A 3000 could easily beat a 2000, but could a 4000 easily beat a 3000?
Sort:
NikkiLikeChikki
Sep 30, 2020
0
#321
@snudoo - since elo is calculated by win probability, then there should be no difference in theory. In practice there might be some differences, but they shouldn’t be egregious, and that would be a function of error in its calculation.
LineBreak wrote:
EndgameStudier wrote:
When a rating gets 3000 and above, it's all the same right? A 10,000 vs a 3000 would be a challenge.
Well . . . an elo is a measure of relative strength. So a theoretical 10,000 vs. a 3,000 would essentially mean that, no matter how strong the 3,000 is (incredibly strong!), the 10,000-elo player would destroy it.
Yes, the 3,000 would play excellent chess. But it would still get checkmated, or be forced to resign, every single game.
We'll probably see 4,000-elo chess engines/neural networks in the near future, as they're already hovering around the 3,500 mark (and we've only just begun to explore the capabilities of neural network engines).
Yes but the thing is chess only has so many "best moves" and if every best move is played in every position, then the game may be a draw no matter what. Let's say for example a 4000 could see the best move in every position, then it wouldn't matter if he was playing against a 10,000, you can't win against the best moves.
SNUDOO wrote:
I'm pretty sure the difference between a 2000 and a 1000 is much smaller than the difference between a 3000 and a 2000. Why? When you get higher, you have to consistently defeat everyone just to maintain your rating. Which means the higher you are, it is like exponentially stronger or something. I'm bad at graphs and statistics so I may be wrong.
Yes but a 1000 also is exponentially weaker, so doesn't that cancel out. He may fall for even the most basic tactical tricks even to a 1400, so a 2000 should easily win.
Sep 30, 2020
0
#324
Not sure there is a theoretical cap as win probability can go from 99% to 99.9% to 99.99% onto the limit. I think right now Magnus could theoretically win like one in 1500 games against Stockfish. That could go to 1 in 10,000 or to one in a jillion. Of course there is the very slim possibility that chess could be solved making a white win 100%, but that would break elo.
What has happened over the past 200 years is a trend that at the highest levels we are getting more and more draws. Sometimes at the current highest levels we are getting all draws, So maybe the limit will be around 4200?
LineBreak wrote:
EndgameStudier wrote:
Yes but the thing is chess only has so many "best moves" and if every best move is played in every position, then the game may be a draw no matter what. Let's say for example a 4000 could see the best move in every position, then it wouldn't matter if he was playing against a 10,000, you can't win against the best moves.
True, there's probably a cap somewhere. I don't know what exactly it'll be. 4,000 is a pretty safe guess, at the moment.
Though, there might come a time in the future when a hyper-advanced AI comes along and crushes a 4,000-elo neural engine with ease (which would then prove that the ceiling is even higher). Only time will tell!
Given that engines still have not mastered mate in 545s other than tablebase numbercrunching, I'd say 10,000
ponz111 wrote:
What has happened over the past 200 years is a trend that at the highest levels we are getting more and more draws. Sometimes at the current highest levels we are getting all draws, So maybe the limit will be around 4200?
You should start a topic about if chess is a draw or not. With best play of course. Assuming chess is a forced win for white (based on a ton of evidence), it would be an interesting topic.
There is! "True of false..chess is a draw with perfect play on both sides" thread.
Oct 2, 2020
0
#329
A point which is often misunderstood is the role of the number of draws. When two players are of similar rating - eg Stockfish and AlphaZero - the score they get is fairly near 50% (actually 57.45%). It is also mostly draws when they are strong players, fundamentally because the average number of blunders in a game is lower, and zero blunders gives a draw. It is also true that a world championship match is mostly draws, because it is between two players of very similar rating who are both very strong.
But the reasoning that because there are a very high percentage of draws between say Carlsen and Caruana (12 draws out of 12 games) this means Carlsen's and Caruana's rating of around 2850 is near the maximum possible must be invalid, because Stockfish and Alphazero have ratings near 3600 on the same scale and disprove the conclusion directly!
Since the reasoning is identical we are no more able to infer 3600 is near the maximum possible rating because two top computer players get a large fraction of draws. It is entirely plausible there is another player who can get over 90% against Alphazero, just like it must be possible for a (silicon) player to get over 90% against Carlsen (unless you believe he has some way of showing the rating of top engines is wrong!) The fact that this hypothetical 4000+ player does not exist now is not evidence against this, any more than the non-existence of Stockfish when Kasparov was having a tough match with Deep Blue version 1 proved a player with a rating over 3400 (600 more than Kasparov) could not exist.
