|
Dias posted:Nepo never recovered from losing game 6, it's gotta be frustrating playing against someone that just draws if you play safe and exploits any little weakness you might create trying to win instead. yeah it’s kinda sad to see
|
#
?
Dec 11, 2021 04:31
|
|
|
|
| # ? May 31, 2024 15:15 |
|
|
I feel bad for Anish Giri. I didn’t watch a lot of the later games, but it kind of felt like he had a hard time watching Nepo perform so badly. I’m sure he’s thinking about how he would have played in the challenger’s seat instead.
|
|
#
?
Dec 11, 2021 04:52
|
|
|
nrook posted:I feel bad for Anish Giri. I didn’t watch a lot of the later games, but it kind of felt like he had a hard time watching Nepo perform so badly. I’m sure he’s thinking about how he would have played in the challenger’s seat instead. "I would have drawn that"
|
|
#
?
Dec 11, 2021 06:02
|
|
|
The Stafford gambit remains the most fun opening, although I think most people have watched the same Eric Rosen videos I have and know how to stop it. But it's real fun when they haven't. Multiple moves assaulting the hapless f2 square and repeatedly double-checking the King into oblivion. 1. e4 e5 2. Nf3 Nf6 3. Nxe5 Nc6 4. Nxc6 dxc6 5. Nc3 Bc5 6. Bc4 Ng4 7. O-O Qh4 8. h3 Nxf2 9. Qf3 Nxh3++ 10. Kh1 Nf2++ 11. Kg1 Qh1# Unfortunately it falls apart pretty quick to 6. h3 and 7. Qf3. I also haven't found a good response to 5. e5 followed by 6. d4. The only response that's ever worked for me requires my opponent to blunder into a lost exchange with 5. e5 Ne4 6. d4 Qa5 7. g3? Nxg3, where 8. hxg3 loses immediately to Qxh1 and 8. fxg3 loses to Qe4+ 9. Qe2 Qxh1. But that's also easily defended by Qe2, Qf3, and Be3 which are all pretty natural moves.
|
|
#
?
Dec 11, 2021 20:12
|
|
|
https://www.chess.com/analysis/game/live/32987373315?tab=review This game felt pretty good even if the computer disagrees sometimes, now if only I could stop blundering one-movers because I just wanna attack...
|
|
#
?
Dec 11, 2021 21:09
|
|
|
On chess.com there's a link to get a 7 day free trial of their premium plans, but I already had a trial about 9 months ago. Is the link for a free trial always there and therefore if I try to get another one I'll get charged, or do they just offer free trials periodically? Has anyone here had more than 1? I don't want to accidentally get stung for $80 or whatever.
|
|
#
?
Dec 12, 2021 01:08
|
|
|
Played an over the board 7 round Swiss with 13+2 time control yesterday, first tournament in person ever for me. Finished 2 wins and 5 losses. All the games were close and competitive other than one where I foolishly went into the Italian as black depsite not knowing the theory and got blown off the board in about 25 moves, and another where I was holding my own in a tense position against a stronger player until I blundered a full rook in such a comedic fashion my opponent spent two minutes thinking before capturing assuming it was a tactic. I found it much harder staying aware of those dastardly long diagonals in person as I'm so used to the flat 2d online board. Overall though was a great experience and looking forward to doing again in the new year.
|
|
#
?
Dec 12, 2021 01:35
|
|
|
Nice. Congratulations on taking the next step into
|
|
#
?
Dec 12, 2021 01:47
|
|
|
Captain von Trapp posted:I hear even some pretty good players do this once in a while.
|
|
#
?
Dec 12, 2021 07:04
|
|
|
Aggro posted:The Stafford gambit remains the most fun opening, although I think most people have watched the same Eric Rosen videos I have and know how to stop it. But it's real fun when they haven't. Multiple moves assaulting the hapless f2 square and repeatedly double-checking the King into oblivion. For e5 the engine suggests something like 5...Kd5 6. d4 c5 (if they take you just develop and attack f2) 7. Kc3 Be7 8. Be2 cxd4 9. Qxd4 Bb4 and you're down a pawn but I guess you are on the offense. They can't try and chase your bishop with a3 and b4 because that puts the queen under attack and there's a line where they hang the rook on a1 trying to save it. I don't know much about gambits for black tho as a filthy hypermodern player.
|
|
#
?
