tygxc wrote:
@8269
"we dont already know chess is a draw" ++ We know Chess is a draw.
"give actual evidence" ++ I have given evidence several times.
1 tempo < 1 pawn
and theres your error.
not even 1 sentence in you make an assumption.
it doesnt matter that a pawn could queen.
Strategy stealing shows Chess cannot be a black win: for every presumed forced black win there exists a corresponding forced white win by losing a tempo.
white cant lose tempo in that way, its physically impossible.
in fact, i can prove it.
@8252
"errors in your logic" ++ There are no errors. What you fail to understand is no error.
"the decisive game/ time per move" ++ The more time / move, the less decisive games and the less mistakes. At unlimited time zero decisive games and zero errors.
"1 error per decisive game" ++ Every decisive game has an odd number of errors.
At higher rates of decisive games there are games with 1 error, with 3 errrors, with 5 errors...
As the error rate and the rate of decisive games goes down like in Tata Steel Masters 2023 there are only decisive games with 1 or 3 errors. As the rate of decisive games goes further down,
like in the ICCF WC Finals, decisive games have 1 error only.
we dont know that. you just made more assumptions.
"comparison between alphazero and your '10^9 nodes' thing"
++ AlphaZero provides some peer reviewed conclusions of extensive calculations.
10^9 nodes/s is the present state of technology of existing cloud engines.
It mainly serves to convert positions to time needed.
and you made some BS conversion between the two, thats the whole point.
current 10^9 cloud engines would run alphazero on 10^6 nodes.
"44,000,000 * 60 Seconds = 2,640,000,000 Nodes per Move for Stockfish 8
50,286 * 60 Seconds = 3,017,160 Nodes per Move for Leela"
(google)
"there were pre set starting openings" ++ No, in that table AlphaZero was free to open.
no, for that table IT EXPLICITLY SAYS THE OPPOSITE.
"As AlphaZero is approximately deterministic given the same MCTS depth
and number of rollouts, we promote diversity in games by
sampling the first 20 plies in each game proportional to the
softmax of the MCTS visit counts, followed by playing the
top moves for the rest of the game."
"your program could catch any errors made"
++ No. The program only calculates to the 7-men endgame table base.
If a white win were reached, then some black move was an error and has to be retracted.
If a black draw or win is reached, then the black moves need not be questioned, but alternatives for the white moves need to be explored.
theres no method of exploration defined in your program. after all, YOU said 10^17.
"different nodes are of comparable strength"
++ Does not matter. The calculation just hops from the initial position to other drawn positions until it reaches a 7-men table base draw or a prior 3-fold repetition.
yeah strength matters. the alphazero evaluations dont take the same space as a stockfish node. your math is therefore off by a factor of at LEAST a thousand.
you are putting a brute force amount of calculations onto a neural network, that isnt able to be done.
"the 'top 4' are going to have equivalent individual chances of finding the correct moves"
++ probability (all 4 top engine moves are errors) = (probability top 1 move is error)^4.
yeah. thats an assumption. one that isnt necessarily true.
"a pre existing set of positions" ++
The task of the good assistants i.e. (ICCF) (grand)masters is to work out suitable starting positions of preferably 26 men for the calculation by the modern computers i.e. 3 cloud engines of 10^9 nodes/s (or 3,000 desktops of 10^6 nodes/s).
engines can find the suitable starting position better.
"jump to the 10^9 nodes" ++ There is no jump. 10^9 nodes/s is the state of the art speed of existing cloud engines now. That is to convert positions to time needed.
"what you erred" ++ I did not err, you failed to understand.
If you are a math student, then you should be able to understand. It is only basic math.
im literally a math student. you are just having a dunning kruger moment, thinking you know something that the rest of the world hasnt figured out yet.
your "top 4 engine moves" is also self contradicting to your own claims. to find the top 4 engine moves would be 4^50 * 10^17 to calculate, as a heuristic.