Author: Robert Hyatt
Date: 10:39:25 11/06/00
Go up one level in this thread
On November 05, 2000 at 12:45:23, Mogens Larsen wrote:
>On November 05, 2000 at 11:28:58, Thorsten Czub wrote:
>
>>fritz, junior, shredder, ... do not see anything.
>>therefore they will lose against gambit-tiger.
>>they follow a paradigm that will die out.
>>
>>"Where do YOU want to go today ?" :-))
>
>It wouldn't be too difficult to find a position where the absolute opposite is
>true, ie. that GT is completely clueless about the position. So it doesn't prove
>anything either way. With a lot of effort almost any program could be
>responsible for a new paradigm based on a single position.
>
>If you consider the limited differences between GT and other programs then
>there's nothing to support a new paradigm. At least nothing remotely
>revolutionary or different. The approach is interesting, but not exactly a novel
>idea. I prefer knowledge programs myself, but I have nothing against the
>so-called "bean counters". They're not mutually exclusive.
>
>Given the information posted by CT your opinions about GT's revolutionary
>concept simply isn't true. But I know about your allergic reaction to facts :o).
>
>Mogens.
Here is a case in point. I don't like to critize any program, because they
all represent a lot of work. But here is an example of simply _not_
understanding a simple endgame feature:
[Event "ICC 20 3"]
[Site "Internet Chess Club"]
[Date "2000.11.04"]
[Round "-"]
[White "SubtleOne"]
[Black "crafty"]
[Result "0-1"]
[ICCResult "White resigns"]
[WhiteElo "2586"]
[BlackElo "2710"]
[Opening "Nimzo-Indian: 4.e3, main line"]
[ECO "E59"]
[NIC "NI.01"]
[Time "21:15:50"]
[TimeControl "1200+3"]
1. d4 Nf6 2. c4 e6 3. Nc3 Bb4 4. e3 O-O 5. Bd3 d5 6. Nf3 c5 7. O-O Nc6 8. a3
Bxc3 9. bxc3 dxc4 10. Bxc4 Qc7 11. Bd3 e5 12. Qc2 Re8 13. Nxe5 Nxe5 14. dxe5
Qxe5 15. f3 Bd7 16. a4 Rad8 17. Ra3 Re6 18. Bb5 Bxb5 19. axb5 a6 20. bxa6
bxa6 21. e4 c4 22. Be3 Rc6 23. Bd4 Qe6 24. Rb1 Ne8 25. Qc1 f6 26. Rb4 Nc7
27. Raa4 Nb5 28. Qa1 h6 29. Qb2 f5 30. e5 Qd5 31. Ra5 Qd7 32. Rba4 Nxd4 33.
cxd4 Qa7 34. g3 Rxd4 35. Kg2 c3 36. Qb3+ Kh7 37. Rxd4 Qxd4 38. Qc2 Kg8 39.
Qb3+ Qc4 40. Qxc4+ Rxc4 41. Ra1 c2 42. Rc1 a5 43. Kf2 Rc5 44. Ke3 Rxe5+ 45.
Kd3 Rd5+ 46. Kxc2 Rc5+ 47. Kd1 Rxc1+ 48. Kxc1 Kf7 49. Kb2 Kf6 50. f4 Kg6 51.
Kb3 a4+ 52. Kxa4 {White resigns} 0-1
If you play thru this game, by move 20 crafty has a passed pawn on the a-file,
which will probably win the game if you simply remove all the pieces right
then. Crafty understands this reasonably well, and wants to trade as every
trade makes that distant passer more valuable. Scan forward and watch white
not only allow trades, but encourage them. And watch it win the pawn on move
48 and _instantly_ lose the game outright. At the point where GT took the
pawn, its eval was -1.0. Crafty was +4. Which was right, here?
That was my point. Or one of my points:
1. A program can correctly evaluate position "X" as good for white, and
be right. It can also evaluate positions Y and Z as good for white and be
dead wrong. Thorsten chooses positions like X, where the program evaluates
it as winning and it is right.
2. take the position above. Crafty didn't expect the pawn to be taken,
because the rook check forces the removal of the last piece for each side,
and it _knows_ that the distant passer is a simple win at that point. It
was right there. Is it right in every case? No. But it is right in
_most_ of the cases dealing with distant passers. Does that then mean that
GT is not worth buying? Of course not. As it has its _own_ strong points where
it does well. Does that mean that Crafty is better? Of course not. It means
that in many endgame positions crafty is better. But in some it will be worse.
And perhaps in kingside attacks it might be worse (This I haven't seen yet as
I haven't been attacked violently and lost badly in any game I have watched).
3. For _any_ program, you can find positions where its positional judgement is
perfect. And you can find positions where it is far off. One of my favorite
program shortcomings is that they will often go for a position like this:
[D]4k/1p4p/////3PP/4K/ w - -
And think white is better. White has two connected passers, black has two
weak isolated passers. Which side would you want here? Crafty says black
is +1.7 with a pure static evaluation, no search at all. This bit of
knowledge was added after a lot of griping by a GM friend of mine. So here
is a case where Crafty is quite smart. I can also show you cases (non-endgame)
where it is quite stupid, although in many of those, the search comes to the
rescue and it still survives its bad judgement (ie junior crafty in the just
finished CCT2 event).
4. I don't call my endgame knowledge a "new paradigm". I call it a steady
evolution of a chess engine, just like we have seen for 25 years now. I don't
call any knowledge that attempts to encourage kingside attacks "a new paradigm"
either. It is just a steady evolution of that _part_ of the evaluation. And
there are obvious counter-measures for this, namely making the opponent know
enough about king safety that it becomes unlikely that a speculative attack will
work.
I'm not "anti-tiger". It certainly plays well. I am most definitely "anti-
hyperbole" which we get a lot of here. Tiger may well be far better than my
program. Using my quad, I'm not getting drubbed off the board yet. But I
might on equal hardware, I don't know. But I am not seeing something "bold and
new" yet.
Thorsten loves CSTal. There is nothing wrong with that. There is a _lot_
wrong with lumping Gandalf and GT into that same mold however. Because that
is based on opinion and not on fact. I doubt either of those two programmers
would say they were trying to implement ideas put forth by CW. Because CW
never put forth any "ideas" in any concrete form. Just lots of smoke and
mirrors and complaints about "I reach won positions but I can't win them because
I am too slow."
Won positions don't count for much in chess, unless they are actually won OTB.
IE I would rather play ugly chess and win every game, rather than playing
beautiful chess and losing every game. People used to put Deep Thought down
in the same way. Yet it won. and won. and won.
The purist might not like that. The pragmatist is satisfied with the result.
There is room for both. But the pragmatist is most likely going to win more
games. IE "beauty is in the eye of the beholder".
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.