It's a good book for its time, but computer chess has advanced so dramatically that much of the practical advice is now obsolete. Modern engines (Stockfish, Leela Chess Zero, AlphaZero derivatives) play at a different level. Many "blind spot" Kongsted identified have been largely solved. Many recommendations about hardware/software are "laughable" by modern standards. Many anti-computer strategies are dated (closed-position approaches won't work against Stockfish). Most 2003 engines were traditionally "alpha-beta" search engines; modern engines use NN + Monte Carlo tree search (MCTS) algorithms. Evaluations functions have changed dramatically, and exploiting engine weaknesses is much harder now. So I don't recommend it, it's not worth it for actual improvement. You can get better, more current information for free online. Ciao 
How to Use COMPUTERS to Improve Your Chess
It's a good book for its time, but computer chess has advanced so dramatically that much of the practical advice is now obsolete. Modern engines (Stockfish, Leela Chess Zero, AlphaZero derivatives) play at a different level. Many "blind spot" Kongsted identified have been largely solved. Many recommendations about hardware/software are "laughable" by modern standards. Many anti-computer strategies are dated (closed-position approaches won't work against Stockfish). Most 2003 engines were traditionally "alpha-beta" search engines; modern engines use NN + Monte Carlo tree search (MCTS) algorithms. Evaluations functions have changed dramatically, and exploiting engine weaknesses is much harder now. So I don't recommend it, it's not worth it for actual improvement. You can get better, more current information for free online. Ciao
Yes it's outdated, however the information can still be useful for playing low to mid level bots on Chess.com since they are programmed to have blind spots, but even old computers play well?
It's a good book for its time, but computer chess has advanced so dramatically that much of the practical advice is now obsolete. Modern engines (Stockfish, Leela Chess Zero, AlphaZero derivatives) play at a different level. Many "blind spot" Kongsted identified have been largely solved. Many recommendations about hardware/software are "laughable" by modern standards. Many anti-computer strategies are dated (closed-position approaches won't work against Stockfish). Most 2003 engines were traditionally "alpha-beta" search engines; modern engines use NN + Monte Carlo tree search (MCTS) algorithms. Evaluations functions have changed dramatically, and exploiting engine weaknesses is much harder now. So I don't recommend it, it's not worth it for actual improvement. You can get better, more current information for free online. Ciao
Yes it's outdated, however the information can still be useful for playing low to mid level bots on Chess.com since they are programmed to have blind spots, but even old computers play well?
They are not programmed to have blind spot, they only limit how deep they look, make them choose move that are lower on the list or even just inject random legal move. That's why it never feel natural. The same bot can hang its queen in front of your pawn then fork you on the next move. Stockfish at its minimum level is around 1350 elo, everything under that, is just different type of limitation put on the engine.
It's a good book for its time, but computer chess has advanced so dramatically that much of the practical advice is now obsolete. Modern engines (Stockfish, Leela Chess Zero, AlphaZero derivatives) play at a different level. Many "blind spot" Kongsted identified have been largely solved. Many recommendations about hardware/software are "laughable" by modern standards. Many anti-computer strategies are dated (closed-position approaches won't work against Stockfish). Most 2003 engines were traditionally "alpha-beta" search engines; modern engines use NN + Monte Carlo tree search (MCTS) algorithms. Evaluations functions have changed dramatically, and exploiting engine weaknesses is much harder now. So I don't recommend it, it's not worth it for actual improvement. You can get better, more current information for free online. Ciao
Yes it's outdated, however the information can still be useful for playing low to mid level bots on Chess.com since they are programmed to have blind spots, but even old computers play well?
They are not programmed to have blind spot, they only limit how deep they look, make them choose move that are lower on the list or even just inject random legal move. That's why it never feel natural. The same bot can hang its queen in front of your pawn then fork you on the next move. Stockfish at its minimum level is around 1350 elo, everything under that, is just different type of limitation put on the engine.
But they are programmed to make mistakes unlike stronger engines like Stockfish. The book may have a different example, but generally if a move doesn't exist in the AIs programming that might have been deliberate then?
Did you really?
But they are programmed to make mistakes unlike stronger engines like Stockfish. The book may have a different example, but generally if a move doesn't exist in the AIs programming that might have been deliberate then?
To the best of my knowledge the main way an engine is nerfed are :
- limit the search time or depth
- make the engine choose a lower ranked move on its list
- make a random move
Engine are just program that use algorithm to choose the best move at the present situation, they have no plan, no tactic, no understanding of positioning, and the nerfing always look an natural. Not long ago I saw a game where Stockfish at 1500 or 2000 elo did a move that was so bad that I though it was a bug. It lose this queen for nothing when 2 other legal move were possible. The cause: There were only 3 legal move, the programming choose this specific move to make Stockfish go down the list of best move, and it lost the queen...
But you will see next week when we start to post bots games again Stockfish, Camille bots don't have those strange behaviors and you can use them for training/Learning at a level that current bots just don't allow.
But they are programmed to make mistakes unlike stronger engines like Stockfish. The book may have a different example, but generally if a move doesn't exist in the AIs programming that might have been deliberate then?
To the best of my knowledge the main way an engine is nerfed are :
- limit the search time or depth
- make the engine choose a lower ranked move on its list
- make a random move
Engine are just program that use algorithm to choose the best move at the present situation, they have no plan, no tactic, no understanding of positioning, and the nerfing always look an natural. Not long ago I saw a game where Stockfish at 1500 or 2000 elo did a move that was so bad that I though it was a bug. It lose this queen for nothing when 2 other legal move were possible. The cause: There were only 3 legal move, the programming choose this specific move to make Stockfish go down the list of best move, and it lost the queen...