Nah even 2300s are easier to beat if you are almost 3000. FIDE system is more accurate. In my opinion 2200 is not enough to be considered a master. It should be 2300 for master and 2600 or 2700 for grandmaster.
Elroch wrote:
A point which is often misunderstood is the role of the number of draws. When two players are of similar rating - eg Stockfish and AlphaZero - the score they get is fairly near 50% (actually 57.45%). It is also mostly draws when they are strong players, fundamentally because the average number of blunders in a game is lower, and zero blunders gives a draw. It is also true that a world championship match is mostly draws, because it is between two players of very similar rating who are both very strong.
But the reasoning that because there are a very high percentage of draws between say Carlsen and Caruana (12 draws out of 12 games) this means Carlsen's and Caruana's rating of around 2850 is near the maximum possible must be invalid, because Stockfish and Alphazero have ratings near 3600 on the same scale and disprove the conclusion directly!
Since the reasoning is identical we are no more able to infer 3600 is near the maximum possible rating because two top computer players get a large fraction of draws. It is entirely plausible there is another player who can get over 90% against Alphazero, just like it must be possible for a (silicon) player to get over 90% against Carlsen (unless you believe he has some way of showing the rating of top engines is wrong!) The fact that this hypothetical 4000+ player does not exist now is not evidence against this, any more than the non-existence of Stockfish when Kasparov was having a tough match with Deep Blue version 1 proved a player with a rating over 3400 (600 more than Kasparov) could not exist
Magnus won that 12 that 12th game, no reason to offer a draw. Only one pawn and piece off the board for each side, they should be rounded up right now and be forced to finish playing off that position!
Oct 4, 2020
0
#332
EndgameStudier a écrit :
kingsleyye wrote:
there is no evidence that a 3000 will beat a 2000 easily
I drew a master at least 3 times in casual games at chess clubs, and I'm only 1553 OTB, 1653 online.
He was drunk or he didn't pay attention.
The old master in my club may loose 2 in 20 against 2000 players....
I think they were trying, they weren't blitz games or anything.
Oct 5, 2020
0
#334
True but... TCEC provides an opening book and many of the openings it chooses are imbalanced in order to make for more interesting games.
But what is their match score? Is it around even?
Oct 6, 2020
0
#336
LineBreak wrote:
NikkiLikeChikki wrote:
True but... TCEC provides an opening book and many of the openings it chooses are imbalanced in order to make for more interesting games.
True, but the engines play both sides of each opening.
If the openings were the sole cause of wins/losses, then we'd see the both engines winning (or losing) in the same openings.
But so far, both Stockfish and Leela have scored victories in openings/defenses that their counterparts didn't. So they each must have strengths/weaknesses that the other doesn't possess (meaning: both engines still have room for improvement).
There is also some randomness in the results of individual games. So you would need to play a lot of games between them in an opening to see if there was a genuine advantage for one.
The real question is whether chess is a theoretically drawn game or not, if not, then that changes the whole story.
NikkiLikeChikki wrote:
True but... TCEC provides an opening book and many of the openings it chooses are imbalanced in order to make for more interesting games.
What I don't like about TCEC is that they don't let the engines play out their own openings they prefer. Many book positions lead to +1 or worse, I saw a garbage french line in there:
1. e4 e6 2. d4 d5 3. Nc3 Nf6 4. e5 Ng8
Like Ng8 what in the world?
Anyways, both engines won as white obviously in that matchup.
I think chess is a drawn game with perfect play. However, 3000 isn't perfect, so a 4000 (if that exists) would destroy a 3000. That 4000 rated player reached that rating somehow. If you study the rating system for example, when you gain 120 elo, you become about twice as good. This is a 1000 rating difference, that means the skill gap is veeery big.
Oct 7, 2020
0
#340
Theoretically, 4000 isn’t possible in chess because it is a draw. The highest theoretical rating is somewhere under 3600. The statistician Kenneth Regan postulated that 3571 is probably the highest an engine can achieve.*
But a 3571 would destroy a 2571 as easily as a 2571 would destroy a 1571.
——
* https://cse.buffalo.edu/~regan/papers/pdf/Reg12IPRs.pdf