Correspondence ICCF Champion Vs Top Engine ?

[Uri Blass]

In the July ratings list I will be around 2350 at ICCF. Any top correspondence player would have an easy time against any unassisted engine with regular correspondence time controls (10 moves in 40 days). The problem with unassisted engines is flawed pruning methods, sometimes they prune out the key moves or don't see the danger until it's too late. I've seen it so many times in my correspondence games.

I wonder how many positions from your corresepondence games you can give when latest stockfish does a losing mistakes at 1 hour per move.
I will be surprised if you find many positions.

[Leto]

Lately I've been avoiding well-researched openings, so any books would be practically irrelevant if I were the correspondence player in such a match.

Always curious, what is your OTB rating - FIDE or National Federation?

Zero. All my life I've been fascinated by computer chess and AI.

[Leto]

In the July ratings list I will be around 2350 at ICCF. Any top correspondence player would have an easy time against any unassisted engine with regular correspondence time controls (10 moves in 40 days). The problem with unassisted engines is flawed pruning methods, sometimes they prune out the key moves or don't see the danger until it's too late. I've seen it so many times in my correspondence games.

I wonder how many positions from your corresepondence games you can give when latest stockfish does a losing mistakes at 1 hour per move.
I will be surprised if you find many positions.

Couldn't we just look at the games Stockfish loses on TCEC. I know TCEC is played at a faster time control but the hardware is quite formidable, certainly better than mine.

[Milos]

In the July ratings list I will be around 2350 at ICCF. Any top correspondence player would have an easy time against any unassisted engine with regular correspondence time controls (10 moves in 40 days). The problem with unassisted engines is flawed pruning methods, sometimes they prune out the key moves or don't see the danger until it's too late. I've seen it so many times in my correspondence games.

I wonder how many positions from your corresepondence games you can give when latest stockfish does a losing mistakes at 1 hour per move.
I will be surprised if you find many positions.

Couldn't we just look at the games Stockfish loses on TCEC. I know TCEC is played at a faster time control but the hardware is quite formidable, certainly better than mine.

You will never be able to drive SF to a position even remotely similar (disadvantages) as in TCEC. If you really believe so, you are highly delusional.
Moreover, why would you limit SF to 1h per move on some mediocre hardware. You yourself mentioned 10 moves in 40 days (usually is 50, but lets not get into details). That is 1000 times longer TC than in TCEC. Are you even aware what three orders of magnitude of extra thinking time actually mean??? That's 10 doublings in time. Even if one doubling is only 20 Elo due to diminishing returns, that's still 200 Elo extra. Do you have any idea how large difference in strength is that on such a high level of play?

[jr66]

In the July ratings list I will be around 2350 at ICCF. Any top correspondence player would have an easy time against any unassisted engine with regular correspondence time controls (10 moves in 40 days). The problem with unassisted engines is flawed pruning methods, sometimes they prune out the key moves or don't see the danger until it's too late. I've seen it so many times in my correspondence games.

I wonder how many positions from your corresepondence games you can give when latest stockfish does a losing mistakes at 1 hour per move.
I will be surprised if you find many positions.

Couldn't we just look at the games Stockfish loses on TCEC. I know TCEC is played at a faster time control but the hardware is quite formidable, certainly better than mine.

You will never be able to drive SF to a position even remotely similar (disadvantages) as in TCEC. If you really believe so, you are highly delusional.
Moreover, why would you limit SF to 1h per move on some mediocre hardware. You yourself mentioned 10 moves in 40 days (usually is 50, but lets not get into details). That is 1000 times longer TC than in TCEC. Are you even aware what three orders of magnitude of extra thinking time actually mean??? That's 10 doublings in time. Even if one doubling is only 20 Elo due to diminishing returns, that's still 200 Elo extra. Do you have any idea how large difference in strength is that on such a high level of play?

I agree for TCEC positions as i underlined above ( positions are made for possible wins ).
About time control, although i know some players use their machine for a long time in a good hardware, many others don't and play also several games in the same time ( sometimes too much hence losses by what we could call "CC zeitnot" )
You will never beleive me but, for what i remember, i did not use often more than say 10h machine for decide a move and if i use an automatical tool, it is just for control and take about 3 h max ( i am perhaps very special in ICCF world I Guess ? )
In this conditions, i think the engine will beat easely all CC players as me in a match because he will have hundreds of Elo more ( one housand if you consider i use now a normal laptop with an AMD 4600H ).....
I don't know because nobody tried it seem ?
I could play a game but, as i said, a new tournament will start and i could not give many time, perhaps ridiculous time per move ( one hour max ??? )
If many people here would be curious if i would be ridiculous, i can try a game if you want because i don't care to lose ?

[Leto]

In the July ratings list I will be around 2350 at ICCF. Any top correspondence player would have an easy time against any unassisted engine with regular correspondence time controls (10 moves in 40 days). The problem with unassisted engines is flawed pruning methods, sometimes they prune out the key moves or don't see the danger until it's too late. I've seen it so many times in my correspondence games.

I wonder how many positions from your corresepondence games you can give when latest stockfish does a losing mistakes at 1 hour per move.
I will be surprised if you find many positions.

Couldn't we just look at the games Stockfish loses on TCEC. I know TCEC is played at a faster time control but the hardware is quite formidable, certainly better than mine.

You will never be able to drive SF to a position even remotely similar (disadvantages) as in TCEC. If you really believe so, you are highly delusional.
Moreover, why would you limit SF to 1h per move on some mediocre hardware. You yourself mentioned 10 moves in 40 days (usually is 50, but lets not get into details). That is 1000 times longer TC than in TCEC. Are you even aware what three orders of magnitude of extra thinking time actually mean??? That's 10 doublings in time. Even if one doubling is only 20 Elo due to diminishing returns, that's still 200 Elo extra. Do you have any idea how large difference in strength is that on such a high level of play?