But you will see next week when we start to post bots games again Stockfish, Camille bots don't have those strange behaviors and you can use them for training/Learning at a level that current bots just don't allow.
I understand what you're saying, but I guess it also depends on the AI too, but generally they do make mistakes and according to the book the computer can have blindspots even if they are not deliberate?
I understand what you're saying, but I guess it also depends on the AI too, but generally they do make mistakes and according to the book the computer can have blindspots even if they are not deliberate?
To the best of my knowledge you don't have AI playing chess, AI even the current top of the line model are horribly bad at chess. You can use AI before and after playing but that's all. As for the "blind spots" that would be limiting the depth of search, on all or certain move. Trouble is you cannot really reliably control what branch get limited.
But they are programmed to make mistakes unlike stronger engines like Stockfish. The book may have a different example, but generally if a move doesn't exist in the AIs programming that might have been deliberate then?
To the best of my knowledge the main way an engine is nerfed are :
- limit the search time or depth
- make the engine choose a lower ranked move on its list
- make a random move
Engine are just program that use algorithm to choose the best move at the present situation, they have no plan, no tactic, no understanding of positioning, and the nerfing always look an natural. Not long ago I saw a game where Stockfish at 1500 or 2000 elo did a move that was so bad that I though it was a bug. It lose this queen for nothing when 2 other legal move were possible. The cause: There were only 3 legal move, the programming choose this specific move to make Stockfish go down the list of best move, and it lost the queen...
But you will see next week when we start to post bots games again Stockfish, Camille bots don't have those strange behaviors and you can use them for training/Learning at a level that current bots just don't allow.
Hi, saying engines "have no understanding" confuses "implementation" (that is, algs) with "emergent behavior" (the strategic play). Modern engines don't "plan" in the human narrative sense, but their evaluation+search produces behavior that functionally mimics strategic understanding. Dismissing this as "just algs" is like saying humans "just have neurons" and therefore lack comprehension. You claim that "nerfing always looks natural": no, artificial handicaps don't produce human-like errors. Bots using random-move injection produce unnatural blunders (e.g., Queen sacrifices with no compensation) that no human at the target ELO would do. Besides, you described a Stockfish bot at 1500-2000 ELO hanging a Queen when 2 safer move existed, attributing it to "programming choosing this specific move to make Stockfish go down the list": the problem here is not just "going down the list", it's that engine evaluation at reduced depth/skill can "mis-rank" moves in ways humans would not. A human at 1500 ELO does not hang a Queen because a probability distribution picked it; they hang because they did not see the threat. The cognitive error model is totally different. About asserting that certain bots (Stockfish, Camille, etc.) "don't have strange behavior", without citing testing methodology or comparative analysis is "anecdotal". Until there is a controlled, peer-reviewed evaluation of bot error patterns vs. human error patterns at matched ELO, this claim is unverified. Regarding "you can use them for training at a level that current bots just don't allow", the problem is that handicapped engines simulate "weakness of calculation", not "weakness of understanding". If the goal is to train against human-like errors at a given ELO, most current bot-handicapping methods are poorly suited. They teach you to punish "alg weaknesses", not "cognitive" ones. What you missed entirely is the human error model: a human player at 1500 ELO blunders due to "selective attention failures", not uniform randomness; besides, he misevaluates positions due to "pattern-recognition gaps", not shallow search; and finally, he makes time-pressure errors due to "cognitive load", not probabilistic move selection. Until bot-handicapping methods explicitly simulate these cognitive constraints, they will produce unnatural play that trains the wrong skills. Ciao
For the case I site, that's what happen, we look into it because I was not understanding how SF 1500elo choose to block a bishop check with a queen when it could : block with a bishop or just move the king. In this case there was only 3 legal move and error injection make it go down the list and take number 3.
As for bot and error/weakening, Camille allow to do it in a completely different way, and as we are going to release free online, I expect to see many weak/joke/training bots.
Like this one, will only move pawn for the first 10 moves and have an over evaluation of the 4 central pawn at 2 point and the lateral one at 1.5. So it start moving pawn then continue
A kind of bot unique to Camille, the broken one. Meaning we try to create a bot but the rule break it (not the intended results ).
Here the 2500 elo bot was unable to beat SF 1500 elo even once over 30 or 40 games.
Goliath lost around 500 elo point with one simple restriction , other profile are just slightly handicap and of course that's against SF, a human understanding the rule could have a wider elo gap.
First, I have to tell you, computers may not help you much. Don't learn to play chess like a computer. It is improper because we cannot calculate hundreds of variants. I mean, you should have your own understanding on an opening or a position. For example, if there is a chance for you to play a positional sacrifice, a computer will tell you how to play every move exactly, but as a human you are different from computers, so before the sacrifice, you need to plan how to attack after the sacrifice, and you can also play well.
If you want to use computer, that's OK. You can use computer to help you know where you make a mistake and what's the best move.
Finally, hope my suggestion will help you.
I've been reading this book by Christian Kongsted and so far I've learned a lot from it like how computers have a blind spot that you can use to your advantage, and how to defend against the computer. It also mentions IBM's Deep Blue and the 1997 rematch that Garry Kasparov lost among other games.
I strongly recommend you read this book to learn how to improve your chess playing the computer!