Rybka is currently without a doubt the strongest chess engine in the world and I admit that even though I was getting handicap in most of games I was quite pessimistic before the match as I had no experience of playing computers in the past. Add to this an octal computer, opening book, equal amount of time for the game and last but not least that I had to play 8 games in 5 days against someone who doesn`t get tired and who doesn`t forgive you any mistake and you`ll probably understand why I was not such a big optimist when starting the match. And something else I`d like to add here. Chess programs nowadays are so strong that supporting an engine in the match vs human is like backing a car in its competition vs runner. I think we can all admire the strong program and give our credit to people who created it but I have to say that those ones who back the engine in such a match have a little understanding from me.
And now to the games. In the first game I thought that I started quite well but then overrestimated my position after Rybka played very unusual Nd7-b8. I sacrificed a pawn and although it probably was not a mistake by itself, such a way of playing is unadvisible vs. computer. For some time I was able to hold the balance but in the end Rybka won in a very convincing manner. Could I imagine at that time that I was not going to lose a single game anymore?
In the 2nd game I had an exchange up but still came very close to losing. I think that if Rybka hadn`t played forcing c3-c4 but just kept pressure on the kingside, things could become very difficult for me. In the game Rybka won some material but I managed to activate my pieces and the game ended in a draw.
In the 3rd I didn`t get much from the opening but I wasn`t afraid of Rybka so much that I wanted to keep the position closed at any cost and decided to open it up by playing e3-e4. Later on the endgame was equal but then I relaxed and allowed Rybka to occupy the d-file after what the things became unpleasant again. Probably Rybka could play better somewhere, maybe winning the pawn on b3 was not the best choice but anyhow I held the draw.
In the 4th game for the first time I came close to winning. I managed to develop an initiative on the queenside without any real counterplay for Rybka but then in the time trouble I missed Qd2-h6! and especially Nf6-e8!, and had to force the draw. As Larry told me, Rybka saw quite an easy win for me but again, what easy for Rybka doesn`t have to be necessarily easy for the human!
In the 5th game Rybka played without the f7-pawn. Though Larry did a good job preparing Rybka`s opening book, I still feel somewhat embarassed that I wasn`t able to win having such a big advantage. As I`m sure you all noticed, in the beginning I had difficulties getting used to Rybka`s play and f7-pawn handicap was not the exception. After I made the mistake early in the opening playing Ne4-f6, the game ended with perpetual check.
By this time we played 5 games and I think Rybka started to get tired. In game 6 Rybka played 1.c4 as she realised that she was having a lot of problems in the game 4. However, we got a kind of Queens gambit semi-open position where my exchange started to play a big role. My play was not perfect and may be Rybka could make a draw somewhere but I think in the opposite colour bishop ending I played quite well bringing my king over from queenside to g4 and back to the queenside.
I was happy with my first win and wanted to win again but didn`t expect that I would win so easily in game 7. Hard to believe that the strongest program in the world played nearly the whole game like a good old Fritz2. Absense of the f7 pawn is only a small excuse as the computer had enough chances to open up the position early in the opening.
And in the last game my only goal was to make a draw and that`s what happenned in the end.
My first conclusions, just after the match:
As I said before, Rybka is very strong and is the strongest engine in the world. She is strong not only in the open positions as some might think but also in defence and that`s probably the reason why it`s so hard to win against her even having f7-pawn handicap.
However, I`m convinced that the program is not unbeatable (when talking about playing normal chess without handicap). In the match we played now the question basically was whether I can convert my material advantage or not and in a way it became a little boring towards the end. I personally would be interested to play a match against this program where the only handicap I`m asking for is that I get White in all games. I`m confident that with the right preparation I`ll be able to represent the human side as well as I did in this match.
And what is it going to take for us to see these games? Do not be bashful... if you have a figure in your head let us know.
> In the match we played now the question basically was whether I can convert my material advantage or not and in a way it became a little boring towards the end.
What kind of adjustments you made in the final games to beat Rybka? Did you make some adjustments to Rybka's play consciously, or did you just "got used" to Rybka's play in some unconscious way?
I followed all games but two from the first to the last move live and it was a great pleasure to be able to. Of course I was behind you all the way; in a match like this I don't understand people backing the engine either ...
I must say though, that you really tortured us in game seven by waiting so long, when you could have ended it twenty or thirty moves earlier!
If the white only handicap match ever comes about, be sure to find me amongst the audience again!
looking forward to see you playing Rybka with no other advantage than playing white side.
first of all: congratulations to your well earned victory over the strongest chess program on earth!
