- Blind mode tutorial
lichess.org
Donate

Training Game #24 with Turbo Eval

#2

"Obviously, the way chess works is once you reach a new height and get all excited, it sends you to the shadow realm of a losing streak"
I have had many experiences with this one, both online and over the board. Just a month earlier, I reached my all-time peak of 2836 bullet on Lichess (on my main account), but then dropped back to 2706 within a week! The same has happened to my FIDE rating. I reached my peak in December of 2024, didn't play in January, then I dropped 112 points in February! Nature of chess I suppose.

I am happy that you are back to writing blog posts, and I hope you will continue to do so.

"Obviously, the way chess works is once you reach a new height and get all excited, it sends you to the shadow realm of a losing streak" I have had many experiences with this one, both online and over the board. Just a month earlier, I reached my all-time peak of 2836 bullet on Lichess (on my main account), but then dropped back to 2706 within a week! The same has happened to my FIDE rating. I reached my peak in December of 2024, didn't play in January, then I dropped 112 points in February! Nature of chess I suppose. I am happy that you are back to writing blog posts, and I hope you will continue to do so.
#3

Maybe someone can explain the reasoning behind the chess.com bot ratings? I can beat a 2000 bot consistently, but when it comes to playing a 2000 human thats another story. Why are the bot's on chess.com rated so high? No one has ever answered my question. I have messaged Levy and chess.com and no one will weight in on the topic. Thanks for listening.

Maybe someone can explain the reasoning behind the chess.com bot ratings? I can beat a 2000 bot consistently, but when it comes to playing a 2000 human thats another story. Why are the bot's on chess.com rated so high? No one has ever answered my question. I have messaged Levy and chess.com and no one will weight in on the topic. Thanks for listening.
#4

@Sethshouse said in #3:

Maybe someone can explain the reasoning behind the chess.com bot ratings? I can beat a 2000 bot consistently, but when it comes to playing a 2000 human thats another story. Why are the bot's on chess.com rated so high? No one has ever answered my question. I have messaged Levy and chess.com and no one will weight in on the topic. Thanks for listening.

I have an opinion about this. I think it's difficult to defeat 2000-rated human players compared to bots because any human above the rating of, let's say, 1200, plays with PURPOSE. Their moves always have a point. If we fail to grasp this idea, we might be in a worse position. But bots do not always play with a purpose. The rating disparity between bots is a result of taking into account the frequency of mistakes, but there is also a deeper factor at play here. While we win because they are programmed to make mistakes at certain moments, we get good positions because of the lack of purpose in the moves of a 2000-rated bot compared to a human. A bot never intentionally sets a trap or does those kinds of things (like pre-move sniping). A bot never thinks "If I place my knight on c4 by going Nh5-g7-e8-d6-c4, I can put pressure on my opponent's queenside pawns". So even though it might play Nh5-g7-e8, there's a chance it can go to g7 again because both Nd6 and Ng7 are equal according to evaluation. They are good tactically, but when it comes to playing with purpose, not so much. Now, Stockfish can find these sorts of Knight manoeuvres using deep calculations. But a 2000-rated bot is not programmed to calculate that deeply, and its policy net (intuition) is not so good. Hence, it would never find an idea such as the following because it's not playing with a purpose. If it were, it would see that it needs to attack B2 and get on with the manoeuvre. But it doesn't play with purpose, so it will not find this.

https://lichess.org/study/embed/3Hhi4Yxg/DdftXUNj#61

This idea is so aesthetically pleasing that it's still stuck in my head. I wonder how this kid found this idea during the game.

