@14330
"Stockfish is very, very good at both." ++ It is weak at strategy, long term planning, openings, endgame fortresses...
"imperfect in ways humans can't help with"
++ The human and the engine complement each other. The human is weak at calculating.
@14328
"ICCF WC is not a valuable source" ++ it is. It is the strongest chess on the planet.
@14330
"Stockfish is very, very good at both." ++ It is weak at strategy, long term planning, openings, endgame fortresses...
"imperfect in ways humans can't help with"
++ The human and the engine complement each other. The human is weak at calculating.
Prove Stockfish 15 is "weak at strategy" compared to humans, with 5 days a move. [Note that it is likely weak at strategy compared to future better engines and better hardware, as well as being weak at endgames compared to a tablebase, thus likely weak with 8 pieces or more even with a tablebase. Sadly, humans are way weaker].
I would guess your glib statement was formed in the era before Stockfish gained a neural network evaluation unit, a change which increased its rating by 100 in a single step. That unit surely does positional evaluation remarkably well (and can do this at every leaf nodes in a tree 60 ply deep).
@14325
"The tablebase generation solution would run out of storage space"
++ There are new storage media, like synthetic DNA.
80 years ago they used ferrite kernels as storage, then came magnetic tapes and disks, CD-ROM, solid state...
It is also possible to store 1 position as 1 bit draw/nodraw, as Tromp has demonstrated a bijection between natural numbers 1, 2, 3... 8726713169886222032347729969256422370854716254.
From the stored 7-men table base a quantum computer can generate 8 men and store it.
Then to 9, 10... 32.
@14328
"ICCF WC is not a valuable source" ++ it is. It is the strongest chess on the planet.
It is not a useful reference for any publications in game theory. It's just anecdotal examples of competitive games.
Your reasoning that the games are perfect because they are draws has been pointed out to be wrong many times.
I am happy to bet you that today's players would lose to those in 10 years time, if they were foolish enough to play with 2024 engines and computers. This means that a crucial part of the reason the players do not win is that they don't play well enough to beat their weak (on an absolute scale) opponents.
Given that both sides have access to the same engines this is no huge surprise. If you acknowledge this, your conclusion of perfection is crucially dependent on the opposition not being as strong as it could be!
@14334
"Prove Stockfish 15 is "weak at strategy" with 5 days a move."
++ Any fortress position does that.
'computer engines did not understand the main ideas and suggested in most middlegame positions that all candidate moves were equivalent' - Jon Edwards
'I accomplished the feat in 38 moves, in a sequence that no computer would consider or find.' - Jon Edwards
'I began to implement an idea that few over-the-board players and no computers would consider or attempt.' Jon Edwards
'It takes chess wisdom to escape with a draw even though Stockfish and co have calculated 2+ scores most of the middle game.' - Langeveld
@14336
"This means that a crucial part of the reason the players do not win is that they don't play well enough to beat their weak (on an absolute scale) opponents."
++ That your the paired error hypothesis.
It makes sense for some errors to come in pairs, but not all errors.
If there were double errors, then there should also be a single error i.e. decisive game.
"Given that both sides have access to the same engines"
++ No. In Preliminaries, Semifinals, Candidates there are still decisive games, despite same engines.
4 of the 17 finalists who qualified are Russian, despite inferior hardware due to sanctions.
In previous years there were decisive games despite same engines.
No, a fortress position is irrelevant if Stockfish never lets you get it.
Jon Edwards is talking about an engine that would beat him. Whose evaluation is to be trusted?
You've given a list of reasons decisive games occur and explained that the players blundered. I bet the Russians made fewer clerical errors.
Where is your documentation of the exact difference in hardware the Russians had?
tygxc knows the engines aren't perfect.
He's already conceded that.
But that means in turn that his other claims are false. Wrong.
But as so often happens with illogic - the 'illogician' doesn't care.
tygxc knows the engines aren't perfect.
Yes, half the time he argues a person magically makes the engine perfect, and his argument for their perfection is that other people were imperfect enough to lose. By this reasoning I am perfect because I beat someone.
He's already conceded that.
But that means in turn that his other claims are false. Wrong.
But as so often happens with illogic - the 'illogician' doesn't care.
@14326
"True or perfect endings are not possible" ++ They are.
"You can't tell the difference between e4 and d4" ++ No difference, both draw.
"The true end will never be achieved" ++ It is being done now: 113 draws out of 113 games.
"There are more than 2 best moves" ++ 1 e4 and 1 d4 are played now, also 1 Nf3.
1 c4 was played in the previous ICCF WC Finals.
"You don't understand what "true end" we are talking about." ++ Please explain.
Explain how a "true end" may be achieved. ICCF WC is not a valuable source, it isn't apart of this party.
First of all define "true end".
When you say, "a perfect game ends in: ???", that would normally be taken to refer to a game with perfect play. The "???" is not a specific number in that case.
E.g., a position with the following diagram is a theoretical draw under both basic rules or competition rules.
It can be drawn in a minimum of two ply (e.g. 1.White offers draw ...Black accepts draw) or in any number of ply (including ℵ₀ under basic rules - arguably limited under competition rules), all with perfect play.
Syzygy plays the positions it addresses perfectly under both basic and competition rules, but from winning positions the "???" is variable depending on which of a large number of solutions it offers is chosen from a forced mate half tree.
E.g. these are both examples of Syzygy playing mate perfectly from the position shown (the first it's best attempt at accuracy)
Or a Nalimov version (which can also play this particular position perfectly under both sets of rules).
Note the discrepency in ???.
@14339
"a fortress position is irrelevant if Stockfish never lets you get it" ++ Stockfish happily goes for +5 evaluation and does not recognise the fortress. It even does not recognise a dead position.
"Jon Edwards is talking about an engine that would beat him. Whose evaluation is to be trusted?" ++ Edwards of course, he knows both.
"I bet the Russians made fewer clerical errors." ++ Maybe they have more discipline. However, not all decisive games are clerical errors or due to illness. Several are due to opening choice or wrong engine evaluation exploited by the human.
"Where is your documentation of the exact difference in hardware the Russians had?"
++ 'the lack of decent equipment is being felt quite notably in Russia, where embargoes have limited access to new hardware and the latest chess tools' - Edwards
'Computers are stronger abroad' - Dronov
Somehow - some pages/posts appear to have just now disappeared.
With posts by several posters disappearing with them.
@14328
"ICCF WC is not a valuable source" ++ it is. It is the strongest chess on the planet.
ok and? it's not rigorously proven so it's not a valuable source.
tygxc why arent you addressing the core fact that any statement about a solution of chess must be rigorously mathematically proven with more than just engine games and your personal feelings?
#14296
3-fold repetition does not ensure a finite game. Some positions that don't have checks make this rule ineffective.
Did you understand why it's unnecessary to ensure a finite game to allow for a solution? The finite number of positions already guarantees a solution.
If no checks can occur in any continuation from a position, that ensures a finite game because the game immediately terminates under the dead position rule.
In such positions, 3-fold repetition will only work after hundreds of continuous shuffling. The 50-move rule is the only and only way to ensure a game will not go that long, that's why the longest chess game mathematically is 8848 moves.
Not in all. In some positions the 3-fold repetition will necessarily occur before 50 moves (assuming you ignore the dead position rule).
There is no true end for chess. Soon, some computers will eventually replace Stockfish, millions of lines may be lurking in the shadows, waiting for our intelligence to find it and do even more crazy stuff. There is no true end.
MEGA is correct.
tygxc continues to assert 'hey these engines are really Strong!'
but he refuses to concede that that doesn't prove anything.
Djokovic is really Strong - but he just lost to young Alcazar.
Carlos A. is a 'marvel'. A genius and athlete.
Apparently even more 'complete' than Novak is.
And that's going a long way.
Will Novak's knees last long enough to 'solve' it?
We don't know and if we're wise we don't care.
But that doesn't mean it isn't interesting.
@14326
"True or perfect endings are not possible" ++ They are.
"You can't tell the difference between e4 and d4" ++ No difference, both draw.
"The true end will never be achieved" ++ It is being done now: 113 draws out of 113 games.
"There are more than 2 best moves" ++ 1 e4 and 1 d4 are played now, also 1 Nf3.
1 c4 was played in the previous ICCF WC Finals.
"You don't understand what "true end" we are talking about." ++ Please explain.