It was a great pleasure for me to follow the match live on internet.
Despite odds it must have been very stressful and exhausting for you to be aware not to make the slightest mistake, which surely immediately would have been punished by Rybka.
Before this match, I was convinced that it would not be possible for a human player to win this match - but surprisingly and fortunately you disabused me (and many others)!
Thanks a lot for giving us a personal insight of your view and feelings along the games.
In the 4th game I hoped you will find a win in your advanced position, but as you said "Rybka saw quite an easy win for me but again, what easy for Rybka doesn`t have to be necessarily easy for the human!"
In 5th game I can imagine that you felt "somewhat embarressed that I wasn`t able to win having such a big advantage" ;-)
Did you ever consider to play 7.Be2, Bf4 or even 7.Nxe5?!
Regarding 7.Be2 I run a computer match, which I mentioned to Larry in the following post:
I think 7.Nxe5 would have been awarded and cited all over the world!
Again thanks for this great performance and hopefully we will see you soon in another well organized, attractive and - last but not least - well sponsored match!
(BTW: I am willing to pay for internet live-coverage, somewhat at about 5 $ a game or 25$ an 8 game match)
I considered 7.Bf4 and even 7.Be2 but didn`t pay much attention to it as my king would have to recapture black bishop on e2 and I wanted to avoid that.
1. It is unfortunate (for us, not for you!) that Rybka did not strive harder to open the games. Particularly this was shown in game 7, first when Rybka failed to play ...c5 instead of ..Nc6, then when she played ..c4?, and then when she waited far too long for counterplay with ...b5 and ...a5. I can see why you might think that you could have chances with no advantage other than White pieces. However, I must warn you: it is quite simple for us to add code that will give Rybka a strong incentive to open the game when told she is playing a human. I very much regret that we did not do this either in Rybka 3 or at least for this match. Rybka 3 already has some incentive to avoid blocked positions in normal chess, enough that it seems that blockading strategies no longer suffice to draw in the hands of amateur players, but this incentive is cancelled out by a material handicap and so Rybka seems to be unaware of the need to open the game in the handicap games. We must fix this before playing further matches with humans.
2. Regarding the Exchange handicap games, the handicap proved to be a bit larger than I expected. One reason for this is that White cannot castle long but Black can. This amounts to a double handicap, material and positional. In some games long ago at rook odds White was allowed to play Kc1 in one move, but this is not an option for us as the normal program will consider such a move to be illegal. You made good use of the queenside castling option in two of the four Exchange handicap games. So I think that in any future matches involving Exchange handicap, it's only fair to say that Black also cannot castle queenside, as if he already had moved the queen's rook. Another issue is that the undefended pawn on a2 is yet a further problem for White in many openings, but I'm reluctant to put it on a3 (as some players do when giving rook odds) as then it constitutes an advantage.
3. You made the comment to me that the Exchange handicap games were in general of more interest than Pawn and move, because at least Rybka can play actively even though down in material. I must say I agree with this view. Also the variety of openings possible is much greater at Exchange handicap than at pawn and move, as the weak Black king severely limits Black's options. I think that the Exchange handicap will likely be a main battleground in the future for Rybka against ordinary GMs, even if it's a bit much to give to an Elite GM like yourself.
4. I have already made one change to the Rybka eval based on the match (mainly game 7) that is testing as a 1 Elo point gain. So at least the match did help Rybka's overall play a tiny bit.
5. I'd like to mention another possibility for future matches (whether with you or another GM). In Japanese they call it "sashi-komi", something like "changing pieces". The idea is this: the first game of the match is played normally, with the GM just getting White. If he wins he must play Black the next game, but if he loses he must take some small handicap (maybe two moves or White c2 pawn for example). If he then wins it's back to even game with GM White; if he loses the handicap game the next game handicap is increased. In all cases a draw means no change for the next game. So if the GM is really able to compete on even terms with only White and to win with a handicap, most or all of the games will be normal games. But if he cannot compete on even terms with just White and can't win with a small handicap, the match would be mostly a handicap match. This format could be used against any strong player; the handicap (or lack of handicap) is self-adjusting. In theory if GM wins twice at the start he might even have to give Rybka a handicap, though I think the chance of this happening is too remote to consider. Anyway this format insures variety.
I hope we get sponsorship for a second match, regardless of the format. Thanks for playing the match and congratulations from the Rybka team on your victory!
> enough that it seems that blockading strategies no longer suffice to draw in the hands of amateur players
They still do, though the amateurs have great problems, Pablo often posts his successful draws against Rybka 3 on CCC.
