@WallaceTwo logic tells me that it is Stockfish level 0, designed for noobs.
I recomment "Game Changer", a book from Mathew Sadler & Natasha Regan on this topic. Good book even if you are not playing chess, even if you are not interested in KI. If you are interested in both you own the book already or you put it on your wishlist right now.
Alpha Zeros style of playing differs pretty much from the "old" chessprogramms. It sacs pawns very likely for positional reasons, even piece sacs for no obvious reasons are possible. Stocjfish simply can'T beat Aplha Zero what is funny - Alpfa Zero das learned Chess only from playing against himself. The vast wisdom of humal players that stockfish owns don#t help. And even more funny: Alpha Zero plays more like a human than stockfish, much more. Without ever teached about openings humans have played.
Read this book - it teaches you much abput KI and chess.
Alpha Zeros style of playing differs pretty much from the "old" chessprogramms. It sacs pawns very likely for positional reasons, even piece sacs for no obvious reasons are possible. Stocjfish simply can'T beat Aplha Zero what is funny - Alpfa Zero das learned Chess only from playing against himself. The vast wisdom of humal players that stockfish owns don#t help. And even more funny: Alpha Zero plays more like a human than stockfish, much more. Without ever teached about openings humans have played.
Read this book - it teaches you much abput KI and chess.
It’s my brother.
Interesting video about AIs learning to drive Trackmania (a racing game):
www.youtube.com/watch?v=a8Bo2DHrrow
www.youtube.com/watch?v=a8Bo2DHrrow
@nooboss "It's not like Stockfish. Stockfish is a classical engine using alpha/Beta search and Alpha Zero is neural net trained with reinforcement learning."
While this statement is not strictly wrong, it is certainly misleading. You are mixing up two things there, search and evaluation. Stockfish uses alpha/beta search, Alpha Zero used PUCT, a different kind of tree search. This part has nothing to do with neural networks, at least not directly. You can write an alpha/beta engine that uses neural networks, in fact the current version of Stockfish is exactly that.
However, the Stockfish that Alpha Zero was matched up against indeed at the time used a hand written evaluation function (although even that evaluation used a sort of learning in the form of tuning, but much less flexibly than the NN of Alpha Zero).
@royalblue04 I will also comment on this since it is a sort of misconception that I have read a lot. "Alpha Zero sacs pawns for positional reasons." is certainly true, however so did Stockfish at the time. I feel like many people still have this outdated image of greedy play from engines however that's not really the case anymore since almost a decade now.
"Stockfish can't beat Alpha Zero." The version from back then with its given settings indeed could not. However Alpha Zero from back then would be crushed by todays Stockfish. Engines have improved very quickly since then (and always have, for some reference to the previous era in chess engine development: www.reddit.com/r/chess/comments/76cwz4/15_years_of_chess_engine_development/ )
As for Alpha Zero playing more human than Stockfish, this is debatable but I once again disagree. Alpha Zero indeed plays even more adventurous than even Stockfish. However exactly that is what makes it less human for me. Because, surprisingly nowadays it is the humans who are greedy compared to engines. Stockfish is often willing to sacrifice pawns or exchanges and sometimes even pieces for long term compensation when many humans are just very reluctant to give up material. In the end, greed is one of the humans defining factors in chess play (ironically). In that sense I see Alpha Zero's style further away from human play. (although I guess you could argue it's closer to the romantic period of chess play in the mid 1800s)
While this statement is not strictly wrong, it is certainly misleading. You are mixing up two things there, search and evaluation. Stockfish uses alpha/beta search, Alpha Zero used PUCT, a different kind of tree search. This part has nothing to do with neural networks, at least not directly. You can write an alpha/beta engine that uses neural networks, in fact the current version of Stockfish is exactly that.
However, the Stockfish that Alpha Zero was matched up against indeed at the time used a hand written evaluation function (although even that evaluation used a sort of learning in the form of tuning, but much less flexibly than the NN of Alpha Zero).
@royalblue04 I will also comment on this since it is a sort of misconception that I have read a lot. "Alpha Zero sacs pawns for positional reasons." is certainly true, however so did Stockfish at the time. I feel like many people still have this outdated image of greedy play from engines however that's not really the case anymore since almost a decade now.
"Stockfish can't beat Alpha Zero." The version from back then with its given settings indeed could not. However Alpha Zero from back then would be crushed by todays Stockfish. Engines have improved very quickly since then (and always have, for some reference to the previous era in chess engine development: www.reddit.com/r/chess/comments/76cwz4/15_years_of_chess_engine_development/ )
As for Alpha Zero playing more human than Stockfish, this is debatable but I once again disagree. Alpha Zero indeed plays even more adventurous than even Stockfish. However exactly that is what makes it less human for me. Because, surprisingly nowadays it is the humans who are greedy compared to engines. Stockfish is often willing to sacrifice pawns or exchanges and sometimes even pieces for long term compensation when many humans are just very reluctant to give up material. In the end, greed is one of the humans defining factors in chess play (ironically). In that sense I see Alpha Zero's style further away from human play. (although I guess you could argue it's closer to the romantic period of chess play in the mid 1800s)
Alpha Zero is a neural network chess engine, it is similar to lc0 or leela chess zero.
If you want to taste how it plays challenge @lol7241bot .
It will automatically accept the challenge but read the bio for time controls. :)
If you want to taste how it plays challenge @lol7241bot .
It will automatically accept the challenge but read the bio for time controls. :)
@royalblue04 whats KI ? AI ? which lang : D thx for book suggestion
KI in german. Forgot to translate tghe shortcut. Funny enough, the book i recommended is in english ;-) AI is correct.
Ditto the Game Changer recommendation from @royalblue04 . I'm a little over halfway through the book. I'm finding it very enjoyable.
I like that I can take new ideas, such as a rook pawn advancement and apply them to my games. Where I get clobbered because I'm unable to make the same calculations that AlphaZero makes. :D
I have started applying ideas around restricting king mobility with good results, though.
I like that I can take new ideas, such as a rook pawn advancement and apply them to my games. Where I get clobbered because I'm unable to make the same calculations that AlphaZero makes. :D
I have started applying ideas around restricting king mobility with good results, though.
This topic has been archived and can no longer be replied to.