Just to point out the obvious, which has probably been pointed out before in the previous 165 posts:
There is no such thing as a 4000 rating in chess, and there never will be. Chess has a wide draw margin and the best possible rating is probably somewhat below 3600. That is, a current 2800 player could draw at least one game out of fifty against God himself. Or, a perfect engine, whichever you're more likely to have faith in.
I know, USCF and FIDE are different algorithms, right? I'm logging off in 5 min btw.
Same basic algorithm, slightly different parameters.
Yeah I guess, I just wish that these are the questions that scientist can try and answer not whats up in Mars, its fucking rocks. Move on!
About 99% of these discussion are already answered in talkchess.com( computer chess programmers forum).
You dont know those cos you dont google for your knowledge.
For example, Engine A is 3000 rated , a new experimental patch on engine A1 has unknown rating but the programmer feel that it is +5 elo.So he have to test 40000 games to reduce statistical error +/- 1.8 elo.
P.S, the experts in there will never talk about non-sense facts like 10000 elo etc like here.
Chess has a wide draw margin and the best possible rating is probably somewhat below 3600
Considering alpha zero had a performance rating of what, ~3500, that's hard to believe. You think a 32 man EGTB would score less than 64% against AZ? I don't know what old calculation that guess is based off of, but it certainly seems wrong.
Based on the draw ratio increasing, I've seen estimates put a perfect player well over 4000.
Just to point out the obvious, which has probably been pointed out before in the previous 165 posts:
There is no such thing as a 4000 rating in chess, and there never will be. Chess has a wide draw margin and the best possible rating is probably somewhat below 3600. That is, a current 2800 player could draw at least one game out of fifty against God himself. Or, a perfect engine, whichever you're more likely to have faith in.
I dont understand that. Probably because how they figure ratings is too complicated for me. But at one time there was no such thing as a 3000 rating either. But then computers started regularly beating other computers and people so I assume they had to give those computers ratings higher than the opponents they were beating.
If the current highest rating is around 3500 or 3600 how do we know there can never be a 4000 rating? What if a new computer program beats a 3600 9 out of 10 times? Wouldn't the rating have to go up?
What if a new computer program beats a 3600 9 out of 10 times? Wouldn't the rating have to go up?
Weak amateur human games end in a draw maybe 1 time out of 100. As the players (human or computer) get stronger, the draw rate increases. From memory (so it may be wrong) Carlsen draws about 60% of his games. Top engines can draw each other >80% of the games.
He's saying current programs are pretty close to perfect, so even a solved version of the game (a database that merely looks up whether a move is winning, draw, or losing) wouldn't be able to score 9 out of 10 against current engines.
I think my question is perfectly reasonable. At a certain point, the players are so good that no matter what the difference in rating is, the game will be very hard.
There will not be a big difference in rating in the first place. To win rating you need win games and most games are draw already.
What if a new computer program beats a 3600 9 out of 10 times? Wouldn't the rating have to go up?
Weak amateur human games end in a draw maybe 1 time out of 100. As the players (human or computer) get stronger, the draw rate increases. From memory (so it may be wrong) Carlsen draws about 60% of his games. Top engines can draw each other >80% of the games.
He's saying current programs are pretty close to perfect, so even a solved version of the game (a database that merely looks up whether a move is winning, draw, or losing) wouldn't be able to score 9 out of 10 against current engines.
How do we know current programs are pretty close to perfect? At one time I'm sure people thought 3000 was unattainable or near perfect since it was so much higher than anyone had ever achieved.
It just seems to me that since computers are in their infancy, it might be premature to say that 4000 will never happen especially since it's only about 500 away from current abilities. I'm sure top computers do draw most of their games but I think that's because they are so similar in ability. But if we took the top computer from 1990 and put it up against the top computer from 2018 I'll bet there would be fewer draws because their abilities are farther apart. I'm sure the top computers from 2118 would most likely draw every game because they will probably be similar in ability too. But I'll bet the top computer from 2118 against the top computer now would have far fewer draws.
How do we know current programs are pretty close to perfect?
I'm not making that argument, smyslovfan is, on behalf of whatever engine programmer he read, or maths re: draw rates suggested.
I tend to agree that draw rates among engines that use the same method to find moves is misleading. Alpha Zero, for example, didn't follow the usual pattern of scoring slightly better with white.
As for technology advancing, yes, we can certainly use our imaginations, but I'm not sure how predictive imaginations are. The people arguing for certain rating limits outside of chess.com are well aware of (or at least they should be) the history of computer chess, and of technology in general.
IMO a 3600 ceiling is too low, but in any case it's probably not too far off.
And I'm willing to consider 4000 is impossible for practical reasons, the same way a 32 man EGTB is not physically impossible, but for practical reasons it may as well be impossible.
I think 4000 might be achievable or close to achievable today if someone wasted the resources of a modern supercomputer for a while. The fastest supercomputer is about a thousand times faster than google's TPUs and would cost a few dollars per second if the Chinese government were to sell time on it, I would say.