Dec 12, 2021 07:47
|
|
|
oh no computer posted:On chess.com there's a link to get a 7 day free trial of their premium plans, but I already had a trial about 9 months ago. Is the link for a free trial always there and therefore if I try to get another one I'll get charged, or do they just offer free trials periodically? Has anyone here had more than 1? I don't want to accidentally get stung for $80 or whatever. I think that always appears. I did the trial and as soon as I cancelled it the banners turned up again
|
|
#
?
Dec 12, 2021 10:50
|
|
|
for game 6, i've seen several places mentioning that 130...Qb1 or 130...Qc2 for black are the only moves that preserve a draw - is this true and if so why? is it because it limits options for white's king?
|
|
#
?
Dec 12, 2021 17:11
|
|
|
So the central broad idea here is that, apparently, black's queen needs to stay in position to check from behind. Nepo lost that when he played Qe6. As for why Qc2 or Qb1 specifically, I would guess that (1) it spots white's rook, preventing white from playing the immediate Nh5, and (2) against Kg4 black preserves the option of Qd1+ (and I guess you then meet Kg5 with Qg1 or Kh4 with Qe1, idea being you keep the knight stuck back on g3). But, you know, lol to working that all out over the board in real time, let alone during time trouble.
|
|
#
?
Dec 12, 2021 17:36
|
|
|
I think I've reached my first rating barrier where it's dominated by players who move very quickly, often choosing to present a lot of tactical complications. I find that I usually get into a favorable position, but either lose by flagging or blunder under time pressure. Of course it's a subjective question, but in such situations how do people usually improve their play for a certain time control? I could imagine serious arguments for playing faster time controls (improve time management?), slower ones (practice sound calculation?), simply playing more games, or doing more puzzles.
|
|
#
?
Dec 13, 2021 06:54
|
|
|
The answers I usually see bandied about are slower time controls, analyse your games, and do puzzles/puzzle rush every day. Just mindlessly bashing out more games or faster games won't help you improve at spotting tactics or analysing positions, it seems.
|
|
#
?
Dec 13, 2021 08:30
|
|
|
Aggro posted:The Stafford gambit remains the most fun opening, although I think most people have watched the same Eric Rosen videos I have and know how to stop it. But it's real fun when they haven't. Multiple moves assaulting the hapless f2 square and repeatedly double-checking the King into oblivion. sorry but the actual most fun opening is the traxler counter gambit.
|
|
#
?
Dec 13, 2021 09:00
|
|
|
Hand Knit posted:So the central broad idea here is that, apparently, black's queen needs to stay in position to check from behind. Nepo lost that when he played Qe6. As for why Qc2 or Qb1 specifically, I would guess that (1) it spots white's rook, preventing white from playing the immediate Nh5, and (2) against Kg4 black preserves the option of Qd1+ (and I guess you then meet Kg5 with Qg1 or Kh4 with Qe1, idea being you keep the knight stuck back on g3). yeah. i saw ben finegold analyzing it and his take was the reason those two moves work is because the queen ultimately needs to land on d1 without the knight having moved first. d1 semi cuts the king off from running up the board the way it wants to, and black will be able to pin the knight. the king won't be able to make progress up the board without losing the knight and so white will be forced into a repetition eventually actually watching some grandmaster analysis of this game 6 kinda opened me up a bit more to the idea that there's more than 1 type of "equal" in a game. like, after the initial time scramble around move 40 in game 6, the game was mostly "equal" according to the engine but that really meant that white had potential winning chances but black had ways to force a draw with perfect play. the players weren't actually in an even balance of power as i normally would have thought of it. 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"
|
|
#
?
Dec 13, 2021 09:10
|
|
|
I feel it should be possible to build this kind of "difficulty" analysis into an engine. Of course it wouldn't be perfect but it could definitely distinguish between "every move draws" and "need to see 10-move line" and hand wave that into a numerical score. Possibly someone has already done this.
|
|
#
?
Dec 13, 2021 20:52
|
|
|
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
|
|
|
One of the coolest unanswered questions is if the opening position is won for white. I think its pretty neat that nobody knows that yet.
|
|
#
?
Dec 14, 2021 00:00
|
|
|
Captain von Trapp posted: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! 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.
|
|
#
?
