White to Move. Rybka couldn't find Qc8! Best it can come up with is Qc5+
Best continuation according to D. Monokroussos is as follows: 1. Qc8 Kg8 2. Bc7! Qxc8 3. gxf7 Kh8
4. Be5 Qc5! 5. Bb2! Nc7 6.Ba1 a4 7.Bb2 a3
8...a2 9.Bb2 a1Q 10.Bxa1.
it just seems to be an anti-computer position, as far as i can tell, based on current pruning algorithims or whatever (i only pretend to know wtf those words mean, i just seem there here a fair amount :).)
This is both worrying and reassuriing: engines are not perfect, so we can' t rely blindly on their analysis, on the other hand there is room not only for
optimizations (say algorithms tuning, better coding) but also for genuine innnovations derived from a more thorough understanding of chess mechanics.
I wouldn 't be surprised if Mathematics had a role to play, Graph Theory for instance.
Vas will most probably figure out why she overlooked ' 2 Bc7 ', although she necessarily evaluates all possible first moves (here ' 1 Qc8 ').
If Rybka were perfect he wouldn' t work full-time to improve her play. Here a weakness has been identified, this is paradoxically good news.
> Here a weakness has been identified
This position pops up about once a year here on the forum. Here is the 2007 version of the discussion:
Use search to find more.
As yanquis pointed it out, this position must be "engine-unfriendly".
/* Steinar */
Its safe to say that: Null move techniques are obviously not compatible with zugzwang and zugzwang is easier to see, and probably more common in the endgame.
As an experiment, it might be interesting to try a chess variant where passing is allowed, and see how often this would get invoked.
/* Steinar */
The bottom line is that current engines can't find zugzwang positions in a manner that doesn't hurt engine strength outside of a small number of endgame positions. You would like to extend this to say that these positions rarely occur outside the endgame. I can't see any reason to believe this is true...
> You would like to extend this to say that these positions rarely occur outside the endgame. I can't see any reason to believe this is true...
More pieces still on the board will mean lesser chance for zugzwang as there is probably some kind of 'non-interfering' move still possible. I don't have exact numbers, but I think it's a non-issue (certainly ELO-wise).
As an example, many fortress configurations fail because of zugzwang. Engines are notoriously poor at finding fortress configurations from either side, so this doesn't affect engine Elo. This certainly doesn't mean these positions are rare or unimportant Elo wise.
On the other side we have best human chess. Full of tactical errors, but with ideas, plans and real sacrifices (not above human horizont, but above computers horizont!) and very much chess knowledge (~600 Elo more than computers). What we see is that tactical skill crushes knowledge by far in chess (chess is a tactical game!) in normal play. But when we come to positions, where tactics isn´t dominant, it becomes different (mostly in endgames). Here knowledge pays off! But when you speak about endgames, never speak about Elo!
The worst thing for computer chess are opening books (and EGTB)! When we speak about engine skill, I want to see the whole picture! Opening, middlegame and endgame. So let´s play three move openings or chess960. When we play three moves openings (or 8-10 moves), we see the faults of the engines (look to CR). If we play endgames without tbs, we see the faults of the engines. If we are skilled, we see the faults from engines with tbs.
Errors are human! No mercy for tin cans!
Let´s come to our theme. Endgame without zugzwang detection isn´t endgame (I´m not interested here in % of escape)! Zugzwang in middlegame is a secret, we will never know. I only know, if you have no clue about endgame, you can´t play middlegame well! If you win anyhow, it only shows your opponent is a patzer!
1. To be #1 in computer chess, you have to polish the +0,02 or 0,03 attidute.
2. To be #1 in analysis, you have to add some knowledge. The danger: Look to F10 and F11. F11 (with less knowledge) isn´t good for analysis but 100 Elo better than F10, which is good for analysis!
But I´m not so sure, you can get both. I´m not interested in #1, but it´s very important for Vas. And never forget: Rybka succeeds in 2005 and 2006 only because of #1.
In testsuites (mostly human games positions) you can see, Fritz 10 and Shredder 10 are much better than Fritz 11 and Shredder 11. They improved their playing strengths (both ~+100 Elo) by discarding knowledge!
What we see now is a trend. Search to more depth is successfull. I don´t believe this trend will survive the next three years! You have to become broader and you have to add knowledge (which doesn´t hurt!).
Maybe Fritz 10 and Shredder 10 will be reanimated on Nehalem EX (32 or 64 cores). Okay you have to improve these engines in scaling! But I see no reason, why search is better for scaling than knowledge (okay; this is the area of the programmers).
These tournaments are nothing for miser!