Explain how a "true end" may be achieved. ICCF WC is not a valuable source, it isn't apart of this party.
First of all define "true end".
When you say, "a perfect game ends in: ???", that would normally be taken to refer to a game with perfect play. The "???" is not a specific number in that case.
E.g., a position with the following diagram is a theoretical draw under both basic rules or competition rules.
It can be drawn in a minimum of two ply (e.g. 1.White offers draw ...Black accepts draw) or in any number of ply (including ℵ₀ under basic rules - arguably limited under competition rules), all with perfect play.
Syzygy plays the positions it addresses perfectly under both basic and competition rules, but from winning positions the "???" is variable depending on which of a large number of solutions it offers is chosen from a forced mate half tree.
E.g. these are both examples of Syzygy playing mate perfectly from the position shown (the first it's best attempt at accuracy)
Or a Nalimov version (which can also play this particular position perfectly under both sets of rules).
Note the discrepency in ???.
True end in this case refers to a perfect game and that there is no way you can go further. A game ends with players play with precision up to 100%. Unfortunately, computers can't have an answer to that. There are no best moves. e4,d4,c4, and f4 are white's choices, while you can tell a difference between e4 and f4, it is hard to tell the difference between d4 and e4. Generally speaking, chess will always have 2 best moves which will result in a similar outcome, 2 best lines or ways that still may have the same outcome, making the game doesn't have a true end.
In this case, if white is SF 16, he would checkmate black quicker than a player using the simple method to checkmate where you box him to the corner and checkmate.
some positions may require white or black to play 100% accurately to win. How to achieve those positions? the opposing team must play a wrong move or line that puts him in a tough situation in the first place, so this situation is removed as it is not perfect in the beginning.
#14296
3-fold repetition does not ensure a finite game. Some positions that don't have checks make this rule ineffective.
Did you understand why it's unnecessary to ensure a finite game to allow for a solution? The finite number of positions already guarantees a solution.
The (hypothetical) generation of a basic chess tablebase makes this explicit. Basic chess has no 3-fold repetition rule nor any rule like the 50 move rule. Players have to agree a draw if there is no stalemate.
The way the tablebase generation works is that it is done for a single combination of pieces on the board at a time. Smaller numbers of pieces are dealt with first historically, and it is clear it is worth dealing with them in order of number of pawns on the board from there (starting with no pawns). See below for why
With a chosen set of pieces, you can start with all random positions with no values being known.
Those that are obviously illegal can be eliminated, but this does not need to be thorough.
Pass 1 identifies which positions have mate on the board. It could also identify stalemate on the board, but this is not essential, so we'll ignore it. The value of these positions is set.
A single partial forward pass is needed, to see which positions have legal captures or promotions to get to a position of known value (in a previously calculated tablebase). Other moves can be ignored for now (because they don't leave the current tablebase).
Pass 2 does retrograde analysis just from those positions whose value was determined in pass 1. This is like playing chess backwards - it works out all positions from which a legal move would reach the position. This can be an uncapture or an unpromotion, but the former would be moving to a larger tablebase, so is not done at this point.
This identifies all positions where a player can play a move to mate. This determines the value of those positions.
Pass 3 does retrograde analysis from those positions. This is a little different because of the different objective. It identifies positions where a player could play so that they get mated, but, more importantly, for some of those positions all legal moves are now seen to lead to being mated. The value of these positions is now clear.
Later passes use the latest set of positions with a clear value to generate a new set of positions with a clear value.
At some point a step occurs when no new positions with a clear value are generated. Not only do we now know the value of all positions with a clear value (winning or losing for the player to play), all the others are now known to have a value of a draw.
In basic chess we can imagine people playing infinite games, ignoring the inevitable 3-fold repetitions, but we can know that if neither player blunders, neither player can win. We consider that a draw for theoretical purposes, although for practical purposes it is merely a permanent failure to win!
Adding in a repetition rule or a n-move rule for drawing makes it finite, which saves a lot of time.
[I'd like to think the above may be a bit clearer than the description at https://en.wikipedia.org/wiki/Endgame_tablebase#Generating_tablebases but it also has some more detail.
"Stockfish is very, very good at both. And imperfect in ways humans can't help with."
Elroch correct and tygxc wrong as usual.