ELO rating

@tpr said in #25:

As ICCF shows: grandmaster + engine beats grandmaster + engine at 5 days average/move in qualifiers, semifinals and finals.

ICCF World Championship 33, started in 2022, had 126 draws and zero decisive games. ICCF World Championship 34, started in 2024, has had 13 draws and zero decisive games.

This is not only true at the championship level. ICCF World Candidates 2024 has had 136 draws and one decisive games. Humans are very simply no longer competitive factors for high level engine play.

(And the one decisive game just features an unexplainable, enormous 100+ centipawn blunder, which is a move that isn't even in the engine's top five suggestions)

@tpr said in #25: > As ICCF shows: grandmaster + engine beats grandmaster + engine at 5 days average/move in qualifiers, semifinals and finals. ICCF World Championship 33, started in 2022, had 126 draws and zero decisive games. ICCF World Championship 34, started in 2024, has had 13 draws and zero decisive games. This is not only true at the championship level. ICCF World Candidates 2024 has had 136 draws and one decisive games. Humans are very simply no longer competitive factors for high level engine play. (And the one decisive game just features an unexplainable, enormous 100+ centipawn blunder, which is a move that isn't even in the engine's top five suggestions)

ColossusChess edited

#32

In order to make this interesting again you would have to play from another starting position and completely prohibit the influence of any outside help, only own human analysis allowed. Allowing only the use of databases or Chess theory, like Lichess does, is not enough since they contain mixed human and engine moves.

Obviously then cheating would be a thing but Cheaters might be catched eventually when the number of suspicious cases is high enough.

This might be off topic but ICCF is completely irrelevant since a long time. They became obsolete even before the Computers got strong when rich bastards started to pay human GMs to analyze their games for them. In order to make this interesting again you would have to play from another starting position and completely prohibit the influence of any outside help, only own human analysis allowed. Allowing only the use of databases or Chess theory, like Lichess does, is not enough since they contain mixed human and engine moves. Obviously then cheating would be a thing but Cheaters might be catched eventually when the number of suspicious cases is high enough.

tpr

#33

#29
"difference between 3 minutes and 5 days is not as big" * It is. The late ICCF GM Dronov proposed to shorten the ICCF time control from the present 50 days per 10 moves to avoid all draws.

"the human assistance pretty much useless now"

No, the human piloting is essential, especially in the opening, where the engines run into a horizon barrier. Chess is most complicated with 26 men: the number of positions rises from 32 to 26 men and then declines. See table 4
https://univ-avignon.hal.science/hal-03483904/document

Also the fact that the TCEC super finals shows a winner proves there is still room for progress. The 50 openings imposed by a human panel are played twice: once with white and once with black. Some openings end 0.5 - 0.5, so the starting position is a draw. Some openings end 1 - 1, so the starting position was a win for one side. However, some end 1.5 - 0.5, that means either the starting position was a draw and one engine erred and lost, or the starting position was a win and one engine failed to win.

Progress will also come once the 8-men endgame table base is available in lieu of the present 7-men endgame table base.

#29 "difference between 3 minutes and 5 days is not as big" * It is. The late ICCF GM Dronov proposed to shorten the ICCF time control from the present 50 days per 10 moves to avoid all draws. "the human assistance pretty much useless now" * No, the human piloting is essential, especially in the opening, where the engines run into a horizon barrier. Chess is most complicated with 26 men: the number of positions rises from 32 to 26 men and then declines. See table 4 https://univ-avignon.hal.science/hal-03483904/document Also the fact that the TCEC super finals shows a winner proves there is still room for progress. The 50 openings imposed by a human panel are played twice: once with white and once with black. Some openings end 0.5 - 0.5, so the starting position is a draw. Some openings end 1 - 1, so the starting position was a win for one side. However, some end 1.5 - 0.5, that means either the starting position was a draw and one engine erred and lost, or the starting position was a win and one engine failed to win. Progress will also come once the 8-men endgame table base is available in lieu of the present 7-men endgame table base.

Pejathaya

#34

The highest stockfish is 3700 where hikaru in chess. Com is 3500 almost as the stockfish

tpr

#35

#32
"ICCF is completely irrelevant"

It is the most relevant as to the truth about chess.

"completely prohibit the influence of any outside help"

That is impossible to control.

#32 "ICCF is completely irrelevant" * It is the most relevant as to the truth about chess. "completely prohibit the influence of any outside help" * That is impossible to control.

tpr

#36

Here is an example of an ICCF game from an ongoing qualifier:
https://www.iccf.com/game?id=1531671
Apparently at 5 days average per move the black human + engine made a mistake and lost.
So there is room for the engine to improve.

Here is an example of an ICCF game from an ongoing qualifier: https://www.iccf.com/game?id=1531671 Apparently at 5 days average per move the black human + engine made a mistake and lost. So there is room for the engine to improve.

Naphthalin

#37

@tpr said in #33:

Also the fact that the TCEC super finals shows a winner proves there is still room for progress.
Exactly -- hence SF progress is measured from games starting with an UHO book aiming at a +1.00 exit eval. There is progress, but games from balanced openings aren't suited to show this progress anymore (which is also why the CCRL rating list is basically a coin toss at the top).

