Carlsen or Nepomniachtchi – Who will win?

[AlexChess]

But...

If engine fans here think that engines rated 3000+ ELO (up to 3900 ELO+ for some ratings lists) cannot easily beat Magnus rated 2900 ELO, what are we doing here???

What is the REAL Stockfish 14.1-dev ELO playing 2 hours / 40 moves + 1 hour / 20 moves + 15 mins + 30 secs / ALL?

Of course it depends on definition of REAL Elo; there are no humans rated high enough to define a human rating for the top engines now, but we can estimate pretty well. The SSDF rating list uses 40/2 hours so that is our starting point. I think the list was accurate in human terms for 2800 level engines 20 years ago, but human ratings are about 100 elo stricter now based on studies of matching human with engine moves, so we subtract 100 from their ratings. They have Stockfish 13 at 3573 on 8 cores, so that becomes 3473. Presumably SF 14.1 on 8 cores would be about 3500 on that scale. But it is known that engine vs engine ratings exaggerate rating differences in human terms, though perhaps not by a huge percentage. So my best estimate for a human FIDE rating for Stockfish 14.1 on 8 cores at 40/2 hours would be roughly 3400. However this isn't testable even in theory, because the rating formula isn't very reliable beyond about 300 elo difference, the results would depend primarily on whether someone can create a book to avoid draws when playing Black. My interpretation of a 3400 rating would be that the engine could win a match from Carlsen giving him pawn and move (i.e. f7 pawn) odds, but Carlsen should win at pawn and two moves or two non-edge White pawns. That's based on b or c pawn being worth about 300 elo at this level, f or g pawn about 400, and playing first about 50 (another 100 for a full extra move). These are roughly the values I get from computer vs computer odds games. Of course at faster time controls the engine ratings would be much higher.

Thank you!
Very interesting explanation

[Uri Blass]

But...

If engine fans here think that engines rated 3000+ ELO (up to 3900 ELO+ for some ratings lists) cannot easily beat Magnus rated 2900 ELO, what are we doing here???

What is the REAL Stockfish 14.1-dev ELO playing 2 hours / 40 moves + 1 hour / 20 moves + 15 mins + 30 secs / ALL?

Most of US do NOT believe that that Magnus rated around 2860 can beat Stockfisk or Komodo Dragon, but we believe those engines would have a tough time beating Carlsen at 2 pawns odds at standard time control, but in T/C of 20 Minutes, Komodo Dragon will have an edge in 10 games.

I think the question is not if humans can beat stockfish or komodo dragon but if humans can get draws against them.

Comp-comp rating say nothing about the rating of computers against humans.

I see that stockfish is still stupid and does not understand that rook is more than a bishop and if I remove Bishop c1 I get higher advantage for black relative to removing rook a1.

It means that maybe it is possible to get a draw against stockfish by sacrificing a bishop in the opening because stockfish will not understand that it is better to get later in a game a rook for a bishop and may fail to get progress and the human may create a fortress that stockfish does not understand later to get a draw against stockfish.

I hate this NNUE that cause this stupid evaluation of stockfish and I believe that it is possible that humans can take advantage of the stupid evaluation in order to beat stockfish in a match with draw odds.
I have no proof for it but I suspect that it is possible that the combination of the ability of GM's and finding weaknesses in stockfish that even I can see can allow GM's and maybe not even top GM's to get a draw against stockfish in most of the cases.

[Cornfed]

I see that stockfish is still stupid and does not understand that rook is more than a bishop and if I remove Bishop c1 I get higher advantage for black relative to removing rook a1.

Actually, I would contend that you are be wrong there!

In the starting position (I presume you are referring to) and thru the earlier parts of the game. The Bc1 is usually worth more than either of the rooks. "Worth" is relative and each pieces 'worth' changes throughout the game. Rooks need a while for space to be cleared out for them, in order for their 'worth' to be felt.

[lkaufman]

I see that stockfish is still stupid and does not understand that rook is more than a bishop and if I remove Bishop c1 I get higher advantage for black relative to removing rook a1.