Dec 14, 2021 00:16
|
|
|
I just had a great bullet game where I absolutely missed the critical line that would’ve led to a crushing victory. I ended up getting a break later and pawn storming my way to a win, but it’s a great position to evaluate. 1. e4 e5 2. Nc3 Nf6. 3. f4 d5 4. fxe5 Nxe4 5. Nxe4 dxe4 6. Bc4? (much better is d4) Nf6 7. d3 Nxe4 8. Bb3? In most games I am way too passive and don’t capitalize on opportunities to start an attack. I would often play something like Bf5 to protect the pawn. But after quickly debating between Qh4+ and Bg4+ (turns out it doesn’t matter much), I went with 8. …Bg4+ 9. Nd2 Qh4+ 10. g3 Nf3+ 11. Kf2 Bc5+ 12. d4  So I have a very small material advantage, but White’s King is being assaulted by active pieces. However my Queen and Bishop are hanging and if I let White’s Queen become active, I can piss away the whole advantage. Black has one move that wins, and a couple of moves that at least keep a small advantage. Everything else loses for Black. 12. …Qf6, threatening the discovered check. Basically anything White does eventually forces the pinned Knight to move Nf4, losing the Queen to Bd1. If White tries to preserve the Queen with say 13. dxc5 Nd2+ 14. Kg1 e3! threatens mate on f2. There are a ton of other variations but all of them are bad for White. 12. …Bxd4 is also ok as it leaves Black up a pawn and White’s pawn structure in shambles. 12. …Qe7 apparently also works according to Stockfish but I don’t really see why.
|
|
#
?
Dec 14, 2021 00:25
|
|
|
I feel like you could build a machine learning model that estimates how many centipawns the average player of a certain rating will lose over the next few turns. It wouldn't be too different from the bots that attempt to mimic human play, I don't think. How accurate they'd be is another question, though.
|
|
#
?
Dec 14, 2021 00:38
|
|
|
The lichess bot seems to do a good job, right down to blundering its way into a loss like I do sometimes.
|
|
#
?
Dec 14, 2021 00:48
|
|
|
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
|
|
|
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. Definitely I can recommend this entire video: https://www.youtube.com/watch?v=DpXy041BIlA&t=2147s But the timestamped highlight is where he programs exactly this engine, consuming 21 billion positions from lichess. There are some problems with this approach that are pretty interesting. The most significant problem is that around half of all chess positions have only ever occurred exactly 1 time. Salt Fish fucked around with this message at 03:00 on Dec 14, 2021 |
|
#
?
Dec 14, 2021 02:58
|
|
|
fart simpson posted:sorry but the actual most fun opening is the traxler counter gambit. weird way to spell halloween gambit
|
|
#
?
Dec 14, 2021 04:15
|
|
|
Spokes posted:weird way to spell halloween gambit fun fact: this is also my most common opening as white. so i agree it owns. e: i have 451 halloween gambit games played as white on lichess and ive won 61% of them and i apparently only have 59 traxlers as black, of which i've won 56% fart simpson fucked around with this message at 05:58 on Dec 14, 2021 |
|
#
?
Dec 14, 2021 05:52
|
|
|
i have some halloween gambit lines memorized up to 10+ moves. this one is my absolute favorite opening which i sadly have only landed a single digit number of times: 1. e4 e5 2. Nf3 Nc6 3. Nc3 Nf6 4. Nxe5 Nxe5 5. d4 Nc6 6. d5 Ne5 7. f4 Ng6 8. e5 Ng8 9. d6 cxd6 10. exd6 Qf6 11. Nb5 Kd8 12. Be3 a6 13. Bb6+ which leads to this wonderful position  black will be forced to shuffle his king back and forth between his two remaining squares while white is free to just really rub his nose in it, taking the rook, a pawn, and the queen with checks on every move until going in for the kill with the white queen
|
|
#
?
