tygxc wrote:
I have proven Sveshnikov right
It is great that you have solved chess. 
End of discussion, I suppose.
Chess will never be solved, here's why
Sort:
Oct 14, 2022
0
#4663
tygxc wrote:
@6118
"you should have used Syzygy"
++ I used Syzygy, but it does not matter.
Syzygy is just more compact than Nalimov, so it fits on a hard disk.
The core information is the same: draw / win / loss.
You've already proved you don't understand tablebases. There's no need to labour the point.
(I notice you're still chopping half of the sentences you respond to to distort the meaning.)
@6116
""Then generate a complete set of games.""
++ Here are 4 relevant KRPP vs. KRP draws.
In all of these the top 1 engine move was table base exact.
This sustains:
A table base exact move is among the top 4 moves of a 10^9 positions / s engine running for 17 s with 1 error in 10^20 positions.
This was extrapolated from the 10,000 + 1,000 AlphaZero games.
@6123
"It is great that you have solved chess"
++ I have not solved chess, I have shown that Chess can be weakly solved in 5 years.
I have shown in two ways that 10^17 positions are relevant to weakly solving Chess.
- top down starting from legal, sensible, reachable, relevant positions
- bottom up calculating an upper bound, a lower bound and an estimate.
3 cloud engines of 1 billion positions / s each reach 10^17 positions in 5 years.
10^9 positions/s/engine * 3 engines * 3600 s/h * 24 h/d * 365.25 d/a * 5 a = 4.7*10^17
Oct 14, 2022
0
#4666
tygxc wrote:
@6118
"you should have used Syzygy"
++ I used Syzygy, but it does not matter.
Syzygy is just more compact than Nalimov, so it fits on a hard disk.
The core information is the same: draw / win / loss.
Likewise, if you used Syzygy, you don't get results for a ruleset incompatible with it.
To be specific, a position can be a Nalimov win and a Syzygy draw (where the 50 move rule prevents a win).
@6129
"where the 50 move rule prevents a win"
++ The 50-moves rule plays no role in weakly solving chess.
A weak solution of Chess without the 50-moves rule is also a solution of Chess with the 50-moves rule. In none of the perfect games we have was the 50 moves rule invoked.
Oct 14, 2022
0
#4668
Also, engines are truly crappy at evaluating positions given 1 billionth of a second. (evaluating their legality alone in that time would be a challenge to specialised hardware like a 300 times faster version of Deep Blue 2's chips, and impossible with a state of the art general processor, never mind any sort of evaluation). You need to evaluate every single one of the positions reachable by a legal move in order to select your magic "four best moves".
Having spent your 40 billionths of a second to evaluate the 40 legal moves, you have not even a good idea of which is best. For that you need to evaluate them to a reasonable depth unless you want to be wrong very often.
@6131
"evaluating positions given 1 billionth of a second"
++ They evaluate 1 billion positions per second.
https://chessify.me/blog/nps-what-are-the-nodes-per-second-in-chess-engine-analysis
"You need to evaluate every single one of the positions reachable by a legal move in order to select your magic four best moves." ++ Yes, that takes 17 s.
"you need to evaluate them to a reasonable depth" ++ Yes: 17 s, that is 17 billion positions.
Oct 14, 2022
0
#4670
tygxc wrote:
@6131
"evaluating positions given 1 billionth of a second"
++ They evaluate 1 billion positions per second.
https://chessify.me/blog/nps-what-are-the-nodes-per-second-in-chess-engine-analysis
"You need to evaluate every single one of the positions reachable by a legal move in order to select your magic four best moves." ++ Yes, that takes 17 s.
That's 17 seconds per position seen by the opponent of a strategy.
Since half the (underestimated) 10^17 positions are seen by the opponent of a strategy, that is:
8.5 * 10^17 seconds, about 280 billion years.
Oct 14, 2022
0
#4671
tygxc wrote:
@6116
""Then generate a complete set of games.""
++ Here are 4 relevant KRPP vs. KRP draws.
Not what I'd call a complete set of games.
You don't say what think time was used. You've produced one game for each position - how do apply your calculation to determine the game theoretic value of each position from those?
In all of these the top 1 engine move was table base exact.
This sustains:
A table base exact move is among the top 4 moves of a 10^9 positions / s engine running for 17 s with 1 error in 10^20 positions.
This was extrapolated from the 10,000 + 1,000 AlphaZero games.
It doesn't contradict it.
You claim it's generally true, so you would need rather a lot of confirmatory examples to sustain the conjecture or even to lend it credibility.
On the other hand you need only one counterexample to discredit it, which I think you'll find among the games I spent so much time preparing for you (which are equally relevant to the correctness of your calculation).
You can settle the matter to your own and everyone else's satisfaction by applying your calculations to the examples I posted here.
I invite you (as you invited me here) to "Show it or shut it". (But this time the onus of proof is correctly placed.)
I'm still following.....fascinated....intrigued....in awe. Epic debate and I don't know who's right, if anyone....although I find myself leaning one way or another occasionally.
@6133
"that is: 8.5 * 10^17 seconds, about 280 billion years."
++ You are not good at arithmetic. There are only 10^17 positions relevant. The engine calculates a billion positions per second. That means 3 engines exhaust the 10^17 positions in 5 years.
Oct 14, 2022
0
#4674
I think @Elroch is probably finding that his arithmetic suffers a similar problem to my problem with Stockfish.
@6134
"how do apply your calculation to determine the game theoretic value"
++ You are confused.
