It is bonkers to claim that a billion positions are enough! Twice as bonkers (on a geometric scale) as claiming 10^17 positions will do (10^8 times as bonkers on a linear scale).
Chess will never be solved, here's why
Sort:
Oct 13, 2022
0
#4643
tygxc wrote:
@6089
"You haven't posted calculations of any of the positions I posted."
++ Yes, I did. I pointed out your problem with your version of Stockfish.
If you repeat a lie three times it doesn't make it true.
Pointing out a non-existent problem with my version of Stockfish on spurious grounds is not, repeat not, posting your calculations for any of my positions.
"the engine top 1 move coincides with the top 1 engine move"
++ No, the engine top 1 move coincides with the 7-men table base correct move.
"As you yourself remarked it didn't always coincide with with the top 1 move" [Deleted text reinserted: of your unspecified tablebase.]
++ No, the top 1 engine move did coincide with the table base correct move.
A tablebase correct move, not necessarily the top, according to your original post.
"In my 2048 second think time per ply game" ++ The position is not relevant to solving chess and you did something wrong with your version of Stockfish.
I'm neither asserting nor denying that it's relevant to solving chess. That would depend on how you plan to do it. (No please not another cut and paste.)
I am asserting that games played from the position are relevant to checking if your calculations do what you claim they do.
Since your calculations don't refer to anything specific in the starting position they should work when applied to any position. All positions are relevant to checking the validity of your method of calculation.
And I did nothing wrong with my version of Stockfish.
"what think time was used" ++ Much shorter, about 30 minutes for the whole line.
"did you even use Stockfish 15?" ++ No, I used 14, as I wrote above.
Yes, apologies. I've located the post.
"It obviously couldn't have been 2048 seconds" ++ No, much shorter was enough to have the top 1 move coincide with the table base correct move.
"Try running it for 2048 seconds per ply." ++ If I already get the top 1 engine move to coincide with the table base correct move, there is no need for 4 alternatives or a longer thinking time.
If you want to check the validity of your calculations, they give error rates as a function of think time, so a range of times is useful. A think time of 2048 seconds is getting closer to the 17s on a 10^9 NPS that you've quoted. A single game with unspecified think time isn't much use.
(It sounds like you still haven't grasped what minimax pathology is btw.).
[Deleted text reinserted: Instead you continue to vacuously assert] "I had a problem with my Stockfish version." ++ Yes you did.
No I didn't. It's your brain that has a problem.
I get full coincidence of the top 1 engine move and in less time than you. So you have a problem.
If you try generating the games I posted with the same think times (almost certainly with either SF14 or SF15) you will find the same phenomenon.
Not my problem; your lack of understanding.
"Yes, it's a relevant position." ++ Glad you at least agree on that. Do you also arrive at coincidence of the top 1 engine move with the table bases exact move?
In your example I do; I haven't generated any games from the position myself.
I think you will also find a few such examples in the games I posted (when you get round to it, that is).
"all positions are relevant to checking the validity of your method of calculation"
++ No, the estimates of perfect play are derived from nearly perfect play.
Positions that cannot result from optimal play by both players are not relevant.
"all positions are relevant to checking the validity of your method of calculation" is correct. Why "++No" and talk about something different? Again you're using the disembodied "relevant".
"If something is relevant then it's relevant to something."
++ 10^17 positions are relevant to weakly solving Chess and 10^44 are not, though legal.
Strongly disagree, but neither is relevant to verifying your calculations.
"proposals of a procedure for weakly solving chess" ++ I have explained the procedure. Start from a drawn ICCF WC game. Analyse 3 alternatives for the last move, then the 2nd to last move etc. until a position from another ICCF WC draw is reached.
Here is another relevant example: the engine top 1 move coincides with the table base exact move
Again no think time or mention of what is playing, so not very useful.
You appear to be saying that only positions and think times where you can demonstrate a good correlation between what your calculations predict and the actual results, or positions where the actual results are unknown, are relevant to verifying your calculations.
I reiterate; All positions are relevant to the verification.
So will you now stop wriggling and post your calculations for the games here? Then we can stop all this pointless discussion about your proposal to solve chess in five years.
Oct 13, 2022
0
#4644
tygxc wrote:
... The table base correct move will always be among those 4, except in 1 case in 10^20 positions ...
And possibly the half dozen or so I've already posted for you on the thread if they don't happen to be in the 10^-20.
But the figure of 10^20 depends on the validity of your calculations.
Why don't you now stop wriggling and post your calculations for the games here? Then you can stop quoting the figure.
@6098
"a billion positions are enough" ++ a billion positions per second
"10^17 positions will do" ++ Yes, 10^17 total relevant.
@6099
"the engine top 1 move coincides with the top 1 engine move"
++ No, the engine top 1 move coincides with the 7-men table base correct move.
"A tablebase correct move, not necessarily the top" ++ It is either table base correct or not.
"That would depend on how you plan to do it." ++ I have explained.
"And I did nothing wrong with my version of Stockfish."
