|
< Earlier Kibitzing · PAGE 7 OF 7 ·
Later Kibitzing> |
| Apr-20-18 | | talwnbe4: ..of course language is somewhat useless unless one understands the words. Take the word "sense" and even worse (better? )the english word "draw", "draw" has a large number of meanings and of course languages change over time, from spelling to morphology (inflection) to meaning, at any given moment - a word might have a number of meanings to any person, let alone many persons. |
|
| Apr-22-18 | | talwnbe4: CHC, as to computers achieving results, then this is perhaps more to your liking ?
https://www.youtube.com/watch?v=sjA... |
|
| May-22-18 | | bobbyperezchess: I don't like the move Stockfish played on its 11th move which is 11...Bf6?!, since it denies control of the weak square d6, which AlphaZero immediately conquers with 12.Nd6! . But still, Van Arkel played this positional error in his game against Roobol. Roobol played 12.Re1!? which is quite okay and still gives white the advantage, but not as strong as AlphaZero's 12.Nd6! . The move played by Stockfish is what most chess players with positional knowledge and understanding will not commit. I prefer 11...d5, though the pawn will be regained quickly with 12.exd5 cxd5 13.Re1!? Re8 14.Nc3 e.g, 14...Nba6 15.Nxe7 Rxe7 16.Rxe7 Qxe7 17.Nxd5 and if 17...Qd8?, 18.Nf6+ Qxf6 (18...gxf6?! 19.Qg4 Kh8 20.Bxb7 ± and white wins the bishop pair while black, with two knights against white's two bishops and a bad kingside structure, obtained a nightmare position and irreparable damage) 19.Qf3! Qe7 (black cannot be as worthy as white after 19...Qxf3 20.Bxf3 ±) 20.Be3 Nb4!? (20...Nc5 21. Bxc5 ± looks good for black, as black has exchanged one of its knights to white's bishop, but could black afford the irreparable structural damage to its pawns?) 21.Rd1 ± with a nice advantage. But we are humans, and we are not (and never) computers. |
|
| Jun-17-18 | | ThirdPawn: This probably can be considered as the Game of the Century. Not only did Stockfish not see 21.Bg5!!!, but also failed to see 20.Qg4! Of course, I saw it, but then I saw the f5 fork and stayed away. I guess that is what separates humans from machines now. In the end, Stockfish must be complimented on lasting another 100 moves as no GM would have lasted 20 more. |
|
| Mar-30-19 | | rcs784: I just discovered something sensational about this game that nobody seems to have noticed yet: Both AlphaZero's stunning sacrifice of the second pawn with 14. e5 (played in at least one of the other published games as well) and subsequent plan of attack were anticipated in a human game, 6 years earlier, between two German players (sadly not in CG.com's database!): Karl-Heinz Podzielny (2408) vs. Hans Georg Emunds (2179), 2011: 1. d4 Nf6 2. Nf3 b6 3. g3 Bb7 4. Bg2 e 6 5. O-O Be7 6. c4 O-O 7. d5 exd5 8. Nh4 c6 9. cxd5 exd5 10. Nf5 Bf6 11. e4 Ne7 12. Nd6 Ba6 13. Re1 Nc8 14. e5 Nxd6 15. exf6 Qxf6 
click for larger viewThis is exactly the same position as after move 15 in the AlphaZero game, but the moves played to reach it were slightly different (different move order, but also Black's knight went d5-e7-c8-d6 instead of Stockfish's d5-c7-e8-d6)--probably the main reason this game seems to have escaped notice so far. But White's idea is exactly the same as AlphaZero's, and his follow-up is remarkably similar as well: 16. Nc3 Bc4 17. Bf4 Nf5 18. Ne4 Qg6 19. h4 h6 20. h5 Qh7 21. Qg4 Kh8 
click for larger viewThis is not quite the same position as that reached after move 20 in AlphaZero-Stockfish (in particular, Black's knight is placed much more actively on f5 instead of b7), but it's similar enough for serious comparison. I'm not sure if Black's knight on f5 is actually a help or a hindrance, though, given that Stockfish responded to AlphaZero's stunning 21. Bg5 sacrifice was 21...f5. Stockfish seems to have always put the knight on b7 instead of the seemingly superior f5, so perhaps Bg5 is a strong move here too. If it was, Podzielny certainly didn't find it (I don't certainly blame him, though; I think a lot of strong GMs probably wouldn't find Bg5 over the board), and the game finished as follows: 22. Nd6 Be6 23. Qe2 Nxd6 24. Bxd6 Rc8 25. b4 a5 26. b5 a4 27. Be4 Bf5 28. Qf3 Bxe4 29. Rxe4 cxb5 30. Rae1 Nc6 31. Qxf7 Qg8 32. Qxd7 Qd5 33. Re6 Rd8 34. Re8+ Rxe8 35. Rxe8+ Rxe8 36. Qxe8+ Kh7 37. Bf4 b4 38. Qg6+ Kh8 39. Qe8+ Kh7 40. Qg6+ Kh8 1/2-1/2 White is an IM; I'm not sure about Black's title or if he had one. Chessbase's database, where I found this game, doesn't have very good information about the tournament