The KRPP vs. KR drawn positions serve to check the thesis that the table base correct move is among the top 4 engine moves for the 10^9 positions/s engine running for 17 s with 1 error in 10^20 positions. As shown in these 6 examples the top 1 engine move is table base exact.
The game-theoretic value of a draw is determined from the 1469 ICCF WC games, that is the closest to perfect play we have: 1104 of those games are with optimal play from both sides.
The game-theoretic value of a draw also follows from a deductive argument. To win you need to queen a pawn, i.e. you need an advantage of 1 pawn or more. 3 tempi in the initial position are worth 1 pawn. 1 tempo in the initial position is not enough to win. You cannot queen a tempo.
"It doesn't contradict it."
++ Yes, that is right. You suggested some 7-men endgame might contradict it, but none did.
"You claim it's generally true" ++ Yes, for positions relevant to weakly solving Chess.
"you would need rather a lot of confirmatory examples to sustain the conjecture"
++ The thesis was extrapolated from the AlphaZero paper. That is what lends it credibility.
"you need only one counterexample to discredit it"
++ Yes: 'No number of experiments can prove me right, one can prove me wrong' - Einstein
"you'll find among the games I spent so much time preparing"
++ The 2 I posted did not contradict.
The KNN vs. KP is irrelevant: 5 men, not 7, not a draw, a known anomaly: an engine can only detect the checkmate if its calculation depth exceeds the distance to mate otherwise the engine is happy to win the pawn increasing its perceived advantage from +5 to +6, but spoiling the win.
Oct 14, 2022
0
#4676
Optimissed wrote:
tygxc wrote:
@6119
"Like Sveshnikov, Kasparov is 100% reliable."
++ Kasparov was proven wrong and I have proven Sveshnikov right:
Chess can be weakly solved in 5 years.
Chess players are more reliable when they speak of Chess than when they speak of themselves.
Speaking of yourself, apparently you don't know what a proof is. You come over, more and more, like an obsessive 11 year old of average ability, with too much self-confidence, who still believes in Father Christmas.
Wait, hold on. You have proof that Father Christmas is a fake? What’s next, Bigfoot isn’t really a big blurry monster roaming the forest? 😉
Oct 14, 2022
0
#4677
mikekalish wrote:
I'm still following.....fascinated....intrigued....in awe. Epic debate and I don't know who's right, if anyone....although I find myself leaning one way or another occasionally.
I’m in agreement with you Mike.
Oct 14, 2022
0
#4678
tygxc wrote:
@6134
[Deleted text reinserted: You don't say what think time was used. You've produced one game for each position -]"how do apply your calculation to determine the game theoretic value"
++ You are confused.
You do your best in that respect.
The game-theoretic value of a draw is determined from the 1469 ICCF WC games, that is the closest to perfect play we have: 1104 of those games are with optimal play from both sides.
Really? I think some of the examples I posted are demonstrably perfect play by both sides. Why don't you check?
The game-theoretic value of a draw also follows from a deductive argument. To win you need to queen a pawn, i.e. you need an advantage of 1 pawn or more. 3 tempi in the initial position are worth 1 pawn. 1 tempo in the initial position is not enough to win. You cannot queen a tempo.
To complete your deductive argument you still need to decide if 3 tempi are worth 1 pawn by weight or volume. You haven't settled that yet.
"It doesn't contradict it."
++ Yes, that is right. You suggested some 7-men endgame might contradict it, but none did.
Only because you haven't got round to checking yet.
"You claim it's generally true" ++ Yes, for positions relevant to weakly solving Chess.
What about positions relevant to checking if it's true?
"you would need rather a lot of confirmatory examples to sustain the conjecture"
++ The thesis was extrapolated from the AlphaZero paper. That is what lends it credibility.
That rather depends on whether your method of extrapolation is correct. Why don't you check and find out. I've provided you with 37 games where, unlike the games in the AlphaZero paper, you have perfect information.
"you need only one counterexample to discredit it"
++ Yes: 'No number of experiments can prove me right, one can prove me wrong' - Einstein
Exactly.
"you'll find among the games I spent so much time preparing"
++ The 2 I posted did not contradict.
I don't think they were among the games I spent so much time preparing, but I think you'll find several there too.
More to the point I think Einstein would probably have been able to find at least one that did contradict.
The KNN vs. KP is irrelevant: 5 men, not 7, not a draw, a known anomaly: an engine can only detect the checkmate if its calculation depth exceeds the distance to mate otherwise the engine is happy to win the pawn increasing its perceived advantage from +5 to +6, but spoiling the win.
It fails on that one in particular. That makes it relevant to verifying if your calculation holds water.
You claim your calculation can determine if the starting position is a win or a draw, so saying that positions that are not draws are irrelevant to checking its correctness makes no sense. Since it also doesn't refer to the number of men neither does saying it has 5 men not 7. (The starting position in your 1469 ICCF WC games probably didn't have 7 men either.)
I invited you here to "Show it or shut it". You have done neither so far. Do you plan to?
And you're still chopping out bits of the post you respond to to alter the context.
Extremely interesting, thanks!
Wits-end wrote:
mikekalish wrote:
I'm still following.....fascinated....intrigued....in awe. Epic debate and I don't know who's right, if anyone....although I find myself leaning one way or another occasionally.
I’m in agreement with you Mike.
That might be the first time in 6140 posts that two people have agreed on anything here. 
@tygxc @Elroch
