Information about a computer account I run on the US chesslive server.

Inside Links

The YaceComp Newsletter
Yace Open Discussion

Outside Links

HiddenDragon v Mark Noble(FM)
US Chess Live
Linkwatcher
Google
Winboard
SCID Database
Frank's Chess Page
Computer Cheating
This Week in Chess
Chessopolis

Don't click this!

Congratulations to Yace! In a tournament featuring 14 of the best free Winboard chess engines, Yace took first place with 36 points out of a possible 52. Since it was held on a single computer, there are no issues of hardware differences giving any of the competitors the edge. http://home.hccnet.nl/leo.dijksman/index.html is the link to the website with the information, select the 1st Division to see the crosstable and you can also download all the games.

Scid 3.0 is available. I haven't tried it yet, but will soon.

posted by matt josephson at 8:23 AM

Now using Yace 0.99.61 for Linux. This version looks to have fixed the bug that caused problems when opponents resigned while it was yaceComp's move. In such a case, Yace had been missing the force command sent from Xboard, and would continue to think about the game. When that happened, Yace would stop playing new games because it wasn't ready for a new game.

Monsters Inc. is a cool movie. See it if you get a chance.
posted by matt josephson at 9:00 AM

Evidence that Yace is improving?

Blitz rating history on USCL:

July--2541
August--2538
Sept--2566
October--2561
November--2581

Standard ratings

July--2418
Sept--2440
October--2447
November--2468

There are a few things to keep in mind about this. I'll have to check when, but I did move Yace to another computer with a faster processor (720 versus 600) and more memory, and in general, I've probably improved the book sources for the books I've made.

But then, the rating pool of USCL has been shrinking, which would be a force that would tend to draw ratings back towards the USCL's Glicko center of 1720. A smaller pool makes it harder for really high ratings and really low ratings appear. So maybe we are seeing an improvement through two things: newer versions of Yace, the growing number of learned positions, and the book learning culling out lines that are bad for Yace to play.

So I feel, yes, we have seen an improvement in Yace's play.
posted by matt josephson at 7:58 AM

Hmm, time flies when one is dating, and when Civilization III gets released, and played, and played, and played.

Have made a new book for Yace that used games played by players with ELOs of 2450+. Along with games for the Halloween Attack. The lightning rating for yaceComp recently achieved a new high at 2719, in some part attributable to Yace's beating computer accounts on time in 1 0 or 2 0 games. Sometimes Yace builds up a sufficient time advantage in those games, and if there are a lot of moves, that time advantage is enough to run the other computer down on time. Might also be in part due to USCL's implementation of a minimum clock charge for every move.

YaceComp has played in 226 Halloween Attacks, though not all as white, as a couple of humans have had the chutzpah to play the fearsome attack with the white pieces.

Right now, for reasons unbeknownst, two Crafty accounts, Cheesemaster and SuperCrafty have noplayed yaceComp. Maybe it's because me, as Mihobbes on the server, gripe about all the Crafty accounts. Hey come on, there are other fine engines that could be used. Anmon, Nejmet, Amyan, Pepito, all would be interesting to see.
posted by matt josephson at 8:04 AM

SCID 2.7 is available for Windows now too. Very nice, and if you go to the site, http://scid.sourceforge.net, it is also now possible to download a database of almost 500,000 games.
posted by matt josephson at 12:27 AM

Highlights of yesterday:

Cheesemaster returned after a long absence. In three blitz games, Yace won 1, lost 1, and drew 1.
posted by matt josephson at 8:27 AM

Thoughts on the AdminBot:

The USCL server does have a distinction in that it tries to provide a more "family-friendly" environment than other chess servers. That's a good thing, since many of the USCL participants are younger than those who play on FICS or ICC. I've never had too much trouble with the Service Reps or the Admins working to maintain such an environment; however, the AdminBot has got to be one of the least well-conceived implementations of enforcement.

In case one is not familiar with the AdminBot, this is the gist of what it does: it monitors certain public channels for offensive language. Based upon the level of the word's offensiveness, a person can get a simple warning, or a warning plus being muzzled in channels for 10 minutes. There may also be even more severe sanctions for even more offensive words; I haven't been particularly inclined to probe those limits.

So one day someone says the word "idiot" and then mentions that he just got a warning from the Bot. Curiosity got me to try it, and sure enough, I got warned for possibly inappropriate language. Interesting I think. Strange too--what other words might get one in trouble? The next morning I asked in the help channel if there were some way to find out what words are off-limits. I was told that I couldn't be told of them because it would make it too easy to circumvent what the Bot is trying to do. So then I'm doing some more thinking--YaceComp uses zippytalks and sometimes quotes from literature. So I ask in the help channel could this be a problem? I am told that if my computer doesn't quote anything of a sexual or obscene nature, there wouldn't be a problem.

Most everyone has heard the quote "If you prick us, do we not bleed?" It's from Shakespeare's Merchant of Venice, part of an impassioned speech where one is pleading the case of being a human being as much as any other man. So I drop this quote in the General Chat Channel, and bang, I get warned and muzzled, because of the word prick. So now I see it is quite possible for one to find literary quotes, that have no sexual or obscene nature to them, that could inflict muzzling. And with no real way of knowing what words could do it. Though trial and error, along with other people's misforutnes, have shown these words to be dangerous: ass (this also happens to be another word for donkey), dick (I have two uncles whose names are Dick), Dickens (yes, that famous author), butt (we wouldn't want to talk about the butt ends of guns now).

Yowch. I'd suggest that the wordlist of Adminbot was a little bit expansive in its scope. Though I have been told that these errors are being fixed, but of course, one has no real way of knowing except through trying, and no one is willing to provide a list of what can't be said.

It was time to make some objections noted. In the course of that, I was told a couple of things. If you did get muzzled for something which wasn't really obscene, all you had to do was get an admin and they'd fix it. But yah, what if there is no admin around? Well you're just going to have to wait 10 minutes. That's not so bad, is it? Perhaps not, but still damn annoying. Besides, one of the reasons AdminBot has been put in place is to monitor things when there are no admins online. Wait a second--I was just told that I shouldn't have to be too annoyed because I can always get an admin to fix things, but that this bot exists for a purpose of adminning when there are no human admins. So on one hand I'm being told that it's not so bad because humans are still in control of it, but on the other hand that its purpose is to control humans when there are no humans around to control the bot.

Pfft. I have other objections as well. Most of us are willing to accept that there are rules in life and places. Generally I don't have trouble with them, though I will protest rules that I don't feel are right. My protest here is not with the rules of USCL, but the manner of administration. It is based upon these expectations I have with rules and their enforcement: that the rule is clearly-defined and their enforcement is mostly fair. AdminBot obscures the rule definition, since we can't be told what will break the rule. Perhaps we can make some good guesses as to what the rule defines, but we can't be certain. And no one is going to tell you--you can only find out through testing or misfortunate choice of words. As for the expectation of fairness, how can it possibly be fair for someone to be punished for something that did not break the rule of obscene or offensive language? This is the inherent problem of creating Bots to administer justice and monitor human behavior--the digital beast cannot comprehend context. And we're asking it to perform that function? Blech.

The final irony of this is that the AdminBot has ended up encouraging behavior that violates the rules. "Oh my goodness, I've just been warned by a Bot!" is the usual reaction to when someone receives a warning. And suddenly people are curious and wondering, just as I started wondering. With that, lots of conversations have suddenly been drawn to this issue--conversations that in the past would have never cropped up or passed quickly from view.

The AdminBot, for all its good intentions, is a bad and inhuman implementation.
posted by matt josephson at 11:41 AM