|
< Earlier Kibitzing · PAGE 2 OF 2 ·
Later Kibitzing> |
| Dec-07-17 | | WorstPlayerEver: Lol what about the fact that 9. e5 is a novelty and S8 is supposed to play without book moves? That points to insecurity. Considered that S8 had 1 minute per move. I let the hardware discussion for what it is. Now, let's see if I can pull the same trick. Lend me a second... 9. e5 Ne4 10. 0-0 Nxc3 |
|
| Dec-08-17 | | Cheapo by the Dozen: Yet another AlphaZero game to which I have the same reaction -- AlphaZero's strategy was a steady and successful grab for space. |
|
| Dec-08-17 | | SChesshevsky: For whatever reason, the restrictions on Stockfish seemed to regulate it to passivity which in turn put itself into a bind over and over. Worse, it also seemed unable to calculate its way to even try to break the bind. So you get a lot of kind of ridiculous looking shuffling of pieces around. This game might be the ultimate example. I kind of get the impression that Alpha might of underachieved. Might be if Stockfish played this way against Komodo or Houdini in tournament condition, it seems possible it could lose nearly every game. Also, as in this game, it seems the restrictions are also more effective as Stockfish might tend toward the instant material gratification without thinking too far ahead or about positional considerations. Reminds me of the tricks grandmasters used to beat the machines when they were still able. |
|
| Dec-08-17 | | talwnbe4: Have to learn to use a phone. Hahaha.. |
|
| Dec-08-17 | | talwnbe4: 36.. c3? is not a move any modern engine would play, Rybka says it's bad as does Stockfish after a minute on a slower computer. 36. ..Ra8 = 36..c3?? +- 1.3 after 10 seconds (stockfish) |
|
| Dec-08-17 | | jeffwadsworth: Yes, 36...c3 makes zero sense. I would love to see the log file of the Stockfish they used in this game. |
|
Dec-08-17
 | | chessgames.com: Our Stockfish 8 can't decide between 36...Ra8 and 36...Rc8, but 36...c6? never entered the picture. That's one a 1-minute evaluation, just like Google claims they used. analysis of move 36...? We're very proud of our dedicated analysis server but it's "only" a $3000 rack mount server. It seems hard to believe that Google Corporation couldn't muster up better hardware. The conspiracy theorists will now say, with plausible evidence, that Google intentionally crippled Stockfish in order to exalt their new technology. I say let Google participate in TCEC next season and see how AlphaZero performs when it's not in a laboratory and doesn't have help from its handlers. |
|
Dec-08-17
 | | varishnakov: It would be very unfortunate if that were true, that Stockfish was crippled. One possibility is it saw so far into the position that it realized it was busted no matter what, so it switched to 36...c3 because everything lost anyway. |
|
| Dec-08-17 | | WorstPlayerEver: <cg>
I'd say we do not have any proof. There are no 100 games. I bet all my chessbucks on it! What I do know is that they did some research on very specific side lines, it must have taken them a while; this did not happen after 4 hours. Then they had to put a label on this project. Because such research is quite expensive. Which Google has to pay for. S8 is like the ultimate victim. For it has a name.. Sites as BBC and chessbase have given no info about time control, conditions met, used equipment whatsoever. 10 games, probably an exclusive collection, is all we have. Bad journalism. IMNSHOTT these AZ guys just want to see money from Google to sponsor their further projects, but they give Google a bad name. Simply because AlphaZero devs claim something which still has to be covered by sources. We're not talking about tic tac toe. Until now the willingness of AZ team/dev to share RELEVANT info about the match is even more cramped than S8's positions with Black. Go figure. They have spread *the match* as if it were world news. On behalf of a few brilliant sacs in rather awkward positions. Which did not contribute to opening theory really afaiac. It's great amusement, but no more than that. Curiosities IMO until proven to be relevant in engine practice. |
|
| Dec-08-17 | | frogbert: Stockfish was crippled in a couple of senses:
1) afaik it didn't use an opening book or endgame tables 2) using a fixed time limit per move instead of a time control with the same average time per move (say, 1 minute) is detrimental to Stockfish, because it can't fully exploit its capability to choose when and where to do deeper selective searches. Regarding 1) - the usage of opening books and endgame tables is "standard" among chess engines these days - due to the known limitations of brute force-ish depth-first search in openings and during the transition to end games. Accordingly, little has been done in order to improve "standard" chess engines' handling of these two phases of the game. What's even more "unfair" is that AlphaZero indeed applied "opening knowledge" (and "middle game knowledge" and "end game knowledge") during play, stored/learned from previous training - although it mostly trained its neural network by playing itself. (As far as I can tell from the info available, it also played games against standard chess engines like Stockfish during training). I don't think we know how much space/storage AlphaZero uses to represent the results of its self-training - I couldn't find that information in the article - but the neural network (after training) can be seen as a huge book of chess knowledge. (The "crazy" part is that it is self-learned, starting only with the rules and goals of chess.) Regarding 2 - "time management", or rather choosing when to look deeper and when not to (selectively), is one of the big improvements in "standard" chess engines over the past decade. So yeah - AlphaZero is amazing, but it wasn't a fair fight, imho. :) |
|
Dec-08-17
| | varishnakov: <cg>
I requested an hour analysis, and it gives 36...c3 as almost indistinguishable from ...Ra8 or ...Re8. This suggests to me that the computer used was just extremely powerful, able to see in 1 min what your version does in 60 (though who knows when ...c3 started to become feasible in its mind--maybe 3 min, maybe 20, maybe 60). |
|
| Dec-08-17 | | daveinsatiable: <WorstPlayerEver> You're quite wrong about Deepmind and publicity seeking. This was just an internal experiment, which makes no claim to be a fair contest. They thought, from the response on here clearly correctly, that it might be of interest to the chess world, so they published a few games.