@Sethshouse said in #3: > Maybe someone can explain the reasoning behind the chess.com bot ratings? I can beat a 2000 bot consistently, but when it comes to playing a 2000 human thats another story. Why are the bot's on chess.com rated so high? No one has ever answered my question. I have messaged Levy and chess.com and no one will weight in on the topic. Thanks for listening. I have an opinion about this. I think it's difficult to defeat 2000-rated human players compared to bots because any human above the rating of, let's say, 1200, plays with PURPOSE. Their moves always have a point. If we fail to grasp this idea, we might be in a worse position. But bots do not always play with a purpose. The rating disparity between bots is a result of taking into account the frequency of mistakes, but there is also a deeper factor at play here. While we win because they are programmed to make mistakes at certain moments, we get good positions because of the lack of purpose in the moves of a 2000-rated bot compared to a human. A bot never intentionally sets a trap or does those kinds of things (like pre-move sniping). A bot never thinks "If I place my knight on c4 by going Nh5-g7-e8-d6-c4, I can put pressure on my opponent's queenside pawns". So even though it might play Nh5-g7-e8, there's a chance it can go to g7 again because both Nd6 and Ng7 are equal according to evaluation. They are good tactically, but when it comes to playing with purpose, not so much. Now, Stockfish can find these sorts of Knight manoeuvres using deep calculations. But a 2000-rated bot is not programmed to calculate that deeply, and its policy net (intuition) is not so good. Hence, it would never find an idea such as the following because it's not playing with a purpose. If it were, it would see that it needs to attack B2 and get on with the manoeuvre. But it doesn't play with purpose, so it will not find this. https://lichess.org/study/embed/3Hhi4Yxg/DdftXUNj#61 This idea is so aesthetically pleasing that it's still stuck in my head. I wonder how this kid found this idea during the game.
#5

Thanks for the input,
So should a 2000 bot be 2000 if it's not going as deep as a 2000 human? The point i am failing to make is, why over rate a bot?

@felew699 said in #4:

I have an opinion about this. I think it's difficult to defeat 2000-rated human players compared to bots because any human above the rating of, let's say, 1200, plays with PURPOSE. Their moves always have a point. If we fail to grasp this idea, we might be in a worse position. But bots do not always play with a purpose. The rating disparity between bots is a result of taking into account the frequency of mistakes, but there is also a deeper factor at play here. While we win because they are programmed to make mistakes at certain moments, we get good positions because of the lack of purpose in the moves of a 2000-rated bot compared to a human. A bot never intentionally sets a trap or does those kinds of things (like pre-move sniping). A bot never thinks "If I place my knight on c4 by going Nh5-g7-e8-d6-c4, I can put pressure on my opponent's queenside pawns". So even though it might play Nh5-g7-e8, there's a chance it can go to g7 again because both Nd6 and Ng7 are equal according to evaluation. They are good tactically, but when it comes to playing with purpose, not so much. Now, Stockfish can find these sorts of Knight manoeuvres using deep calculations. But a 2000-rated bot is not programmed to calculate that deeply, and its policy net (intuition) is not so good. Hence, it would never find an idea such as the following because it's not playing with a purpose. If it were, it would see that it needs to attack B2 and get on with the manoeuvre. But it doesn't play with purpose, so it will not find this.

This idea is so aesthetically pleasing that it's still stuck in my head. I wonder how this kid found this idea during the game.

Thanks for the input, So should a 2000 bot be 2000 if it's not going as deep as a 2000 human? The point i am failing to make is, why over rate a bot? @felew699 said in #4: > I have an opinion about this. I think it's difficult to defeat 2000-rated human players compared to bots because any human above the rating of, let's say, 1200, plays with PURPOSE. Their moves always have a point. If we fail to grasp this idea, we might be in a worse position. But bots do not always play with a purpose. The rating disparity between bots is a result of taking into account the frequency of mistakes, but there is also a deeper factor at play here. While we win because they are programmed to make mistakes at certain moments, we get good positions because of the lack of purpose in the moves of a 2000-rated bot compared to a human. A bot never intentionally sets a trap or does those kinds of things (like pre-move sniping). A bot never thinks "If I place my knight on c4 by going Nh5-g7-e8-d6-c4, I can put pressure on my opponent's queenside pawns". So even though it might play Nh5-g7-e8, there's a chance it can go to g7 again because both Nd6 and Ng7 are equal according to evaluation. They are good tactically, but when it comes to playing with purpose, not so much. Now, Stockfish can find these sorts of Knight manoeuvres using deep calculations. But a 2000-rated bot is not programmed to calculate that deeply, and its policy net (intuition) is not so good. Hence, it would never find an idea such as the following because it's not playing with a purpose. If it were, it would see that it needs to attack B2 and get on with the manoeuvre. But it doesn't play with purpose, so it will not find this. > > > > > > This idea is so aesthetically pleasing that it's still stuck in my head. I wonder how this kid found this idea during the game.
#6