circumstances, but both players were over 50 years old at the time, which leads me to think the game may have been played in a German senior open or similar event. Chessbase has no record of any other human playing this double pawn sacrifice before or since, other than one game from 2018 in which Russian IM Dmitri Saulin, (undoubtedly inspired by AlphaZero) tried it against a lower-rated player but went wrong and lost. But 6 years before AlphaZero, Karl-Heinz Podzielny anticipated not just the initial sacrifice but AlphaZero's entire follow-up prior to the piece sac. The fact that AlphaZero, with no pre-packaged opening book or database of human games, replicated Podzielny's play almost move for move only serves as a further testament to the genius and essential soundness of the whole variation. Not bad for a "lowly" IM. |
|
Apr-12-19
 | | OhioChessFan: Hashtag Caissars |
|
| Apr-12-19 | | scholes: Maybe that was correspondence game |
|
Apr-12-19
 | | PawnSac: < ChessHigherCat: But humans are the only decently developed species! Outside of mankind, all the creatures have developed quite indecently and traipse around in public without any underwear! > classic CHC! lmao |
|
Apr-12-19
| | PawnSac: < ChessHigherCat: (If you wonder why I'm interested in "dead" languages, it's because I'd rather read interesting ancients than waste time trying to communicate with boring contemporaries). > another kudos.. hehe. So what Greek lit are u reading? |
|
| Apr-12-19 | | ChessHigherCat: Hi <PawnSac>, thanks for the kind words! I'm just reviewing my grammar but fortunately the textbook has lots of interesting original texts. Unfortunately not all of them are of a high philosophical level though. Yesterday I read a passage describing a sea battle between the Spartans and Athenians written by Anaxagoras (who finished Thucydides' incomplete version of the Peloponnesian War), which is pretty brutal (an oxymoron). Today I tackled an excerpt from an Athenian court speech composed by Lysias in 394 BC for a middle-aged homosexual defending himself against a charge of assault brought against him by a fellow Athenian, Simon, who was his rival for the affection of a young slave.
The subject matter isn't very uplifting or enlightening but it's still fascinating seeing the world as viewed from the minds of people thousands of years ago. Anyway, thanks again for your interest and I look forward to more of your posts, which I always enjoy. Sorry to move from the sublime to the ridiculous, but this excursion brought me back to a particularly hilarious homoerotic fantasy of the great anti-gay crusader KingFish/Big Pawn: <But on this page, I have <CHC> shaking in his boots, looking to crawl under a blanket fort with <moron>, <desperate to escape [my] penetrating presence>. No vacancies in my blanket fort for that creep! <pdxjjb: <ChessHigherCat> Back on the machine learning thing, I don't think the truth is quite as optimistic as the way you played it back. In practice, the machine won't just continue to improve, and in some cases it will get worse if you overtrain it> Please don't think this is a sarcastic question, since I haven't had time to follow AlphaZero very closely (although I'm still a big fan), but would you say that AZ has ceased to improve now due to overtraining/overfitting or is it still getting better? |
|
Apr-13-19
| | OhioChessFan: <CHC> can't man up on Rogoff, so he tries to win a debate with a non-existent opponent on this page. Most liberals spend a lifetime winning debates with people on the same side agreeing the other side is wrong. Typical liberal. Then they run into people who don't agree, and all of a sudden they don't have the intellectual chops to defend their position. |
|
| Apr-13-19 | | ChessHigherCat: What debate? I never saw the post before because I had Beepy on ignore at the time and just thought it was hilarious: Big Macho Man: < I have <CHC> shaking in his boots, looking to crawl under a blanket fort with <moron>, <desperate to escape [my] penetrating presence>. Ha. ha, ha, ha. ha, Ha. ha, ha, ha. ha, Ha. ha, ha, ha. ha!!!! As to serious debate on Rogoff, forget it, it's all some kind of demented Championship Wrestling routine with the Trumputinettes. |
|
| Dec-18-20 | | WorstPlayerEver: 16... Ne8= |
|
| Dec-18-20 | | WorstPlayerEver: Sloppiness of A0 Team confirmed. After the brilliancy 21. Bg5, they probably lost interest in the game. So in this position A0 'misses' the simple 33. Bxd6. Which.. obviously reduces the sea snake status of the actual game by hmm let's say 50%. 
click for larger view |
|
Dec-18-20
 | | beatgiant: <WorstPlayerEver>
Are you claiming that 33. Bxd6 is a lot better than the game line? Based on what? The d-pawn does fall several moves later. And it's understood that AlphaZero uses less materialistic evals than traditional engines. |
|
Dec-19-20
| | beatgiant: I think the point of 33. Qxa8 is that White is able to trade queens and reach an easily won endgame, which isn't possible after 33. Bxd6 Nd7 and the knight will have time to reach f6 and prevent that. |
|
| Dec-19-20 | | SChesshevsky: <I think the point of 33. Qxa8 is that White is able to trade queens and reach an easily won endgame...> When I was first starting, played a very experienced expert. At one point he sat back, staring at the board, and says to himself but out loud, "Hmm, I have so many ways to win." Think, based on A0's millions of games Monte Carlo historical basis, it's also sure it has many ways to win after 20...Kh8. The continuation that includes 33. Qxa8 does seem pretty straight forward though. But might be the only way to really understand A0's choices is to compare variations with some other, at least somewhat similar, Monte Carlo historical based software. |
|
| Dec-19-20 | | WorstPlayerEver: Oh dear, A0 grabs a Rook instead of a pawn and.. yessir.. "it is less materialistic.." Nonsense; it's made by humans, remember?
33. Bxd6 Nd7 34. Rxa8 Qb1+ 35. Kg2 Nf6 36. Qf5 Qd1 37. Be7 Qxg4+ 38. Qxg4 Nxg4 39. Rxa7 +5 SF12 40 ply Black can resign. 
click for larger viewGuess what??? 39 moves is exactly one third of 117 moves. No? The line is pretty well forced, though. |
|
Dec-19-20
| | beatgiant: <WorstPlayerEver> <Oh dear, A0 grabs a Rook instead of a pawn and.. yessir.. "it is less materialistic.."> No, in both lines A0 wins a rook, in yours he also grabs the d-pawn immediately and in A0's he grabs the d-pawn half a dozen moves later. <Nonsense; it's made by humans, remember?> Yes but it's made by humans based on different principles. The actual game reaches the following position after move 38. 
