< Earlier Kibitzing · PAGE 2 OF 3 ·
|Oct-19-04|| ||bunti: where can one download this analysis? |
|Oct-19-04|| ||clocked: It is good that <AJ> finally admits it is drawn, but he has NOT shown that Mig's line does NOT draw as well.|
<bunti> check out http://www.chessbase.com/games/2004... and http://www.angelfire.com/games3/AJs...
|Oct-19-04|| ||bunti: <clocker> Thanks. Because I do not know a lot about computers and especially after that incident with allegations of computer viruses between benjamin lau and Life Master AJ, I am unwilling to download from unknown sources. Based on the wording of AJ's latest post, I am unable to determine if he is saying that he simply found another line that draws, or if he is saying that MG's line is indeed not a draw. If anyone has downloaded AJ's analysis and has the time to do so, could they post some of it here if they are not worried about breaking any copyright laws. |
|Oct-19-04|| ||clocked: <bunti> <AJ> has done a book's worth of analysis, a summary would be a disservice. I would recommend just looking at the chessbase link. |
|Oct-19-04|| ||Dionyseus: I just checked Mig's line with Shredder 8 and in just a few seconds it was clear that it's a draw, and not just that, but that there were two draws just like Mig hinted at. 43.d6+ Kd7 44.Kb6 Rf5 45.Rb4 g5 46.Kxb7. Here, Mig says 46...Rf4 draws, and he's right.
47.Rb3 Kxd6 48.Re3 g4 49.b6 Kd5 50.Ka7 Rf3 51.Re2 Ra3+ 52.Kb8 Rb3 53.b7 g3 54.Kc8 Rc3+ 55.Kd7 Rb3 Clearly nothing can be done to progress. Draw |
Mig also said that 46...Kxd6 might also draw, and he's right again. 46...Kxd6 47.b6 Rf7+ 48.Ka6 Ke5 49.Rc4 Rg7 50.Rg4 Rg6 51.Kb5 Kf5 52.Rg1 Rg7 53.Ka6 g4 54.Ra1 g3 55.Rg1 Ke4 56.Rg2 Kf3 57.Rxg3+ Kxg3 58.b7 Rg6+ 59.Ka7 Rg7 And no progress can be made, it is a dead draw.
Because my Shredder 8 was able to find the two drawing lines so quickly, my guess is that Life Master AJ did not install all the 5-piece endgame tablebases.
|Oct-20-04|| ||Dionyseus: Just downloaded Life Master AJ's annotation for this game, and I'm now absolutely certain he does not have all the 5 piece endgame tablebases installed. |
In this line 43.d6+ Kd7 44.Kb6 Rf5 45.Rb4 Kxd6 (my Shredder 8 says this is a bad move) 46.Kxb7 g5 47.b6. Here, AJ says that Ra5 is nearly forced, but that's not true, my Shredder 8 immediately sees that Ra5 loses. The correct response is 47...Rf7+! 48.Ka6 Ke5 49.Rc4 Rg7 50.Rc7 Rg6 51.Rc5+ Kf4 52.Rc4+ Kf5 53.Rc5+ Kf4 54.Rxg5 Kxg5 obvious draw.
Here's another indication that AJ does not have all the 5 piece endgame tablebases installed: He says that with the white king in b7, white pawn in b6, and white rook in g7, black rook on a1, and black king in c5, after nearly half an hour of analysis with Fritz 8, "this position is definitely won for white."
Yes, that is correct, but had you installed all the 5 piece endgame tablebases, your Fritz 8 would have told you that immediately, and it would have told you that it's a mate in 25 and that black's best defense is Kd6, and that black's worst defense is Ra7+. It would have also shown you how to mate with any of the other 18 black defenses.
|Oct-20-04|| ||LIFE Master AJ: <D> I have the tablebases ... the set on discs. I do NOT use them on a regular basis ... I am convinced they have dozens (if not hundreds) of errors. |
When I first got them, I checked them against the ECO of Endgames, I found like 50 mistakes. This was enough for me to put them on the shelf, and never use them.