Only for you the serious analysing engines :-): Fritz 10, Shredder 10 (and maybe 11), Zappa and maybe Rybka (here I´m not so sure; I´m sure she bluffs me sometimes! :-().
PS: I don´t have Hiarcs and Sjeng!
PS2: I payed €50 for single Fritz 10, Deep Fritz 10 costs €120 when it was released. If you have Nehalem (4 or 8 cores) it´s a bargain offer! :-)
But I become angry! :-(
All was okay before Fruit and then Rybka comes out. Searching for best moves was trend (and little improvement), but Fruit and Rybka shows, this isn´t the right approach for playing strength! You have to go in the deep and you will see the nuggets! But I don´t believe, there are nuggets in 3,000 meters (depths=30). And you can see it in CR play. She is digging in 2000 meters (depths=20). But here she sees all, also winning moves! :-)
Now we have only one best engine and the other goes the same way and become stronger (but not enough) and weaker at the same time. It´s a disaster!
PS: I don´t think CB is much inteested in money because ot the Fritz engine. They make their money with smiling Vishy and others ...
I do not have Fritz10 and Shredder10 so I cannot compare them in test suites to Fritz11 and Shredder11 but even if
Fritz10 and Shredder10 are better in test suites it does not mean that the programmer improved the playing strength by discarding knowledge.
It is possible that adding knowledge is the reason that they got worse in test suites when the knowledge that they added is productive in games and counter productive in test suites.
Test suites are not about random human games positions but usually about positions when some sacrifice is the best.
There are many human game positions when sacrifice is not the best move and it is possible that shredder11 and Fritz11 find better moves in these positions.
> No ideas, no plans and no real sacrifices!
You may like to search for Rybka 3 Vs. other (non Rybka) engines, I recall these were full of sacrifices that the other engines weren't expecting.
This isn´t the story I tell about!
But Thinker is bullshit! An engine, which can´t give a line and doesn´t follow UCI protocol!
Okay, if they get such successfull as Bobby in 1970, I will have a look on this engine!
What I have seen are only sacrifice R vs B or N and maybe additional a pawn. This is no sacrifice! Okay, eval is always alarmed!
And Rybka overvalue Queen. No surprise, because she has no clue about endgame!
You don´t understand me!
PS: Have a look to the games of the great Petrosian!
Roland made it sound as if humans play real sacrifices all the time while Rybka doesn't, "go watch Petrosian games" or "Fischer's Bg4" is unsatisfactory since he didn't went to look for real sacrifices on Rybka games.
I claim that Rybka plays sacrifices where humans aren't able to see why they work immediately on a regular basis (with the disadvantage that Rybka doesn't play them against them, because Rybka doesn't play strong humans much or games happen in private.)
It is my belief that all engines designers have, as for now, overlooked some fundamental but
relatively simple properties of chess mechanics.
There is room for breakthrough innovations. Who first ?
> There is room for breakthrough innovations. Who first ?
Well, the breakthrough was the null-move invention. This greatly increased engine playing strengths, the (very) small drawback was it would fail in zugzwang situations. But those are so rare compared to normal situations there is no need to fix anything.
I do not consider null-move (rationale) as a progress in the understanding of chess mechanics, let-alone a breakthrough.
But rather as a very efficient algorithmic trick.
In other words, they know about combinatorics, not about geometry or topology. Having no intution of shapes and fluids, they lack a dimension
in their "reasoning".
>It's almost always a win in real play, ...
Hahaha! Very good joke! :-)
my rybka has enormous problems to find mate in six here, maybe it's a matter of engine settings i don't know
(the solution is 1.Rh1 d4 2.Ra1 d3 3.Ra3 bxa3 4.Ka1 a2 5.Ba3 Kxf6 6.Bb2)
i know there are special engines for solving this sort of problems, but shouldn't rybka find mate in six relatively quickly in a legal chess position?
> i have a new one:
No, you don't, it's old.
i've found this position in chessmaster X, sorry but can't remember the author, sure people here know the name.
- moving the pawn away from d4
- capturing the rook at a3 with the pawn
- capturing the rook at f6 with the king
Granted, mate in six is pretty extreme. Doesn't it find it at all? I'd assume it would if you just gave it some time...
/* Steinar */
Rybka is not designed to find the fastest mate but to win games.
Rybka finds alternative win so she does not care about the fastest mate.
Rybka suggests 1.Nd6 cxd6 2.Rxd6 f6 3.Bxf6+ Kf7 4.Bxh8 Rxh8 5.R5xh6 Rxh6 6.Rxh6 and white is winning.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill