yeah it will be able to calculate more moves ahead
A 3000 could easily beat a 2000, but could a 4000 easily beat a 3000?
Sort:
Nov 20, 2021
0
#546
Isn't this kind of a silly question? Ratings are based on win probability. A rating difference of 1000 means roughly a 99.7% chance of the higher rated player winning. If the 4000 player didn't have a 99.7% chance of winning, then they wouldn't be rated 4000. It's as simple as that.
If there is such a thing like 4k and 3k rated players/engines at the same time, then certainly 4k one crushes 3k one since otherwise by making many draws, it cannot maintain its own rating.
Some people here saying that there will be many draws, no, many draw happens among similar rating engines, it is mathematically impossible to have two players of huge gap and they make many draws.
Some people here saying that there will be many draws, no, many draw happens among similar rating engines, it is mathematically impossible to have two players of huge gap and they make many draws.
Nov 20, 2021
0
#548
Yes, presuming such a player could exist (non-trivial), the 4000 would need to achieve around 99.97% to justify his rating. Doing so would require effort (top performance always does). While such dominance of such a strong player might seem incomprehensible, bear in mind that the present top engines are already capable of getting close to 100% against the top humans.
A lower (but dominant) score by a 4000 against a 3000 could be achieved with less effort (eg in a simultaneous display).
Nov 20, 2021
0
#549
Exactly. Anyone who thinks there would be a bunch of draws has a fundamental misunderstanding of what ratings are supposed to measure. A ratings difference of 1000 BY DEFINITION means a 99.7% chance of the higher rated player winning. If they don't win that much, then the rating difference isn't 1000... BY DEFINITION!
The question is can you get to 4000 elo?
Stockfish's 3800+ elo now is based on imbalanced openings. Stockfish 12 and later versions will simply draw Stockfish 14.1 in every start position game due to NNUE and insane search.
Nov 20, 2021
0
#551
If it was "every" they would technically have the same rating for chess. Forced openings are a different game.
A variant of chess with a specified distribution of forced openings has its own separate Elo scale.
Nov 20, 2021
0
#553
The math is simple. A player rated 1000 points higher than another should score 99.7 out of 100 points.
In chess, a 4000 rating is probably impossible. But in a game that is not a theoretical draw it doesn’t matter whether the lower rated player is 300 or 3000. A 1000 point difference requires the higher rated player to score 99.7/100.
SmyslovFan and NikkiLikeChikki both understand the math.
Nov 21, 2021
0
#556
andrewpalmer123 wrote:
yeah the higher rating the more likely the draw
It is worth noting that this is a separate mode of variation to the expected score.
- Expected score is a monotone increasing function of difference in rating.
- Probability of a draw is a monotone increasing function of rating for a given difference in rating
To put it another way, variance of result is a monotone decreasing function of rating for a given difference in rating.
Dec 27, 2022
0
#557
There will never be anywhere close to 4000, unless chess is solved, and even then they may be around 3700.
Let’s take stockfish (maybe 3600-3700?) against magnus carlsen(2800) as an example. Stockfish easily beats Magnus every time. In order for one player to be 1000 points above another, there has to be a big difference between them, otherwise they would just be the same rating.
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.
If the ratings were accurate to playing strength, then a 4000 would mop the floor with a 3000 as if the 3000 were nothing at all.
Heck, even a 500-point difference is monstrous.
Take Stockfish 15 NNUE (3500+ Elo) and pit it against Crafty 25.2 (3000 Elo). See how it turns out.
I've read here that there are no enough human vs engine games to accurately compare their ratings. That's not true: in ICC Madchess v1.4 played thousands of games vs human players and got 2100 classical rating and 2400 blitz. Other engines too. Also in lichess.
That same engine in CCRL list is 2200 rating. That proves CCRL rating can be directly compared to human rating (give or take some points, but pretty accurate).
yes, 4000 ELO will deferentially crush the 3000 ELO player like stockfish 14.1 on 250 treads (3862) will crush stockfish 14.1(3689) itself.