The 50 openings imposed by a human panel are played twice: once with white and once with black. Some openings end 0.5 - 0.5, so the starting position is a draw. Some openings end 1 - 1, so the starting position was a win for one side. However, some end 1.5 - 0.5, that means either the starting position was a draw and one engine erred and lost, or the starting position was a win and one engine failed to win.
The openings in TCEC are very, very carefully chosen to be as close to +1.00 (50% W/D) as possible in order to allow 1.5-0.5 results as a consequence of one engine understanding/playing the position a bit better than the opponent. Even a +0.80 position would produce almost exclusively double draws, with the rare exception where both engines underestimate the position.

Progress will also come once the 8-men endgame table base is available in lieu of the present 7-men endgame table base.
Running SF with or without 7-men EGTBs already now makes a difference of <5 Elo (again, starting from UHO openings and not regular startpos), going from 7-men to 8-men probably makes even less of a difference. Plus, if you start from balanced openings, you won't even get any endgame positions which aren't just a dead draw in the first place.

@tpr said in #33: > Also the fact that the TCEC super finals shows a winner proves there is still room for progress. Exactly -- hence SF progress is measured from games starting with an UHO book aiming at a +1.00 exit eval. There is progress, but games from balanced openings aren't suited to show this progress anymore (which is also why the CCRL rating list is basically a coin toss at the top). > The 50 openings imposed by a human panel are played twice: once with white and once with black. Some openings end 0.5 - 0.5, so the starting position is a draw. Some openings end 1 - 1, so the starting position was a win for one side. However, some end 1.5 - 0.5, that means either the starting position was a draw and one engine erred and lost, or the starting position was a win and one engine failed to win. The openings in TCEC are very, very carefully chosen to be as close to +1.00 (50% W/D) as possible in order to allow 1.5-0.5 results as a consequence of one engine understanding/playing the position a bit better than the opponent. Even a +0.80 position would produce almost exclusively double draws, with the rare exception where both engines underestimate the position. > Progress will also come once the 8-men endgame table base is available in lieu of the present 7-men endgame table base. Running SF with or without 7-men EGTBs already now makes a difference of <5 Elo (again, starting from UHO openings and not regular startpos), going from 7-men to 8-men probably makes even less of a difference. Plus, if you start from balanced openings, you won't even get any endgame positions which aren't just a dead draw in the first place.

tpr

#38

#37
"going from 7-men to 8-men probably makes even less of a difference."

It will make a huge difference, as 7 men cannot be materially balanced and 8 men can.

"if you start from balanced openings, you won't even get any endgame positions which aren't just a dead draw in the first place" * Well in ICCF human + engine from the initial position and 5 days/move there are still decisive games in qualifiers, semifinals and candidates. Only in the finals have the strongest ICCF (grand)masters + engines at 5 days/move now reached perfect play with all draws.

#37 "going from 7-men to 8-men probably makes even less of a difference." * It will make a huge difference, as 7 men cannot be materially balanced and 8 men can. "if you start from balanced openings, you won't even get any endgame positions which aren't just a dead draw in the first place" * Well in ICCF human + engine from the initial position and 5 days/move there are still decisive games in qualifiers, semifinals and candidates. Only in the finals have the strongest ICCF (grand)masters + engines at 5 days/move now reached perfect play with all draws.

corvusmellori

#39

@tpr said in #36:

Here is an example of an ICCF game from an ongoing qualifier:
www.iccf.com/game?id=1531671
Apparently at 5 days average per move the black human + engine made a mistake and lost.
So there is room for the engine to improve.
No, there is room for the human to stop trying to doubt the engine, which is way, way better than the human can ever be.
https://lichess.org/6637Cdtf#37

This is a completely inexcusable single-move blunder. The engine - even the very weak lichess webfish - immediately and at extremely low depth points out that QxN is the only move, and NxN is a 600cp blunder.

This isn't one engine outplaying another. This is a human interfering with the engine and thus causing a loss.

@tpr said in #36: > Here is an example of an ICCF game from an ongoing qualifier: > www.iccf.com/game?id=1531671 > Apparently at 5 days average per move the black human + engine made a mistake and lost. > So there is room for the engine to improve. No, there is room for the human to stop trying to doubt the engine, which is way, way better than the human can ever be. https://lichess.org/6637Cdtf#37 This is a completely inexcusable single-move blunder. The engine - even the very weak lichess webfish - immediately and at extremely low depth points out that QxN is the only move, and NxN is a 600cp blunder. This isn't one engine outplaying another. This is a human interfering with the engine and thus causing a loss.

Naphthalin

#40

The only thing your ICCF example proves is that humans still make mistakes, and the one in the position is a blunder within fractions of a second. It would be a (mildly) interesting experiment to join ICCF and just run SF17.1 for 3min on each move and get 100% draws this way, but I personally don't know anyone who could be bothered to do that.

About the TB8: it's of course speculation, but your claim "it will make a huge difference" is based on nothing beyond your personal belief, disregarding that the gain of using TBs up to 7 compared to not using TBs has declined to almost nothing for recent engines. The only thing your ICCF example proves is that humans still make mistakes, and the one in the position is a blunder within fractions of a second. It would be a (mildly) interesting experiment to join ICCF and just run SF17.1 for 3min on each move and get 100% draws this way, but I personally don't know anyone who could be bothered to do that.

This topic has been archived and can no longer be replied to.