++ Your version makes mistakes and mine does not.
"A think time of 2048 seconds is getting closer to the 17s on a 10^9 NPS that you've quoted."
++ Agree, but if my engine top 1 move is already table base correct I do not need more.
"I haven't generated any games from the position myself." ++ I did.
"only positions and think times where you can demonstrate a good correlation between what your calculations predict and the actual results, or positions where the actual results are unknown, are relevant to verifying your calculations."
++ I posted 2 of your irrelevant positions above and my top 1 engine move coincided with the table base correct move, while yours did not. You deny you have a problem.
I posted 2 relevant positions games and again the top 1 engine move coincided with the table base correct move.
Oct 13, 2022
0
#4647
tygxc wrote:
@6099
"the engine top 1 move coincides with the top 1 engine move"
++ No, the engine top 1 move coincides with the 7-men table base correct move.
Here you're contradicting your own statement.
You said
I previously posted calculations of two of your irrevelant positions. The engine top 1 move coincided with the top 1 engine move.
I gently pointed out
Firstly, I think you'll find the engine top 1 move coincides with the top 1 engine move whatever game played by an engine you consider.
Now I find you attributing your howler to me.
There are multiple tablebases and multiple correct moves for each, so you should say "a correct move" not "the correct move".
"A tablebase correct move, not necessarily the top" ++ It is either table base correct or not.
We already knew you don't understand tablebases. What you say is true for a particular tablebase but not necessarily for different tablebases.
Certainly it is generally not "the top or not" in different tablebases.
"That would depend on how you plan to do it." ++ I have explained.
"And I did nothing wrong with my version of Stockfish."
++ Your version makes mistakes and mine does not.
Not true. SF15 is rated higher than SF14. Your version of Stockfish is just too bone idle to play enough games to make mistakes.
"A think time of 2048 seconds is getting closer to the 17s on a 10^9 NPS that you've quoted."
++ Agree, but if my engine top 1 move is already table base correct I do not need more.
Your calculations take the results of multiple games to determine error rates. That's why I provided you with multiple games. You're not going to get very far with a single game with unknown think time if you're going to apply your calculations to it.
"I haven't generated any games from the position myself." ++ I did.
One as far as I know. Is that enough to talk about "games" plural?
[Deleted text reinserted: You appear to be saying that]"only positions and think times where you can demonstrate a good correlation between what your calculations predict and the actual results, or positions where the actual results are unknown, are relevant to verifying your calculations."
++ I posted 2 of your irrelevant positions above and my top 1 engine move coincided with the table base correct move, while yours did not. You deny you have a problem.
Yes, I deny I have a problem.
You have a comprehension problem.
Some of mine had mistakes, but I think it's likely that some of those with a similar think time to whatever was your average think time did not. (I've not fully checked my games against the tablebases. You're supposed to be doing that.)
If you like you can run the same set of positions I posted with the same fixed think times and we can look at how closely the error rates match. But preferably first just get on with posting your calculations based on the games I've already run.
I posted 2 relevant positions games and again the top 1 engine move coincided with the table base correct move.
Good.
As I said your positions are relevant because all positions are relevant to verifying your calculations.
Now, after all that obfuscation, will you stop wriggling and post your calculations for the games here? Then the rest of us can discuss the topic.
Oct 13, 2022
0
#4648
MARattigan wrote:
You appear to be saying that"only positions and think times where you can demonstrate a good correlation between what your calculations predict and the actual results, or positions where the actual results are unknown, are relevant to verifying your calculations."
There is a phrase for that - confirmation bias. This is a slightly different version of it as @tygxc is using it to fool himself that he can prove the result rather than that the result itself is true.
[Formatting due to chess.com. Don't ask me...]
Oct 13, 2022
0
#4649
As for habitually leaving out bits of your text to make it appear you've said something completely different, there's probably a phrase for that too.
Oct 13, 2022
0
#4650
MARattigan wrote:
As for habitually leaving out bits of your text to make it appear you've said something completely different, there's probably a phrase for that too.
Misrepresentation.
@6106
"There are multiple tablebases and multiple correct moves for each" ++ Yes, and yes, sometimes
"true for a particular tablebase but not necessarily for different tablebases"
++ You do not understand table bases.
The game states and the correct moves are the same in all table bases.
"SF15 is rated higher than SF14."
++ That may well be, but on my desktop with less time the engine move coincides with the/a table base correct move and on your computer it does not, so you have a problem.
"Your calculations take the results of multiple games to determine error rates."
++ Yes: 1469 ICCF WC games and 1000 AlphaZero games.
"Some of mine had mistakes" ++ So you have a problem with your Stockfish configuration.
"preferably first just get on with posting your calculations based on the games I've already run."
++ Preferably get on with relevant positions like the 2 I posted.
Oct 13, 2022
0
#4653
Kasparov once said "computer will never beat top chess human players"...
@6114
Yes, Kasparov said: "Never shall I be beaten by a machine!" in an interview in 1989.
https://www.chesshistory.com/winter/extra/computers.html
"Chess will never be solved" falls into the same category:
people not wanting something to happen and therefore stating it cannot be done.