One of them is a very famous endgame problem by BENKO - it should end in a draw, with correct play. (The tablebase gives it as a win for White.)
|Oct-20-04|| ||LIFE Master AJ: <Dionyseus> You probably believe that the TB's are flawless. |
|Oct-20-04|| ||clocked: HAHAHAHAHAHAHAHAHAH, the tablebase is not OPINION. BENKO is human! I would love to see ONE "mistake" |
|Oct-20-04|| ||LIFE Master AJ: Table-Bases should be a TOOL ... just like an analysis engine. They should AID analysis ... NOT replace it!!! |
Kramnik - or a member of his team - forgot this, and it nearly cost him the World's Championship - when Leko won game eight of their match!!!!!
|Oct-20-04|| ||acirce: What?! What had the 8th game to do with tablebases? And no, tablebases is more than just a "tool". They actually give you the objective truth about a position. You said you'd checked them against the ECO of endgames. Let's see one of those 50 mistakes. |
|Oct-20-04|| ||clocked: Look at http://www.angelfire.com/games4/lif... he finally admits that the tablebase solution "might" work! I don't think he understands what the tablebases are. |
|Oct-21-04|| ||Woodpusher: Life Master AJ: First of all, what the heck is the "ECO of Endgames"? The Encylopeadia of Chess Openings of Endgames? Huh? Anyhow, if you find discrepancies between your printed materials and tablebase, you should consider the possibility that the book is wrong and tablebase is right. |
|Oct-21-04|| ||Dionyseus: I'm 100% certain AJ has not installed the tablebases properly.
Tablebases are flawless, period. Having all 5 piece tablebases installed means that your computer contains flawless knowledge of all 5, 4, and 3 piece endings. Without even making a calculation, it will tell you every possible solution, win loss or draw, to a 5,4, and 3 piece endgame.
And that's not all, there are already some 6 piece endgame tablebases, but it is not yet complete. |
Looking at the "Goldsby Position," http://www.angelfire.com/games4/lif...
I see that AJ gives Qg6 as the absolute best move. Actually, Qg6 is the 13th best move! Qe6+, Qg1, Qf2, and Qe3 are just as good as Qg6, they all mate in 21 moves.
The absolute best moves are Qb4+ and Qc5+, both which mate in 18 moves.
Anyways, after 1.Qg6, AJ says Rd1 is the best move. Actually, Rd1 is the third best move and it allows a mate in 10. The best move is Kd8, which allows a mate in 20. btw, Kf8 is second best, allowing a mate in 19.
Anyways, let's continue with your line. 1.Qg6 Rd1 2.Qg5+ Kf8 3.Qf4+ Ke7
Here you say 4.Qb4+ Kd8 5.Qb3 is the best line. Well actually that's the second best line, and 4.Qg5+ Kf8 5.Qf4+ is just as good, both mate in 10. The best line mates in 8 and that is 4.Qh4+ Kf8 5.Ke6 Kg7 6.Qg4+ Kh6 7.Qxd1 Kg5 8.Qd4 Kg6 9.Qg4+ Kh6 10.Kf6 Kh7 11.Qg7#
|Oct-21-04|| ||Sneaky: <I see that AJ gives Qg6 as the absolute best move.> Well, if your criteria is "quicker is better" then I guess AJ is wrong, but perhaps AJ's move implies a method which leaves the inferior side with fewer defensive options. If we all carried around 7 gigabytes of RAM in our brains we'd be better off playing it like tablebase dictates but for practical purposes perhaps AJ's method is more clear and therefore "the best."|
The only "flaw" I have heard about tablebase is that there are some endgame studies it fails on because they were constructed to allow one (or both) of the players to have castling rights, and the tablebase programmers skipped the whole castling situation. But these are not serious positions, they are composed puzzles designed to fool the solver, so I don't really consider that to be a defect.
|Oct-21-04|| ||clocked: An early "flaw" was the exclusion of the en passant rule. |
|Oct-21-04|| ||LIFE Master AJ: So it is called the "Encyclopedia of Chess Endings." BFD
No, I suspect the TB's are flawed. Probably something to do with the algorithm that generates them. This is not an ignorant opinion, but the highly informed opinion of a Life Master. |
In endings of only 3-5 pieces, they are often correct. (I have spent hundreds of hours analyzing the endgame of "King-plus-Queen," versus "King-plus-Rook." (No pawns on the board.) In this ending the computer is right ... often finding a quicker way to win than the books do. (I have a position named after me in this ending ... I think I know what I am talking about here.)