click for larger viewwhich, to my eye, is very similar to the position at the end of your line above. White's king and bishop and Black's king and knight are placed slightly differently. Why do you think yous is much more resignable than the game's? |
|
| Dec-19-20 | | SChesshevsky: <...Guess what??? 39 moves is exactly one third of 117 moves...> Get the feeling that A0 play raises a lot more questions than gives answers. Seems at times it gets a much better, likely winning, position then it continues play like a sadistic troll. Once it's figured it's winning, eliminates any sort of counter play. Like in the 33. Bxd6 line, Black does get some checks. Meaningless but something. And A0 doesn't even want to allow that. Then appears A0's plan is just to wipe out all SF's pieces. Put a voice synthesizer on A0 and I'd probably also expect it would provide some nasty trash talk from move 20...Kh8 on. Wondering if this chess cruelty was programmed in or is just a nature of how it learned how the game is played? If it's nature. Maybe want to reconsider expanding AI's power too much. |
|
Dec-19-20
| | beatgiant: <SChesshevsky>
Or maybe Black's threshold for resignation was set rather high. Most humans would have resigned at least after losing a pawn (move 64) or a second one (move 87). |
|
Dec-19-20
 | | AylerKupp: <<SChesshevsky> Think, based on A0's millions of games Monte Carlo historical basis, ...> One way to train a chess neural network is using Supervised Training, where the engine is provided with a large set of master level games, thus providing a historical basis for its play. But AlphaZero's neural network training set used Reinforcement Training with AlphaZero playing games against itself and, whenever it won a game it gave itself a "reward" and was thus reinforced for its play. So its training set did not have any historical basis, unless you count the games that AlphaZero played against itself. And don't confuse the training of a chess neural net with the use Monte Carlo Tree Search (MCTS) during the game. When using MCTS the engine, during the game, conducts a set of simulated games (playouts) and determines the winning probability (I think that it should be scoring probability but I could be wrong) of the various moves. An example of use of MCTS with a classical engine containing a hand-crafted evaluation function (no neural net) is Komodo MCTS. In AlphaZero's case (like with LeelaC0) the situation is a little bit blurry because AlphaZero does not perform pure random playouts but supposedly uses the results in its neural net to predict the results of these playouts. No, I don't know how that works. |
|
Dec-19-20
| | AylerKupp: <<WorstPlayerEver> Oh dear, A0 grabs a Rook instead of a pawn and.. yessir.. "it is less materialistic.."> It's not explicitly materialistic as I explained to <SChesshevsky> above. It determines what move to play on the basis of its playouts to determine which candidate move has the best winning probability. And it's not surprising that playouts that result in obtaining a material advantage will likely result in winning probabilities higher than those playouts that don't. |
|
Dec-19-20
| | AylerKupp: <<beatgiant> Or maybe Black's threshold for resignation was set rather high.> All that engines do is report an evaluation to the GUI, although I don't know what GUI or equivalent was used in the AlphaZero vs. Stockfish matches. It's the GUI that determines whether it's time to resign (typically on the basis of having the position evaluation exceed a specific threshold a certain number of consecutive moves, common to both White and Black), whether the position is a draw as a result of either a 3-fold repetition of position, the 50-move rule, a tablebase result, or other criteria dependant on the GUI and the information it has. And, of course, the operator may resign for the engine. I think that the reason that there were so many long games in the AlphaZero vs. Stockfish matches is that recognition of many of the win/draw/loss situations were not programmed into the GUI. |
|
| Dec-20-20 | | WorstPlayerEver: There's a huuuuuge difference: Black has no Nf6 after 33. Bxd6. So it's no use playing through at that level, I suppose. I will illustrate this by giving three funny lines, starting from the diagram in my last post (after 39. Ra7): 39... Ne5 40. Bf8 Bf7? 41. f4 Bd5 42. Kg3 Nc6 43. Rd7 Ba2 44. Bg7 Kg8 45. Bh6 Nd4 46. Rg7 Kh8 47. Ra7 Nf5 48. Kg4 Bb1 49. Bf8 Kg8 50. h6 +8 39... Ne5 40. Bf8 Bd5 41. Kg3 Kg8 42. Bg7 Nf7? 43. Bh6 sac sac Nh6 44. Kf4 Nf7 45. a3 Kg7 46. Rd7 Bc6 47. Rc7 Bd5 48. Ra7 (NB White deliberately lost a tempo here: zugzwang) Kh5 49. Ra6 Kh5 50. Rb6 Ng5 51. a4 +8 39... Ne5 40. Bf8 Bd5 41. Kg3 Kg8 42. Bg7 Nc6 43. Rc7 Nb4 44. a3 Nd3 45. Bh6 (45... Nb2 46. Rg7 Kh8 47. Rd7) Bf7 46. Rf7 Kf7 47. b3 +7.5 and White is simply two passers up. So A0 team ahs ben trolling SF. Prove me wrong. |
|
 |
|
|
< Earlier Kibitzing · PAGE 7 OF 7 ·
Later Kibitzing> |
|
|
|
Kibitzing
sac: 24.Qd6
icon.