Oct 14, 2022
0
#4655
tygxc wrote:
@6106
"There are multiple tablebases and multiple correct moves for each" ++ Yes, and yes, sometimes
"true for a particular tablebase but not necessarily for different tablebases"
++ You do not understand table bases.
The game states and the correct moves are the same in all table bases.
Not under competition rules.
Here's Nalimov and Syzygy in the KNNKP position I posted after Black's first move, but with ply count set to 0.
Nalimov shows 12 winning moves.
Syzygy shows 1 winning move.
Here's a related ply count 0 position.
Nalimov shows a win.
Syzygy shows a draw.
"SF15 is rated higher than SF14."
++ That may well be, but on my desktop with less time the engine move coincides with the/a table base correct move and on your computer it does not, so you have a problem.
No. You still have a comprehension problem.
SF's error rates don't necessarily improve with more think time. They can deteriorate.
Get off your arse and generate your own set of games with the same think times if you don't believe me.
"Your calculations take the results of multiple games to determine error rates."
++ Yes: 1469 ICCF WC games and 1000 AlphaZero games.
Conveniently chosen so that you can't tell how many mistakes were made nor how the think times compare with SF at different think times, not even the same ruleset and mostly terminated by agreement or the TCEC draw rule.
"Some of mine had mistakes" ++ So you have a problem with your Stockfish configuration.
No I don't. I have a problem with the mental capacity of the person I'm talking to.
"preferably first just get on with posting your calculations based on the games I've already run."
++ Preferably get on with relevant positions like the 2 I posted.
Then generate a complete set of games. I've spent enough time on it. You're the one who claims your calculation works. Try doing something to test it.
The positions you posted are neither more nor less relevant to verifying the correctness of your method than the ones I already posted.
You only need one counter-example to show it doesn't work. I've generated four sets of games for you here. See if you can find a counter-example among those. Stop prevaricating - post your calculations for those games.
@6116
"Nalimov shows 12 winning moves. Syzygy shows 1 winning move."
++ That does not matter. 1 winning move is enough. A table base does not even have to show winning moves, DTZ or DTM: just the information if the position is a draw or not is enough,
as the rest can be deduced from looking at the legally reachable positions.
"Syzygy shows a draw." ++ No, Syzygy shows a win that exceeds 50 moves without capture or pawn move, that is a win just the same. The 50-moves rule plays no role in weakly solving Chess. Black can achieve the game-theoretic value of the draw without invoking the 50-moves rule.
"SF's error rates don't necessarily improve with more think time. They can deteriorate."
++ I do not believe that. If you believe that, then reduce your time to eliminate your error.
"you can't tell how many mistakes were made" ++ I can tell by statistics.
"how the think times compare with SF at different think times"
++ Derived from the AlphaZero paper
"not even the same ruleset" ++ A more decisive rule set: table base win claims that exceed 50 moves without pawn move or capture are allowed, but such claims do not happen.
"mostly terminated by agreement" ++ Because neither side can win. Likewise none of the decisive games ended in checkmate and none in a 7-men endgame table base win claim,
neither exceeding 50 moves without pawn move or capture or not.
"preferably first just get on with posting your calculations based on the games I've already run."
++ Preferably get on with relevant positions like the 2 I posted.
"Then generate a complete set of games." ++ OK, I will generate some more games.
"The positions you posted are neither more nor less relevant to verifying the correctness of your method than the ones I already posted." ++ My positions are relevant, yours are not.
"You only need one counter-example to show it doesn't work." ++ A relevant one.
"post your calculations for those games." ++ I have shown why they are not relevant and I have shown for 2 that you make an error with your calculation.
Oct 14, 2022
0
#4657
tygxc wrote:
@6116
"Nalimov shows 12 winning moves. Syzygy shows 1 winning move."
++ That does not matter.
It does if you use Nalimov and pick one of the others, but you should have used Syzygy! You said it didn't matter which tablebase you used, right?
Oct 14, 2022
0
#4658
tygxc wrote:
@6114
Yes, Kasparov said: "Never shall I be beaten by a machine!" in an interview in 1989.
Like Sveshnikov, Kasparov is 100% reliable.
@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.
@6096
"Every new position requires an evaluation"
++ Of course: each of the 3 engines evaluates 10^9 positions/s so as to arrive at the 3 white alternatives to the ICCF WC drawn game and the top 1 black defences to those.
"or ranking the top defensive moves. The latter is of course inadequate" ++ No. If the end result is a table base draw or a prior 3-fold repetition, then that retroactively validates all black moves.
"if you use a cursory evaluation you have more chance of missing a good move"
++ There is no cursory evaluation: it is the selection of the top 4 white moves during 17 s on the cloud engine (corresponding to 4.7 h on a desktop): the move of the drawn ICCF WC game and 3 top alternatives. The table base correct move will always be among those 4, except in 1 case in 10^20 positions, while there are only 10^17 relevant positions.