But when the TB's find a longer way to win ... or miss simple (or not so simple) drawing lines ... like in the case of repeated tests using Pal Benko's world-famous compositions ... I begin to suspect a flaw somewhere.
The problem with you simple-minded fish is that you think the computers are infallible, even God-like. Nothing could be further from the truth. Try analyzing a deep game of strategy ... several games of Petrosian's come to mind. (I know of at least three where the evals don't matter ... and you could give the computer a month, and it will still not come up with the winning plan.)
Another example was the game that Leko drew just before he lost. (13?) ALL the computers were showing an eval of: "MINUS-slash-PLUS" or that is was a simple win for Black ... by some 2-4 points. In fact, we headed for some simple book draws. (The Philidor Position?)
Get a clue.
|Oct-21-04|| ||AgentRgent: Mr. Goldsby, I accept your assertion that some tablebases may be incorrect. However, i'm sorry to say that your arguments are simply lost among your unabashed arrogance and rude behavior. Stop taking yourself so seriously, and others might pick up the slack. Just some advice, take it or leave it. |
|Oct-21-04|| ||Woodpusher: AJ, I think you are comparing apples and oranges. Tablebase is only for positions with 5 or less pieces. There is work being done on a 6-piece tablebase but it's not going to be done for a long time. The computers all failed on the Kramnik-Leko ending because computers are in fact not very good at endgames. They were not using tablebase to find the result of that ending because there were more than 5 pieces on the board! That doesn't mean that tablebase is flawed. Tablebase is not even a computer, it's just a table. You could print it in book form if you wanted to kill a forest.|
It could be that there is a bug in the algorithm which produced the 5-piece tablebase to begin with, if that's the case then I think it's your duty to inform the programmers.
Anyhow, the bottom line is, to prove that tablebase is flawed you need to provide a single instance where it produces an incorrect result. Unless you produce that then I have to assume you're just full of hot air.
|Oct-21-04|| ||LIFE Master AJ: <Woodpusher>
# 1.) What you want has already been documented on my website.
# 2.) Congrats! You have been added to the magic list.
|Oct-21-04|| ||acirce: <LIFE Master AJ> If you on your website have provided a concrete example where the tablebases are faulty, can you please give us the URL? |
|Oct-21-04|| ||Shadout Mapes: I finally found AJ's analysis, guys.
http://www.angelfire.com/games3/AJs... scroll down. the format is .cbv
Indeed, he provides no winning lines at the end on Mig's analysis, and gives a vague winning plan that doesn't sound exactly winning.
AJ, what would you recommend for white after 43.d6+ Kd7 44.Kb6 Rf5 45.Rb4 g5 46.Kxb7 Rf4 47.Rb1 (I see no differences between this and Rb3) 47...Kxd6 48.b6 g4 49.Rb5 (once again, name a better move. it looks logical, stop the g pawn from moving) 49...g3 50.Rg5 Rf3 51. now what do you suggest to win? give me moves, not "...Rg8... advancing his king to the 8th row, followed by putting the white pawn on the 7th rank." In that setup, how could white avoid a perpetual? The king is caught behind his pawn, the rook to the black pawn.
And I concur with ughaibu, Cyphelium is a great contributer, and friendly on top of that.
Looking forward to your response, Mr. Goldsby.
|Oct-21-04|| ||Woodpusher: Shadout, First, I followed the link you gave, searched up and down the page, and found nothing related to tablebase or endgames. Second, please simply define the position which Mr. Goldsby believes tablebase has incorrect, that way I can follow the analysis that you have reprinted and check against my copy of tablebase. |
|Oct-21-04|| ||Minor Piece Activity: It's in a Linares file I think. |
|Oct-21-04|| ||clocked: <shadout Mapes> You began writing that post 2 days ago!!!???|
<Woodpusher> the link is for THIS endgame. Goldsby has NOT provided an example of what you are requesting.
< Earlier Kibitzing · PAGE 2 OF 3 ·