I quite like the idea of point 5. I'd call them dynamic handicaps! Does the GM want a normal game? He should earn it! Though I'd like if the next level handicaps weren't as drastic, such as that losing the first game gives you time handicap, and losing the second gives you draw odds, then draw odds+time handicap, two moves or material handicap, that seem more drastic and make fans unhappy should come later.
> Do you know whether Pablo got draws with R3 using default contempt, or with a reasonably high value set?
It is default settings so Contempt was 15. The thing is that people isn't going to change the parameters just to play against Pablo, and changing parameters is akin to using ChessTiger's Anti-Human setting that already is invulnerable to Pablo tactics. So maybe Rybka is capable of making Pablo fail every time with a higher contempt but people is interested in a program that does it on the default settings (as in a Rybka that becomes aware of the stonewall and changes contempt automatically.) That's yet to be seen.
> The steps should be clearly increasing with each loss, not just different.
But we don't know the value of certain handicaps so it's valid to change from a unclear handicap to another, as long as it's on the same range so the steps are clearly increasing, at least until we can figure out the values of the handicap. Anyway, I'd like to see as much normal games as possible , so, instead of mixing draw odds (that I agree it doesn't work), what about 3 move book, then time handicap, then book+time handicap, then enter material odds? That would guarantee that at least 4 games look normal, those are the ones I'm most interested in. I wonder of other normal game handicaps can be mixed.
As for handicap steps, we could logically do White plus shortened book as the first step or time odds, and then both together, but I think it's not logical to use short book and time as separate steps, as which is larger is unclear and depends on the magnitude of each. There are many other options that stay within the realm of standard chess, such as requiring Black to open 1...h6 for example or stipulating some unsound gambit like 1e4 d5 2exd5 c6?. Note also that two move handicap is equivalent to requiring White to open with 1Nf3 and 2Ng1 against any defense, so this is also a valid handicap that does not change the starting position or force the human to make any particular moves. Same with three moves (Black must open 1...Nc6 and 2...Ng8 against anything).
> There are many other options that stay within the realm of standard chess, such as requiring Black to open 1...h6 for example or stipulating some unsound gambit like 1e4 d5 2exd5 c6?.
Oh yeah, I've been a great supporter of the unsound gambit handicap, I thought that would never be considered as I've never seen it happen and watched the pawn handicap over and over. Forcing 1 move feels natural, since I've played amateurs that may answer h6?! against me, and the game feels normal. For a two forcing moves I'd prefer to see h6+a6 played than Nc6+Nb8 because the pawn moves at least seem to make some sense "You can't put your bishops on b5/g5; I can chase you away with these moves if you play Bc4/Bf4, etc.", but in the second case it's just like adding "pass" as a rule, as there's no difference (just like creating a game in where black loses a pawn and call it a thematic opening instead of pawn and move, how we got there doesn't change where we are.) Though I'm aware we'll inevitable get to these weird looking games as Rybka will play better than the GM.
> My guess is that it would be about 200 ELO weaker if it always tries to go for open positions.
lkaufman said the other day that changing Rybka's playing style had a loss of 30 ELO, I wouldn't call that dramatic (I'm aware you didn't.)
More important is the use of a 1CPU 32bit for matches, I'd like to know if the Rybka at my PC can beat a GM at normal chess! But it seems 1CPU is now rare and 32bit is coming next.
> They already sell notebook computers with 2 cpus and 64 bit Vista in the local store here for $675, so I think there is no point any more in trying to demonstrate performance on 1 cpu or 32 bit systems.
Hi larry would u please give details of this package.
Are they second hand or new ?
I want to buy one.
I have analysis the games with my "simple" Rybka3. (some games also in real time).
I have gotten all, I underline ALL, the movements played by the special version of Rybka for the match with Milov.
I am a little sceptic on these versions "anti-human" of engines.
And I can show this.
A few days to find the time to do one "compare analysis", and I will write the file in this forum.
My implementation requiring Rybka's evaluation were unnecessarily overcomplicated, though.
We called it "Progressive Odds".
-even, draw for color
-pawn and move
-pawn and 2 moves
-knight (odds giver gets white)
-rook (odds giver gets white)
-queen for knight (odds giver gets white)
I don't recall anyone being weak enough to need more odds. The rook pawns always started in their ordinary position (no a3 at the beginning).
by the way, I also would like to add that the exchange sacrifice games were much more natural and normal-chess like than the pawn and move handicap. the pawn and move dictates the strategy of the game too much, I suppose, whereas with the exchange, there are so many options for the defender.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill