|
What blows my ~1100 mind is how they're so good tactically that there's no tactics. All of my games have Benny Hill level tactics, wild swings of the eval, etc. These guys, it's like watching an arm wrestling match between hydraulic presses. Incredible power, little motion. (Image from Game 1)
|
#
¿
Nov 28, 2021 16:05
|
|
|
|
| # ¿ May 15, 2024 01:53 |
|
|
Doctor Malaver posted:Disappointing game from Carlsen. He went directly from prep to draw. Not something you'd expect from the world champion, and one of the best players ever, playing with white pieces. He's also playing one of the best players ever, and both are good enough that even Stockfish can find only a couple centipawns worth of complaint about each move on average. To get a decisive result, one of the players is probably going to need to make a mistake. I can understand why both are reluctant to risk being the one who makes that mistake.
|
|
#
¿
Dec 1, 2021 02:48
|
|
|
Dias posted:Game 7 N/V: yo game 6 was so sick tho Yasser Sierawan's "Winning Chess Openings" goes through all the usual openings at a level suitable for beginner and club level, and concludes with the following advice: If white, play the King's Indian Attack. As black against 1. e4, play the Pirc. As black against 1. d4, play the King's Indian Defense. All three involve essentially the same moves, so there's no real memorization. They avoid traps and leave you with a solid defensive structure. You do pay a small price of potentially letting the opponent get a head start on the center, and they're not considered great openings at the grandmaster level. But at low levels (e.g., me), who cares? It's going to come down to tactics anyway. All you need from an opening is not to shoot yourself in the foot right out of the gate.
|
|
#
¿
Dec 4, 2021 22:19
|
|
|
totalnewbie posted:I think he's completely lost all confidence in his play and has already acknowledged Carlsen as just the better player (which he is) and basically given up on the match, but can anyone really blame him? Magnus has 74 more Elo points than Ian. That implies an expected score of 0.6 per game, which over a 14 game match is a pretty solid favorite. In my opinion it's not an otherwise-even match where one guy just happened to crack under the pressure, it's that Magnus really is significantly, measurably better than everybody else. Good on Ian for fighting his hardest. I'm not sure anyone else could have done better. In a couple years though? We'll see. Firouzja is a walking buzz saw.
|
|
#
¿
Dec 9, 2021 04:33
|
|
|
Apsyrtes posted:I wouldn't say you had "a lot of blunders" -> just one and it was getting your bishop trapped. I hear even some pretty good players do this once in a while. Really though, I agree with the "resignation was the blunder" A 2-point material deficit early in the game is at worst a learning experience, and more probably there'd be an opportunity to take advantage of an opponent mistake at some point.
|
|
#
¿
Dec 11, 2021 02:59
|
|
|
fart simpson posted:so it's kinda cool to think of it that way, that "both players have equal winning chances" or "neither player has a chance to win" or "one player is 'winning' but there's a trick for the other player to force a draw" are 3 entirely different types of "equal" In principle a perfect chess engine would have only three evaluations: "White to mate in N", "Black to mate in N", and "Draw". That we humans have a bunch of weird shades of "draw, but Player X is more likely to blunder" is not necessarily super easy for an engine to evaluate. There have been some efforts to develop engines that play like people, blunders and all, by training based on a corpus of games played by humans at a particular ELO. Those could probably be used in this context. Might be an interesting research problem!
|
|
#
¿
Dec 13, 2021 23:51
|
|
|
jiggerypokery posted:If you had chess.com's database you could probably get a long way pinging off moves actual players have made in most positions with just position lookup. The lichess database of almost three billion rated games is available for free download if anyone is a amateur chess coder and wants to give it a try. I don't think pure position lookup would work though; I suspect it only takes a handful of moves in the game for a typical player to get into "this position has never been seen before" territory.
|
|
#
¿
Dec 14, 2021 00:58
|
|
|
There's no proof that chess is a draw with best play, but it seems pretty likely for a number of reasons. For instance, endgame tablebases have given us a lot of insight into restricted versions of chess with fewer pieces, and in fact they show us that most endgames reached by grandmasters and computers alike are drawn with ideal play. It's the mistakes, and the brilliancies that provoke mistakes, that give human chess its interest.Sub Rosa posted:One way you can see this is that in terms of engine evaluation, you pretty much need to get to +/-2 before a game isn't a draw. In practice with the best engines, chess not only is a draw with perfect play, it's usually a draw with a handful of inaccuracies or even some mistakes. So if black is still able to hold with imperfect play, the idea that white could win with perfect play becomes increasingly unlikely. In my opinion long-time-control computer chess still has some room left to grow in interesting ways at the highest levels, for instance with play starting from positions other than The Starting Position. Say, give one engine a position that seems slightly but meaningfully better (say, the +2 that you mentioned) and see if it can convert (or save the draw, if on the receiving end). In essence this explores the space of positions that realistically might or might not be a draw with best play.
|
|
#
¿
Dec 15, 2021 00:45
|
|
|
I don't think I understand this Lichess puzzle, possibly because I've only recently managed to work my way this far up (~1900) in my puzzle rating. Materially the solution boils down to "you win a pawn", but it's a full 5.0 points worth of eval at the default depth. What endgame principle am I missing?
|
|
#
¿
Jan 13, 2022 04:19
|
|
|
Puzzles have taught me that chess is easy, it's just that I'm an idiot.
|
|
#
¿
Jan 27, 2022 04:40
|
|
|
jesus WEP posted:Daniel Naroditsky I watched two of his "watch me do puzzles" videos and my puzzle rating went up 200 points basically instantly. 
|
|
#
¿
Feb 1, 2022 23:27
|
|
|
Lichess has a great practice page with all the basic mates plus many interesting patterns: https://lichess.org/practice As a terrible beginner I once stalemated a king-and-rook vs. king endgame, and these practices were super helpful. I am still a terrible beginner, but I don't stalemate that endgame!
|
|
#
¿
Aug 21, 2022 21:24
|
|
|
Chromatics posted:I don't really understand the mechanism either, is it that they are looking at an engine in their phone away from the table or something? The tweet mentions a 15-minute broadcast delay as an anti-cheating measure. That suggests they're worried about a confederate outside the match area. You could get a little single-board computer like a Raspberry Pi Pico, plug in a little dongle to get cell phone data, and attach one of the teeny tiny motors they use to make cell phone vibrate. Then the confederate watches the stream, plugs the position into Stockfish, and texts you "Nxe6" or whatever the best move is. Your little widget, carefully concealed in your shoe or collar or something, then vibrates that move in Morse code. The whole thing could be smaller than a pack of playing cards. Is this a ridiculous and contrived thing to do? Sure, but people have done stupider.
|
|
#
¿
Sep 5, 2022 22:08
|
|
|
neaden posted:Sure, a company that just bought Magnus' chess app would never do something to make their cash cow happy, private companies always act in the interest of justice and fairness. I'm sure chess.com has secret proof that Hans cheated on an OTB tournament they didn't run but just won't reveal it. Although they don't publicize the details, it's not like chess.com is using some super-double-secret algorithm to catch cheaters. The basics are well-known: does the selected move the top engine moves weirdly often, is the timing of the moves unusually lengthy even when only one move is reasonable, how is centipawn loss statistically distributed, how are move ranks statistically distributed, etc. There are AIs like Irwin that automate a lot of this. A GM who knows what playing a human GM "feels" like is usually pretty reliable too, although a tired and peevish GM could certainly get it wrong. So far no evidence has been presented in either direction, but the as-played moves in the tournament should give plenty of data that chess experts will probably be able to use pretty effectively to prove or eliminate the possibility of cheating.
|
|
#
¿
Sep 7, 2022 01:55
|
|
|
Helianthus Annuus posted:Magnus has a recent history of sabotaging top-level chess by withdrawing from this tournament and by refusing to defend his title -- if anyone deserves a ban, it's him. By historical World Chess Championship standards you're a weirdo if you're not a prima donna with Dr. Evil style demands about how to run chess championships.
|
|
#
¿
Sep 9, 2022 01:23
|
|
|
Starsfan posted:Yeah I think Magnus himself mentioned this at one point before this whole thing blew up (not necessarily related to Hans Nieman I think) but if he had the ability to get just 2 computer moves in an entire game at his discretion he would literally never lose to anyone at his level of strength. These players are so closely matched in skill that the smallest edge at the critical moment would make the difference. Top-level chess players are surreal. If you gave me two computer moves I'm not even sure my rating would improve.
|
|
#
¿
Sep 21, 2022 03:34
|
|
|
Starsfan posted:I mean that move accuracy thing isn't the be all and end all, like if a top level GM was playing one of us (and trying their hardest) their move accuracy would likely approach or be at 100% At my basically barely-four-digit-ELO blitz level I've had a few extremely high percentage games. They tend to go along the lines of "oh his queen just wandered into traffic, I guess capture it?" and sure enough Stockfish also thinks it's a good idea to capture queens that have wandered into traffic.
|
|
#
¿
Sep 27, 2022 01:20
|
|
|
Powered Descent posted:By this logic, everyone who ever took a plea bargain is guilty. When it's "confess and get a year, go to trial and get ten" you're definitely going to get some false confessions. When the penalty is "get banned on chess.com" that's probably less likely. On the other hand, a "secret evidence, secret decision" system is fine for a silly game website. But once real money and reputations are on the line there really does need to be a more transparent solution.
|
|
#
¿
Sep 30, 2022 22:36
|
|
|
tractor fanatic posted:If you're worried about cheating it's going to be increasingly hard to take online chess seriously. It will only get easier and easier to cheat and there's basically no way to stop it all. Online money is very much real money, and it might even be criminal to cheat in a money game, but all this just makes it harder to take online chess seriously. With OTB at least you can imagine there will be physical security measures, but there's nothing stopping increasingly sophisticated engine use online. At the money/prizes competition level that might be true. But if cheating gets only you to an ELO where you'd end up in a random 10 minute rapid game with me, you should probably just give up on chess. And if for some reason you turn out to be a bot that perfectly replicates the experience of playing a terrible novice, I guess I don't actually have that much to complain about. It would be a substantial realism improvement for most of Chess.com's bots. But yes, actual money competitions will probably eventually have to be pretty much exclusively OTB.
|
|
#
¿
Oct 4, 2022 23:50
|
|
|
neaden posted:Duck chess is good. Turning Leela Chess Zero into a godlike duck chess engine would probably take an hour of coding and a day of letting a rack of GPUs run. I think it would be kind of fun to get crushed under the Terrible Rubber Tail.
|
|
#
¿
Oct 6, 2022 00:14
|
|
|
neaden posted:Dlugy is on reddit trying to clear his name now, but a lot of what he's saying doesn't make much sense like he played titled Tuesdays with his students suggesting moves and some of them cheated. If your students are suggesting moves by throwing darts at a chessboard, you're cheating. You can't get external help, period. Magnus himself is on video doing this though, so 
|
|
#
¿
Oct 12, 2022 23:13
|
|
|
If we're doing "who's better" and you're willing to take a look at a position in an OTB lunch-break game where both players are probably triple-digit strength, here's an interesting(?) position (white to move): Note that both players very bad, so don't expect anything subtle. I was black and assumed that I was toast, since white's up +3 in material. I'm not actually sure why Stockfish rates it as about -4, so that's why it's interesting to me. I'm guessing it's because so many of white's pieces are undefended and the black queen is poised for Qxb2, causing a bunch of subsequent problems.
|
|
#
¿
Oct 26, 2022 01:29
|
|
|
I read an interesting article about a neural network a research group had built to play Stratego. It's a game of incomplete information, which chess isn't, but what interests me is the setup phase where you initially place your pieces. Has anyone done some thought about Chess960 but where each player can choose their piece ordering in secret before the game? It would be an interesting exercise to try to come up with an ideal ordering that works best against an arbitrary opponent configuration. I wonder if the ideal ordering would be the same for white and black - it may well not be.
|
|
#
¿
Dec 3, 2022 03:42
|
|
|
|
| # ¿ May 15, 2024 01:53 |
|
|
Apparently it's very difficult to build a good Stratego engine. Many of the techniques that work for chess engines like Leela and Stockfish NNUE can't deal with the vast state space of Stratego. It's a much more complicated game in terms of the number of states, in fact much more so than even Go or poker. This particular engine, DeepNash, uses a novel game theory approach that avoids having to make an explicit model of what the opponent's pieces might be. Despite this engine performing in the high 90s to 100% win percentage over the other top engines, it's still beatable by the best humans.
|
|
#
¿
Dec 3, 2022 22:16
|
|