Actually, I would contend that you are be wrong there!

In the starting position (I presume you are referring to) and thru the earlier parts of the game. The Bc1 is usually worth more than either of the rooks. "Worth" is relative and each pieces 'worth' changes throughout the game. Rooks need a while for space to be cleared out for them, in order for their 'worth' to be felt.

No, Uri is right here. Many tests I've run show that even it the initial position, a rook is much more valuable than a bishop, though the gap is much closer to one pawn than to the nominal two pawns difference. Removing Bc1 and Ra8 is clearly winning for White; even removing Ra1 and Bc8 is clearly winning for Black. Just play it out with any top engines many times and you will confirm this.

[Uri Blass]

I see that if I remove pieces of both sides stockfish's evaluation is relatively better and it does not claim that the value of bishop is bigger in this case.

When I remove a rook for one side and a bishop for the other side stockfish evaluate correctly the better side when the main problem is evaluating unbalanced positions when one side has a big advantage.

It may be interesting to have a rating list when you remove part of both pieces.

For example the following position
[fen]rn1qkbnr/pppp1ppp/8/8/8/8/PPPPPPPP/1NBQKBNR w Kkq - 0 1 [/fen]

I wonder if stockfish is right when it considers it to be equal or old engines are right when they usually consider black to be better.

I am surprised that stockfish10 that does not use NNUE consider white to be better.

[Cornfed]

I see that stockfish is still stupid and does not understand that rook is more than a bishop and if I remove Bishop c1 I get higher advantage for black relative to removing rook a1.

Actually, I would contend that you are be wrong there!

In the starting position (I presume you are referring to) and thru the earlier parts of the game. The Bc1 is usually worth more than either of the rooks. "Worth" is relative and each pieces 'worth' changes throughout the game. Rooks need a while for space to be cleared out for them, in order for their 'worth' to be felt.

No, Uri is right here. Many tests I've run show that even it the initial position, a rook is much more valuable than a bishop, though the gap is much closer to one pawn than to the nominal two pawns difference. Removing Bc1 and Ra8 is clearly winning for White; even removing Ra1 and Bc8 is clearly winning for Black. Just play it out with any top engines many times and you will confirm this.

Sorry, I should be more clear - I was actually talking about from a human player prospective, not an engines of the starting position.

I wonder if doing what following this idea of Uri's but with removing the Queen vs removing the 2 Rooks looks like from an engine perspective. I would presume the queens worth is increased a tad just because it can get active earlier and participate in the game while Rooks...have to wait their turn more often than not to express their worth.

[lkaufman]

I see that stockfish is still stupid and does not understand that rook is more than a bishop and if I remove Bishop c1 I get higher advantage for black relative to removing rook a1.

Actually, I would contend that you are be wrong there!

In the starting position (I presume you are referring to) and thru the earlier parts of the game. The Bc1 is usually worth more than either of the rooks. "Worth" is relative and each pieces 'worth' changes throughout the game. Rooks need a while for space to be cleared out for them, in order for their 'worth' to be felt.

No, Uri is right here. Many tests I've run show that even it the initial position, a rook is much more valuable than a bishop, though the gap is much closer to one pawn than to the nominal two pawns difference. Removing Bc1 and Ra8 is clearly winning for White; even removing Ra1 and Bc8 is clearly winning for Black. Just play it out with any top engines many times and you will confirm this.

Sorry, I should be more clear - I was actually talking about from a human player prospective, not an engines of the starting position.

I wonder if doing what following this idea of Uri's but with removing the Queen vs removing the 2 Rooks looks like from an engine perspective. I would presume the queens worth is increased a tad just because it can get active earlier and participate in the game while Rooks...have to wait their turn more often than not to express their worth.

In the initial position, the queen is worth clearly more than the two rooks, regardless of whether you go by engine eval, by GM eval, by engine playouts, or by human playouts (Dzindzi used to bet on this!). I think it has more to do with the presence of so many other pawns and pieces on the board than it does on the precise start positions of the queen and rooks, although both factors play a role. Regarding rook for bishop, I don't believe that the result would be different if top GMs played it out instead of engines; I don't know about the result at amateur level. These days NNUE engines playing hyper-bullet are a pretty good substitute for top GMs playing tournament level chess.