@Sethshouse,
Thank you for making me think this deeper. To answer your question, a 2000 bot might go deeper, but they are easier to beat because they don't have a purpose in their moves. They are good tactically because they calculate better than a 2000, but when the position is not tactical, the human has the upper hand because of the difference in thinking patterns. It requires deeper calculations if you are going to make positional decisions based on calculations like a bot does. But a human makes positional decisions based on logical assumptions like "b2 is the weakest point, my rook should stay to guard the king, the bishop doesn't have direct access, so I should put my knight on c4". This kind of thinking doesn't require calculation, so a human is better in such positions where positional decisions have to be made. But bots make decisions based on calculation, so if their search tree doesn't reach a position with a Knight on c4, they won't get this idea. Stockfish calculates deeper, and its intuition is better, so its search tree reaches a position with a knight on c4. But for a 2000-rated bot, this is unlikely. That is why it's easier to beat a 2000-rated bot. They are better tactically, but they lack a logical thinking pattern. And also, their better tactical ability is sometimes hindered because they are programmed to make mistakes at certain moments.

Now, the reason for ranking a 2000-rated bot the same as a 2000-rated human is because of the factor considered in ranking them. I think the abilities are being considered here. A 2000-rated bot might be 2200 level tactically due to the calculating ability, but 1800 level positionally due to the lack of logical thinking ability. Hence, by taking the average, they have ranked the bot to be 2000.

There's also another possibility. As I said earlier, a human has the upper hand because of the logical thinking pattern. Because of this, a 1900 human might be able to beat a 2000-rated bot in 3/3 games. But it might take a 2100 bot to beat the same 2000-rated bot in 3/3 games. So again, taking the rating average, the bot might be ranked 2000.

@Sethshouse, Thank you for making me think this deeper. To answer your question, a 2000 bot might go deeper, but they are easier to beat because they don't have a purpose in their moves. They are good tactically because they calculate better than a 2000, but when the position is not tactical, the human has the upper hand because of the difference in thinking patterns. It requires deeper calculations if you are going to make positional decisions based on calculations like a bot does. But a human makes positional decisions based on logical assumptions like "b2 is the weakest point, my rook should stay to guard the king, the bishop doesn't have direct access, so I should put my knight on c4". This kind of thinking doesn't require calculation, so a human is better in such positions where positional decisions have to be made. But bots make decisions based on calculation, so if their search tree doesn't reach a position with a Knight on c4, they won't get this idea. Stockfish calculates deeper, and its intuition is better, so its search tree reaches a position with a knight on c4. But for a 2000-rated bot, this is unlikely. That is why it's easier to beat a 2000-rated bot. They are better tactically, but they lack a logical thinking pattern. And also, their better tactical ability is sometimes hindered because they are programmed to make mistakes at certain moments. Now, the reason for ranking a 2000-rated bot the same as a 2000-rated human is because of the factor considered in ranking them. I think the abilities are being considered here. A 2000-rated bot might be 2200 level tactically due to the calculating ability, but 1800 level positionally due to the lack of logical thinking ability. Hence, by taking the average, they have ranked the bot to be 2000. There's also another possibility. As I said earlier, a human has the upper hand because of the logical thinking pattern. Because of this, a 1900 human might be able to beat a 2000-rated bot in 3/3 games. But it might take a 2100 bot to beat the same 2000-rated bot in 3/3 games. So again, taking the rating average, the bot might be ranked 2000.