To evaluate player's strength it would be more accurate to see what they did in unusual positions. AI can select what is an unusual position throughout in his big data and accumulate ACPL of players' choise in those cases.
So when my bullet game has 100ACPL, its like WC games back in the day? :D
Well I was thinking of that. The article does mention that some world championships back in the days had more ACPL than some bullet games played by club players nowadays. Yet I doubt the average club player would be able to beat Steinitz.
This begs the question : what does ACPL actually measure in terms of player strenght?
This begs the question : what does ACPL actually measure in terms of player strenght?
Amazing work absolutely amazing work, Like the best read on Lichess in a good year.
I would like to know who is the most accurate Player? This I would find more interesting than to know what was the most accurate game (which is likely a draw).
I think by measuring the average accuracy, it should be possible to plot a graph where you can read off the elo strength of each player. For instance, I expect a GM to play more accurate on average than an average club player. There is certainly a strong correlation with strength and accuracy, but it would be interesting to know how reliably strong the correlation is.
The average mistake a player does is certainly also dependent on style. Wild positions will deviate more, short dull games, rich of exchanges and a quick draw will not have many blunders in them, mainly because alternative moves are often almost equally good. But in wild positions the difference in the evaluation of possible candidate moves can deviate quite a lot.
For example: the second best moves in wild positions can be much weaker than in dull positions.
I think by measuring the average accuracy, it should be possible to plot a graph where you can read off the elo strength of each player. For instance, I expect a GM to play more accurate on average than an average club player. There is certainly a strong correlation with strength and accuracy, but it would be interesting to know how reliably strong the correlation is.
The average mistake a player does is certainly also dependent on style. Wild positions will deviate more, short dull games, rich of exchanges and a quick draw will not have many blunders in them, mainly because alternative moves are often almost equally good. But in wild positions the difference in the evaluation of possible candidate moves can deviate quite a lot.
For example: the second best moves in wild positions can be much weaker than in dull positions.
Even Tal wasn’t bad.
This was a very interesting piece - thank you! And, Capablanca's accuracy is the stand-out for me, looking across the historical trend. Given that he was known for "sight of the board", maybe his deep intuitive skills were akin to the AI approach, giving accuracy in a different (deeper?) way than brute-force calculation?
Guess that best games in all history were Capablanca/Alekhine matches.
Empateeeeeeee mr chese mgnus
Given that both competitors probably had a very intense preparation with engines, and they are so accurate, I wonder if they would be able to draw 14 games against Stockfish if they wanted to. Or would Stockfish slowly build up an advantage for every game and end up winning ?
This topic has been archived and can no longer be replied to.