I joined chessgames as a non-sub to see what it was like.
It was getting pelters from another place and I wanted to see for myself.
Of course I got involved in an anti-computer rant.
Some lad bet me a years chessgames sub I could not show him,
as I had claimed, a 1400 RHP lad playing a move and winning where in the
same position Houdini would choose a different move and lose.
So I showed him Game 4867930 the White huiman played Qh3 and
mated Black - A computer would choose Qf3 and being a piece & pawn
it would lose.
The lad bought me a years sub! I told him not too but he went ahead.
So there I am all alone. (except for my benefactor) If any of you lot are
there then say hello on my private forum thingy.
I need to know who my friends are. I'm Sally Simpson.
There are some really big guns floating about on there and I just know I will be ruffling feathers.
I don't want to upset any of you RHP lads by mistake.
(you will see why 'Sally Simpson' in the profile.)
I am going to start a games collection but very little else.
My spare chess time is taken up on here and I aint leaving here.
Originally posted by greenpawn34Well yes, but if black swaps queens instead of taking the rook then he's a piece up.
I joined chessgames as a non-sub to see what it was like.
It was getting pelters from another place and I wanted to see for myself.
Of course I got involved in an anti-computer rant.
Some lad bet me a years chessgames sub I could not show him,
as I had claimed, a 1400 RHP lad playing a move and winning where in the
same position Houdini would ...[text shortened]... ction but very little else.
My spare chess time is taken up on here and I aint leaving here.
Hi Deep Thought.
The human lad played 7.Qh3 and won because Black took the Rook. (Qc8 mate)
The computer chooses 7.Qf3 because it expects after 7.Qh3 Qxh3 and
it wants a Knight on f3 instead of h3. Either way it matters not White is
a piece and pawn down and lost.
The 1400 player won a game the best computers would lose.
It is a very basic example of a computer failing to play chess.
It cannot recognise a basic simple trap in this six and two three's position.
It cannot play to expect a blunder. It cannot hope, it sulks.
Another from the 1400 RHP lads. (Black to play)
In the actual human game Black played 1...Kc7 2.Ka6? Kb8 0-1.
One of the computer experts tested this position.
Critter and Stockfish took the draw as top choice but Komodo liked 1...Kc7!.
(there is hope for this program.)
However when the test were run with Komodo's tablebase switched off it
Komo listed it's top three as 1...Kb8, 1...Kc7, 1...Kd7 (? that losses).
This is quite a serious error and I can only assume the lads who put this
thing together skipped endgame routines because 'why bother' the TB's
will kick in and we get 100% accuracy.
OK. but how does thing assess engamges before the TB's start up?
A Master game example. Bernstein - Tarrasch, St Petersburg 1914. (White to play)
Bernstein had to win this game so he could play in the second phase of this brilliant tournament.
White is lost but tried one last gamble 69.Rxb2+!.
If Tarrasch has played the clumsy, though very plausible....
(see Blog for 100's of examples of plausible moves losing on the spot.)
70.Kxb2 Be5 wins.
No computer would play 69.Rxb2+ even though it is the best practical chance.
They go for Rxa6 as it staves off the coming mate the longest.
They have no idea of what a blunder is, they do not have ideas,
they cannot play Chess.
Studying with one (at our level) at your peril.
It will drop you in a position laden with tricks and traps, some will be,
yes, will be, chess is very rich in tactics, 10-15 moves or even longer deep.
It will see them, but you won't. It will dismiss them, can you afford to v a Human?
We keep getting told how strong they are, all I am doing is pointing out
a weakness. 1400 players are winning games from positions these things would lose.
The work being done on them is soley to beat other computers, not to help
humans play chess. It can;t. it cannot think like a human.
(Tarrasch of course played 70.....axb7 0-1.)
Problem Tests
by Larry Kaufman
In the last issue, we reprinted the "One-Hour CCR test" and made the comment that we hoped to offer an endgame test to be averaged in with the "One-Hour" results for more accurate rating predictions, since the "One-Hour" test does not deal with the endgame. Such an endgame test, devised by Franz Maresch, has been published in both "Ply'' and "PC Schach" magazines, together with the results of a fair number of programs. His test consists of 35 problems, on which your computer is to spend 1O minutes each in infinite level, with one point awarded for each correct solution at the end of the ten minutes. In some cases there are more than one correct solutions, while in other cases no credit is awarded unless one or more follow-up moves are also found in ten minutes each. To convert the point score to an estimated USCF rating (for endgame play only), I suggest multiply ing by 30 and then adding 1650. This means that a perfect score would earn a 2700 rating (about U.S. Champion level) while a zero score would rate 1650. In my opinion this is the best endgame test for computers that I have seen so far, as it covers all major classes of simplified endgames and tests both knowledge of endgame theory and calculating ability. I would hope that some readers might try running the test using the scoring method Igave for the "One-Hour'' test, namely allowing only two minutes per problem and awarding one point for each correct move after 15", 30", 1', and 2', disregarding the follow-up moves since they will generally be found if the first move is found. This would allow for the points of the two tests to be combined directly to make a point score ranging from 0 to 240 for the sixty problem set, which we might call the "CCR three hour test". That score would then be multiplied by five and added to a constant to be deter mined to estimate ratings more accurately than any other test has been able to do. In the meantime, we'll have to be content to average the "One-Hour'' and "Maresch" rating estimates.