What they've done is very impressive, but what they haven't (yet) done or attempted to do is make a world-beating chess computer. If they want to they will, but they probably have bigger fish to fry. |
|
| Dec-08-17 | | WorstPlayerEver: <daveinsatiable> Internal experiment? Lol it was all over the *news* 100 games? Any proof of this? No. But keep telling I am wrong. Kinda luv that :) |
|
| Dec-08-17 | | SChesshevsky: If it was simply an internal experiment, the paper released is either an example of trying to fool the public or fool themselves. They clearly imply the superiority of their activity based in the Stockfish match. Has the feel more of PT Barnum than legitimate science. |
|
| Dec-08-17 | | dehanne: A Karpovian nightmare. |
|
Dec-09-17
 | | maxi: I absolutely love it. AlphaZero sacs a meager pawn and is one bishop up by move 18. It is practically impossible to activate black's white square bishop. Capablanca would have been proud! |
|
| Dec-11-17 | | s4life: <WorstPlayerEver> You are acting too butthurt for your arguments to be considered beyond ranting A predecessor of this system defeated the best Go player in the world a year ago.. chess as a computational problem is peanuts compared to that effort .. and they had a year to make it better.. |
|
| Dec-11-17 | | WorstPlayerEver: <s4life>
"chess as a computational problem is peanuts compared to that effort .." Then I don't understand why you let me wait. It seems A0 now has the perfect opening book. I am willing to pay :) |
|
| Dec-11-17 | | Bluegrey: In the position we were discussing Stockfish wanted to play 36...Rd4 with an eval of abt 0.3. However the line continues 37. Be3 Rd7(Semi-forced) 38. Rh4 Be7(best according to SF)
39. Rh6 Kg7
40. R×h7 K×h7
41. Qh5+ Kg7
42. Bh6+ Kh7
43. Bf8+ Kg8
44. B×g6 B×f8
45. B×f7(wins Queen) etc
The line continues and ultimately many moves later,it reaches a fortress type position that it evals at +1.5. Very complicated lines but as far as I can see it should be a draw but SF from afar went for c3 to avoid getting into this line
So at move 36, if black instead plays Ra8, this same tactic seems to work as well, white starts with Be3 preparing to sac on h7.If black shuffles with the rook,white would double on the h file and ultimately sac on h7 or g6 depending on whatblack does. There seems to be some super deep tactics in all lines I've checked so far and some very interesting middle game zugzwangs |
|
| Dec-11-17 | | Bluegrey: <chessgames.com> Copied my analysis from a few days ago above |
|
Dec-20-17
 | | Penguincw: Video analysis of this game: https://www.youtube.com/watch?v=nk-.... |
|
| Dec-20-17 | | Chess Is More: <Penguincw: Video analysis of this game: https://www.youtube.com/watch?v=nk-....> That is great, thanks for the link. |
|
| Dec-24-17 | | Ulhumbrus: 36...c4 looks like a concession as the pawn is no longer protected and this suggests that 35 Bc1 threatens something. |
|
| Dec-27-17 | | 50movesaheadofyou: It just fusking screws you up to come here and see all this talk about Stockfish being crippled when I spent a whole day and half a night replaying the moves on my wood board believing the match was all fair and being amazed by the games. And now you tell me some bad English coffee was spilled onto Stockfish's circuits. Come on man, don't go reopen that #2 whisky bottle of "human intervention" from '97. Garry drank some and told a funny lie, you know. |
|
| Feb-02-18 | | talwnbe4: It does seem like SF is busted on move 36..
36..Ra8 37. Be3 Bg7 38. Bf4 Bf8 39. Rd1 Ra3 40. Bg5 +- 1.35 with a continuing positional weakness |
|
 |
|
< Earlier Kibitzing · PAGE 2 OF 2 ·
Later Kibitzing> |
Kibitzing