You have a good point about the openings, TCEC does sometimes use openings that makes it more likely an engine would lose.
I have doubts that the extra time helps an engine much, I'd estimate the elo increase in the range of 50, not anywhere close to 200. I was a part of the World vs correspondence chess grandmasters matches that playchess.com ran many years ago and we've had people that would run the machines on infinite analysis for days and I remember them saying the engines top move choice doesn't change often given the extra time.

[Chessqueen]

In the July ratings list I will be around 2350 at ICCF. Any top correspondence player would have an easy time against any unassisted engine with regular correspondence time controls (10 moves in 40 days). The problem with unassisted engines is flawed pruning methods, sometimes they prune out the key moves or don't see the danger until it's too late. I've seen it so many times in my correspondence games.

I wonder how many positions from your corresepondence games you can give when latest stockfish does a losing mistakes at 1 hour per move.
I will be surprised if you find many positions.

Couldn't we just look at the games Stockfish loses on TCEC. I know TCEC is played at a faster time control but the hardware is quite formidable, certainly better than mine.

You will never be able to drive SF to a position even remotely similar (disadvantages) as in TCEC. If you really believe so, you are highly delusional.
Moreover, why would you limit SF to 1h per move on some mediocre hardware. You yourself mentioned 10 moves in 40 days (usually is 50, but lets not get into details). That is 1000 times longer TC than in TCEC. Are you even aware what three orders of magnitude of extra thinking time actually mean??? That's 10 doublings in time. Even if one doubling is only 20 Elo due to diminishing returns, that's still 200 Elo extra. Do you have any idea how large difference in strength is that on such a high level of play?

You have a good point about the openings, TCEC does sometimes use openings that makes it more likely an engine would lose.
I have doubts that the extra time helps an engine much, I'd estimate the elo increase in the range of 50, not anywhere close to 200. I was a part of the World vs correspondence chess grandmasters matches that playchess.com ran many years ago and we've had people that would run the machines on infinite analysis for days and I remember them saying the engines top move choice doesn't change often given the extra time.

What was your last Correspondence rating and do you feel like plying versus Komodo Dragon2 MCTS with some odds depending on your last rating?
http://talkchess.com/forum3/viewtopic.p ... &start=210

[Leto]

In the July ratings list I will be around 2350 at ICCF. Any top correspondence player would have an easy time against any unassisted engine with regular correspondence time controls (10 moves in 40 days). The problem with unassisted engines is flawed pruning methods, sometimes they prune out the key moves or don't see the danger until it's too late. I've seen it so many times in my correspondence games.

I wonder how many positions from your corresepondence games you can give when latest stockfish does a losing mistakes at 1 hour per move.
I will be surprised if you find many positions.

Couldn't we just look at the games Stockfish loses on TCEC. I know TCEC is played at a faster time control but the hardware is quite formidable, certainly better than mine.

You will never be able to drive SF to a position even remotely similar (disadvantages) as in TCEC. If you really believe so, you are highly delusional.
Moreover, why would you limit SF to 1h per move on some mediocre hardware. You yourself mentioned 10 moves in 40 days (usually is 50, but lets not get into details). That is 1000 times longer TC than in TCEC. Are you even aware what three orders of magnitude of extra thinking time actually mean??? That's 10 doublings in time. Even if one doubling is only 20 Elo due to diminishing returns, that's still 200 Elo extra. Do you have any idea how large difference in strength is that on such a high level of play?

You have a good point about the openings, TCEC does sometimes use openings that makes it more likely an engine would lose.
I have doubts that the extra time helps an engine much, I'd estimate the elo increase in the range of 50, not anywhere close to 200. I was a part of the World vs correspondence chess grandmasters matches that playchess.com ran many years ago and we've had people that would run the machines on infinite analysis for days and I remember them saying the engines top move choice doesn't change often given the extra time.

What was your last Correspondence rating and do you feel like plying versus Komodo Dragon2 MCTS with some odds depending on your last rating?
http://talkchess.com/forum3/viewtopic.p ... &start=210

2336 ICCF (https://www.iccf.com/player?id=516044&tab=3 ) but this rating period ends tomorrow and I'm forecasted to be 2352 in the July rating list. I'd only be interested in something resembling correspondence time controls, I could handle a move every two days or one move a day.

[MonteCarlo]

There was one recent test in these forums with some pretty strict restrictions for the unassisted engine, operated by me, against jefk.

The engine played black, used maximum 2 hours per move, cleared hash each move, used neither opening books nor tablebases, and was on low-to-middling hardware based on NPS for SF and LC0.

By his own admission, jefk played a bit cautiously since he didn't know my exact binary/config choices (this is important, I think, as otherwise it would be much easier to guide the unassisted engine somewhere difficult for it), but from a human chess player's perspective black was never in any difficulties that game.

Perhaps against a higher ranked correspondence player playing more aggressively to exploit the limitations (especially no TBs) things would be different, but I suspect that beating "engine slaves" is just extraordinarily difficult these days, and maybe requires luck and sheer number of attempts more than anything else.

Cheers!

[jr66]

In fact, we beat more often humans blunders ( moves inversed or forgotten etc ) than engines in CC today !
@ MonteCarlo
A Book is now integrated in SF NNUE and so, you cannot play wihout book i guess ?
No TB can be indeed an advantage but engines had also many endings integrated before and i don't know if they removed them now ?
Regards
Jacques