into equations, whatever the processing power, computers will just suck at go.
I code, therefore I am.
Fritz with the queens on is a different animal. It's a monster
If I were on the programming team I'd take that as an expert suggestion to add a negative adjustment to the evaluation of queen exchanges.
-
- - You can't take something off the Internet! That's like trying to take pee out of a swimming pool.