Since the "One-Hour'' test has come out, there have been results pouring in from the readers of all the magazines that have reprinted it, especially Eric Hallsworth's "Computer Chess New Sheet",
Dedicated Computer"One-Hour" "One-Hour'' " Maresch""Maresch" Average
TASCR30 (ver. 2.2)Points
63.5Rating"
2515Points
30"Rating"
2550"Rating"
2533
M Genius 6803065.52535(26)24302483
M Berlin Pro61.52495(25)24002448
CMach Gid3 30M6425202423702445
Meph Vane 680206024802524002440
M Montreaux572450(26)24302440
Saitek RISC2500522400(25)24002400
Meph Port 680205123902524002395
Meph Lyon6800045.523352524002368
Meph RISC1502380(23)23402360
Meph Vane 68000472350(24)23702360
Meph Berlin442320(24)23702345
Meph Almer680204122902423702330
Novag Sapphire3822602524002330
Meph Port 68000422300(23)23402320
Fid Elite 9(34)22202423702295
Novag Diablo412290************
Meph Almer68000302180(22)23102245
Fld Des. 2325292170(22)23102240
Novag Jade II/Zircon II322200************
Sal Pres/GK2100302180************
Meph Nigel Short302180************
Fld Mach Ill192070(20)22502160
Novag Ruby282160************
Meph Roma 68000272150************
Meph Dall 68000252130************
Meph Amsterdam242120************
43
Dedicated Computer"One-Hour" "One-Hour" "Maresch" "Maresch" Average
Points"Rating"Points"Rating""Rating"
Meph Academy
20
2080************
Meph Polgar192070************
Meph Supermon2192070************
MephMM52221001019502025
Flcl Excel 312.52005************
Flcl Elite C122000************
Sal Trav Ch old132010818901950
Exe Adv Star Ch91970618301900
PC Software"One-Hour'' "One-Hour'' "Maresch" "Maresch" Average
Points"Rating"Points"Rating""Rating"
Meph Genius 37526302824902560
Meph Genius 27426202824902555
Hlarcs 37226002824902545
CM40006425202824902505
TheKlngTascbase622500(28)24902495
MChessPro 4.058.524652824902478
MChessPro 3.15624402824902465
Hlarcs 2.158246027(2460)2460
Rebel 6.06024802624302455
WChess6024802524002440
Kallisto 1.82542420************
Gideon Pro5524302524002415
Fritz 2532410************
Fritz 36024802223102395
Socrates 3.05424202323402380
Nimzo462340(25)24002370
Fritz 1472350************
Rex2.3(32)2200************
GnuChess182060************
black to move problem 10
problem 11
black to move problem 12
problem 13
problem 14
problem 15
black to move problem 16
problem 17
problem 18
black to move problem 19
black to move problem 20
problem 21
problem 22
black to move problem 23
problem 24
black to move problem 25
problem 26
black to move problem 27
problem 28
black to move problem 29
problem 30
problem 31
problem 32
problem 33
black to move problem 34
problem 35
Originally posted by greenpawn34Hi gp,
Hi Deep Thought.
[fen]rn2kbnr/ppp1pppp/8/8/3P4/2PQ4/PP3PqP/R1B1K1NR w KQkq - 0 7[/fen]
The human lad played 7.Qh3 and won because Black took the Rook. (Qc8 mate)
The computer chooses 7.Qf3 because it expects after 7.Qh3 Qxh3 and
it wants a Knight on f3 instead of h3. Either way it matters not White is
a piece and pawn down and lost.
The 1400 ...[text shortened]... y chess. It can;t. it cannot think like a human.
(Tarrasch of course played 70.....axb7 0-1.)
all except the earliest chess machines use a search routine where they select the move whose consequences maximizes the score for each player up to a move horizon, and you are entirely right - they cannot play for a trap except possibly as a consequence of a move horizon effect where if the player falls for it the pay-off is within the machine's move horizon and if the player doesn't fall for it the penalty for the machine is outside its move horizon.
Komo listed it's top three as 1...Kb8, 1...Kc7, 1...Kd7 (? that losses).
That's interesting. It's move list should be 1. ... Kc7, 1. ... Kb8, 1. ... Kxb7, 1. ... Kd7 and 1. ... Kd8 the first three of which draw. The only explanation I can hazard a guess at is that they've suppressed captures in the move list to avoid one type of machine idiocy, so the move list is wrong. It'd be interesting to see the algorithm.
I think you are right, but overstating it. Computers make tediously accurate opponents, or if set to a beatable level, just play stereotypic poor chess and don't make the kinds of errors humans do. On the other hand I find Crafty useful to chew over OTB games and point out tactical opportunities I've missed, which I sometimes learn lessons from. I learned K+B+N vs K by playing it out against a table base, and they will play all EGTB positions without result changing errors, but I wouldn't take strategic advice off one, and as you say - they miss all potential traps.
Hi Deep Thought.
" I learned K+B+N vs K by playing it out against a table base..."
Recently I was reading a post on another site from a good player who says
he can mate a computer with KBN - K very easily but doing it as an excercise
v a human for two games and he found he was having to stop to re-think his strategy.
The human did not play the 'best' moves. Those that lead to the longest mate.
Instead the human played natural moves, which in theory should have been mated
quicker but the lad said though he won both times it was 'different' and 'tougher'.
I never made a comment on it but I'm thinking:
He was seeing the patterns 3d, possibly for the first time and he toiled
OTB to drag them down.
Plus sometimes the best move to play is not the best move to play.
Something a computer knows nothing about.
Is it now pointless to study won Rook endings with a top box to brush
up on your technique, say KRP v KR because suddenly the thing can see
it's going to get mated with best play, it sulks and plays only moves to prolong
the game and stave off the coming mate rather than put down all the
human tricks and obstacles one would face OTB.
These things are now too strong to be of any practical use.
You need to dumb them down to a human level.