This would require rewriting a top engine's code in an efficient, massively parallel form, which is non-trivial.
I think 4000 might be achievable or close to achievable today if someone wasted the resources of a modern supercomputer for a while. The fastest supercomputer is about a thousand times faster than google's TPUs.
This would require rewriting a top engine's code in an efficient, massively parallel form, which is non-trivial.
Wasted?
No commercial or strategic value. The actual applications include life sciences, weather forecasting, industrial design and pharmaceutical research.
If he wasn't constipated yeah
What if a new computer program beats a 3600 9 out of 10 times? Wouldn't the rating have to go up?
Weak amateur human games end in a draw maybe 1 time out of 100. As the players (human or computer) get stronger, the draw rate increases. From memory (so it may be wrong) Carlsen draws about 60% of his games. Top engines can draw each other >80% of the games.
He's saying current programs are pretty close to perfect, so even a solved version of the game (a database that merely looks up whether a move is winning, draw, or losing) wouldn't be able to score 9 out of 10 against current engines.
As I have pointed out before, rating differences still lead to the same scores by definition. The difference is that the stronger player loses less and thus doesn't need to win as much to achieve a certain rating difference.
Some top computer matches have seen the extreme version of this, where all the wins are by one side.
Just to point out the obvious, which has probably been pointed out before in the previous 165 posts:
There is no such thing as a 4000 rating in chess, and there never will be. Chess has a wide draw margin and the best possible rating is probably somewhat below 3600. That is, a current 2800 player could draw at least one game out of fifty against God himself. Or, a perfect engine, whichever you're more likely to have faith in.
I dont understand that. Probably because how they figure ratings is too complicated for me. But at one time there was no such thing as a 3000 rating either. But then computers started regularly beating other computers and people so I assume they had to give those computers ratings higher than the opponents they were beating.
If the current highest rating is around 3500 or 3600 how do we know there can never be a 4000 rating? What if a new computer program beats a 3600 9 out of 10 times? Wouldn't the rating have to go up?
I'm confused to, if a a 3500 rated engine wins against another 3500, it will keep going up. Even if it wins against a 3000, it will still go up. The ratings do change after each game, even in draws. But as I pointed out CHESS MAY NOT BE A DRAW EVEN IF PLAYED PERFECTLY. Therefore we cannot assume that even engines playing PERFECT moves all the time would result in a draw. Therefore, one side may still win and the ratings will still change.
What if a new computer program beats a 3600 9 out of 10 times? Wouldn't the rating have to go up?
Weak amateur human games end in a draw maybe 1 time out of 100. As the players (human or computer) get stronger, the draw rate increases. From memory (so it may be wrong) Carlsen draws about 60% of his games. Top engines can draw each other >80% of the games.
He's saying current programs are pretty close to perfect, so even a solved version of the game (a database that merely looks up whether a move is winning, draw, or losing) wouldn't be able to score 9 out of 10 against current engines.
How do we know current programs are pretty close to perfect? At one time I'm sure people thought 3000 was unattainable or near perfect since it was so much higher than anyone had ever achieved.
It just seems to me that since computers are in their infancy, it might be premature to say that 4000 will never happen especially since it's only about 500 away from current abilities. I'm sure top computers do draw most of their games but I think that's because they are so similar in ability. But if we took the top computer from 1990 and put it up against the top computer from 2018 I'll bet there would be fewer draws because their abilities are farther apart. I'm sure the top computers from 2118 would most likely draw every game because they will probably be similar in ability too. But I'll bet the top computer from 2118 against the top computer now would have far fewer draws.
Chess programs are NOT close to perfect. They cannot even solve simple mates in 67 moves, like this one:
Just because Stockfish can memorize a million openings, doesn't mean it can calculate logic, patterns, triangulated positions, super advanced endgames..etc, which a PERFECT engine could. Engines can't even solve some 13 move puzzles. So much for calculating MILLIONS of moves ahead LMAO.
Just to point out the obvious, which has probably been pointed out before in the previous 165 posts:
There is no such thing as a 4000 rating in chess, and there never will be. Chess has a wide draw margin and the best possible rating is probably somewhat below 3600. That is, a current 2800 player could draw at least one game out of fifty against God himself. Or, a perfect engine, whichever you're more likely to have faith in.
Many things that are "obvious" are false. This could easily be one of them.
The first serious attempt to build an AI to play chess achieved a 64% score against an engine which is the pinnacle of many decades of development. In achieving this score AlphaZero didn't lose at all. What you are missing is that what matters to the advance in chess standards is the plus score it achieved, not that it managed to do so while avoiding even a single loss.
This lack of any draws in this match obviously (I use the word justifiably!) does NOT imply that there can be no stronger chess player. What it strongly suggests is that for future matches between top players the trend will continue that the stronger player will not lose very much, if at all. This is NOT the same as (indeed it bears no resemblance to) saying that the stronger player will not WIN at all.
Surely this should be clear after multiple posts on the point?
I know, USCF and FIDE are different algorithms, right? I'm logging off in 5 min btw.