“Weakly solving only needs 1 strategy to draw for black, not all black moves.
So instead of w^(2d) positions only w^d = Sqrt (w^(2d)) positions
That is where the square root comes from.”
assumes pre existing knowledge of which positions to evaluate
Chess will never be solved, here's why
Sort:
Feb 12, 2023
0
#6402
Feb 12, 2023
0
#6403
Tygxc you do realize that Sveshnikov was exaggerating right. Like that should be obvious to you
@8750
"Everything after Tromp's number"
++ Tromp's number 10^44 is relevant for strongly solving Chess to a 32-men table base.
For weakly solving Chess Sqrt (10^37 * 10 / 10^4) = 10^17 positions are relevant.
.
tygxc wrote:
@8750
"Everything after Tromp's number"
++ Tromp's number 10^44 is relevant for strongly solving Chess to a 32-men table base.
For weakly solving Chess Sqrt (10^37 * 10 / 10^4) = 10^17 positions are relevant.
Incorrect. 10^44 is the only relevant and reliable number for both solutions currently, until you can prove differently, and you cannot, and have not in several years of posting the same thing over and over. There's not a single valid number in your red equation above.
@8751
"assumes pre existing knowledge of which positions to evaluate"
++ No, it does not assume any knowledge of which positions to evaluate.
It is just like Schaeffer did for Checkers. Each pawn move and each capture along the seeded line shrink the relevant search space and the stored boundary.
@8752
"Sveshnikov was exaggerating"
++ Sveshnikov was right.
Feb 12, 2023
0
#6409
tygxc wrote:
@8752
"Sveshnikov was exaggerating"
++ Sveshnikov was right.
He was right to exaggerate. It was a great quote, reflecting how computers were transforming chess. But it wasn’t literal. Otherwise he wouldn’t have said that
Feb 12, 2023
0
#6410
tygxc wrote:
@8751
"assumes pre existing knowledge of which positions to evaluate"
++ No, it does not assume any knowledge of which positions to evaluate.
It is just like Schaeffer did for Checkers. Each pawn move and each capture along the seeded line shrink the relevant search space and the stored boundary.
Then how come schaeffer didn’t evaluate the square root of positions?
Feb 12, 2023
0
#6411
tygxc wrote:
@8751
"assumes pre existing knowledge of which positions to evaluate"
++ No, it does not assume any knowledge of which positions to evaluate.
It is just like Schaeffer did for Checkers. Each pawn move and each capture along the seeded line shrink the relevant search space and the stored boundary.
That’s argument from repetition.
You literally contradict yourself here. 10^17 isn’t the search space of a weak solution, it’s the total positions that make up the weak solution. 10^34 / 44 is the search space.
Feb 12, 2023
0
#6413
You also still haven’t addressed the fact that I have objectively proven that the strategy stealing method cannot work.
ardutgamersus wrote:
i eat glue
This is sort of what I was talking about with the idea of the chess engines themselves not necessarily being programmed correctly.
Some things are counterintuitive. 
@8759
"how come schaeffer didn’t evaluate the square root of positions?"
He evaluated 10^14 relevant positions of the 5*10^20 legal positions.
Checkers is no Chess.
A Checkers board is more crowded: 24 men on 32 squares as opposed to 32 men on 64 squares.
Checkers has only 2 kinds of men, Chess 6.
A Checkers board has more edge effects: 16 of 32 squares on the edge as opposed to 30 of 64.
shimel42 wrote:
Never is too long.
(and assumes things stay stagnant/progress at whatever current rates are thought possible in terms of computing...)
Never, as storage sits currently (or with any serious/reasonable predicted advances). There's not enough matter in our solar system to do the job. So unless you are going to invent FTL travel before you solve chess...
@8761
"How do you determine which move by white to do?"
++ At first sight all legal white moves and 1 black response.
On closer inspection some clearly wrong moves like 1 e4 e5 2 Ba6? need no inspection.
The idea is to take the top w white engine moves, e.g. w = 4.
As proven the table base exact move is among the top 4 engine moves when running 17 s on a 10^9 nodes/s engine with 1 error in 10^20 positions.
@87660
"10^34 / 44 is the search space"
++ How do you arrive at that?
Feb 12, 2023
0
#6419
It’s short hand for 10^34 (your claimed number) OR the established 10^44
Feb 12, 2023
0
#6420
tygxc wrote:
@8761
"How do you determine which move by white to do?"
++ At first sight all legal white moves and 1 black response.
On closer inspection some clearly wrong moves like 1 e4 e5 2 Ba6? need no inspection.
The idea is to take the top w white engine moves, e.g. w = 4.
As proven the table base exact move is among the top 4 engine moves when running 17 s on a 10^9 nodes/s engine with 1 error in 10^20 positions.
Wdym “as proven”
you LITERALLY ASSUMED A FALSE ERROR DISTRIBUTION
also, for that 17s running, it would take over a million years to calculate 10^17 positions.
your nodes/sec is what is required to calculate a position with that accuracy every 17 seconds
If you are running 17s then BY DEFINITION it would take
++ Quantum computers are commercially available and they can run Stockfish if translated from C++ to Python. Quantum computers may be much faster in generating 8- or 9- men endgame table bases in the foreseeable future.
Quantum computers *cannot* run the full instruction sets of C++ or Python. I already schooled you on this before. They can run a limit subset of instructions that do not include things like looping, etc. Quantum computers cannot output intermediate results or use them in further processing, because reading results from quantum matrixes destroys them. What good does it do you, for example, to increment a counter when you cannot store and read back what your counter was at a given point in time? You would have to solve chess in one pass. Good luck with that
.
++ Solving Chess does not need a single Floating Point Operation, so peteFLOPS are irrelevant. Indeed strongly solving Chess to a 32-men table base with all 10^44 legal positions is beyond existing technology.
It doesn't matter what CPU measure you use.
++ There is solid basis for that. Per Tromp there are 10^44 legal positions, but as the 3 displayed randomly sampled positions show, the vast majority has > 3 rooks and / or bishops per side, so they never can result from optimal play from both sides.
Therefore Gourion's 10^37 is a better estimate. It is a bit too restrictive: positions with 3 or 4 queens do occur in perfect games with optimal play from both sides. So multiply by 10 to incorporate these. That is where the * 10 comes from.
However, a random sample of 10,000 positions as counted by Gourion show none can result from optimal play by both sides. That is where the / 10,000 comes from.
Weakly solving only needs 1 strategy to draw for black, not all black moves.
So instead of w^(2d) positions only w^d = Sqrt (w^(2d)) positions
That is where the square root comes from.
So
Yields: Sqrt (10^37 * 10 / 10,000) = 10^17 positions relevant to weakly solving Chess.
Everything after Tromp's number is just conjecture and assumptions. Gourian = not proven or peer reviewed. Dividing by a further 10,000 is just a number you had to use to reach your 10^17 goal, and has no basis. Taking the square root is a *possibility*, but not proven to work for Chess. So, the only real number we're working with here is 10^44. All the rest of your stages of implausible reductions depend on assumptions you cannot prove.
So, your premise that is we assume all of this unproven steps reducing from 10^44 to 10^17 are perfect, that your prediction of error rates from engine evaluations holds up, *and* that 3 human beings can correct for errors that slip through without making mistakes themselves in 5 years. That is laughably weak.