[MartinBryant]

This report yesterday suggests that Carlsen may not want to defend his title next time anyway!
https://www.theguardian.com/sport/2021/ ... motivation

[carldaman]

I read that a bit differently. It sounds more to me that Carlsen may be making a play here to change the World Championship and Candidates format, and perhaps even to return to the former chaotic era of hand-picked opponents, all in the name of 'feeling motivated', of course.

Lately, the FIDE Candidates qualification system has left a lot to be desired in that several of strongest players don't get to qualify, while other weaker players can somehow slip in. For example, Wesley So is one of Carlsen's toughest opponents, but he did not qualify for the last Candidates, mainly because of the poor selection criteria.

Maybe FIDE will pay attention to these comments and get their act together before another Kasparov vs Short scenario happens again.

[Chessqueen]

I see that if I remove pieces of both sides stockfish's evaluation is relatively better and it does not claim that the value of bishop is bigger in this case.

When I remove a rook for one side and a bishop for the other side stockfish evaluate correctly the better side when the main problem is evaluating unbalanced positions when one side has a big advantage.

It may be interesting to have a rating list when you remove part of both pieces.

For example the following position
[fen]rn1qkbnr/pppp1ppp/8/8/8/8/PPPPPPPP/1NBQKBNR w Kkq - 0 1 [/fen]

I wonder if stockfish is right when it considers it to be equal or old engines are right when they usually consider black to be better.

I am surprised that stockfish10 that does not use NNUE consider white to be better.

Also the first move is worth almost a pawn, if I told you that Stocfish can give the e2 pawn odds and still draw to Komodo Dragon, most people here would say NO that is impossible, but to move first is worth a pawn and here it is

[pgn][Event "Computer chess game"]
[Site "MININT-UB2PIMJ"]
[Date "2021.12.16"]
[Round "?"]
[White "Stockfish_14_x64_bmi2"]
[Black "Dragon-2.5-64bit-avx2"]
[Result "1/2-1/2"]
[BlackElo "3600"]
[Time "08:08:36"]
[WhiteElo "3640"]
[TimeControl "600+3"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPP1PPP/RNBQKBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "100"]
[WhiteType "program"]
[BlackType "program"]

1. d4 Nf6 2. Nf3 d5 3. c4 Be6 4. Qb3 dxc4 5. Qxb7 Bd5 6. Qb5+ c6 7. Qa4
Bxf3 8. gxf3 Qxd4 9. Nc3 e6 10. Be3 Qd6 11. Bxc4 Be7 12. Ne4 Nxe4 13. fxe4
O-O 14. Be2 Nd7 15. Rd1 Qc7 16. O-O Bd6 17. Qc2 Bxh2+ 18. Kg2 Be5 19. f4
Bf6 20. e5 Be7 21. Rh1 g6 22. Rh3 Rfd8 23. Rdh1 Nf8 24. Bc4 Rab8 25. b3 Rb4
26. Kf3 h5 27. Rg3 h4 28. Rg4 c5 29. Qh2 Rxc4 30. bxc4 f5 31. Rxh4 Bxh4 32.
Qxh4 Qb7+ 33. Kf2 Qb2+ 34. Kf3 Rd3 35. Qh8+ Kf7 36. Qf6+ Ke8 37. Rh8 Rxe3+
38. Kxe3 Qd4+ 39. Kf3 Qe4+ 40. Kf2 Qxf4+ 41. Kg2 Qd2+ 42. Kf3 Qd1+ 43. Kf2
Qc2+ 44. Kf3 Qd3+ 45. Kg2 Qe2+ 46. Kh3 Qf3+ 47. Kh2 Qf2+ 48. Kh3 Qf3+ 49.
Kh2 Qf2+ 50. Kh3 Qf3+ {3-fold repetition} 1/2-1/2[/pgn]