Dec 14, 2021 06:10
|
|
|
cheetah7071 posted:I feel like you could build a machine learning model that estimates how many centipawns the average player of a certain rating will lose over the next few turns. It wouldn't be too different from the bots that attempt to mimic human play, I don't think. How accurate they'd be is another question, though. D34THROW posted:The lichess bot seems to do a good job, right down to blundering its way into a loss like I do sometimes. You might be talking about these already, but Maia Bots are neural network-powered(!!!) bots designed to choose the same move a human player would at various levels (Maia1 = 1100, Maia5 = 1500, Maia9 = 1900) and are a good alternative imho to the standard lichess bots. I think they play in a fairly human fashion although with less variety and they are less prone than lichess bots to make massive, massive blunders which can feel like the bot is throwing the game.
|
|
#
?
Dec 14, 2021 09:59
|
|
|
https://www.chess.com/news/view/magnus-carlsen-defend-world-chess-title-alireza-firouzjaMagnus Carlsen posted:"If someone other than Firouzja wins the Candidates Tournament, it is unlikely that I will play the next world championship match," This really rubs it in to Nepo. I rolled over you so easily, it wan't even fun. edit: I was always team Nepo, but this makes me even more bummed he didn't win game 6. With 5 equal games and a loss, I bet Magnus would have had sufficient motivation. Doctor Malaver fucked around with this message at 20:51 on Dec 14, 2021 |
|
#
?
Dec 14, 2021 19:54
|
|
|
Salt Fish posted:One of the coolest unanswered questions is if the opening position is won for white. I think its pretty neat that nobody knows that yet. Plenty of people know it is not a forced win for white with reasonable certainty. We don't really need the hard proof of a 32 piece tablebase to say that.
|
|
#
?
Dec 14, 2021 20:41
|
|
|
Sub Rosa posted:Plenty of people know it is not a forced win for white with reasonable certainty. We don't really need the hard proof of a 32 piece tablebase to say that. I don't really understand... most things about chess (or machines), but i'm not sure how they would arrive at this conclusion. Do you have a link or an explanation or anything i could read?
|
|
#
?
Dec 14, 2021 20:50
|
|
|
I think it's basically heuristic, with the main one being that draw rates tend to increase with more skilled play (as I understand it this extends to the top-level machine play).
|
|
#
?
Dec 14, 2021 21:26
|
|
|
Spokes posted:I don't really understand... most things about chess (or machines), but i'm not sure how they would arrive at this conclusion. Do you have a link or an explanation or anything i could read? It's a combination of a ton of things. I'm a titled correspondence player, and correspondence allows engine use. We have a deadlock of draws once you get above around 2300 rating on ICCF. Anything that isn't a draw past that is due to something like a transposition error, typically. Probably lower than 2300 since NNUE. Used to, the people who play correspondence could study engines and spend hours looking for flaws in engine play to find winning lines that the engine would misevaluate and see if other players would trust the misevaluation. But since NNUE basically everything is a draw. There just aren't serious enough flaws anymore with NNUE presuming sufficient depth/time/cpu. Even with the sharpest lines with both sides trying to win, still a draw. Basically the only real room for improvement in engines is a lower time controls. Engines occasionally win or lose a game in blitz, and development is focused there. The idea of which engine is better is basically based on these short time controls. 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. Even in human chess, a super GM who wants a draw as black knows how to park the bus and get one. To try to break through white has to basically play something dubious to get outside of theory and white ends up getting punished just as often for not accepting the drawing line. You can never get an engine into that sort of trouble. Like in terms of prep the thing to do these days is to find these positions where the engine says only one move is good, but it is a move that a human probably won't play. You don't look for objectively winning lines, because there just aren't any. You look for the best chances to cause your human opponent to make an error. So in short, to win a game of chess, you basically must force an error from your opponent. Everything I know of chess and engines leads to this conclusion. Which means "chess with perfect play is a draw" is not worth doubting enough that we should shy away from saying "we know chess is a draw" just because we will never be able to formally prove it.
|
|
#
?
Dec 15, 2021 00:28
|
|
|
That reminds me: Is there any NN chess engine that is trained in a format that values draws as less then 1/2 ? I assume it would play somewhat differently.
|
|
#
?
Dec 15, 2021 00:41
|
|
|
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
|
|
|
Has any tournament tried weighting draws? Like 2/3 of a point for a Black draw and 1/3 for a White draw?
|
|
#
?
Dec 15, 2021 00:57
|
|
|
Aggro posted:Has any tournament tried weighting draws? Like 2/3 of a point for a Black draw and 1/3 for a White draw? Not directly, but you will have tournaments with tiebreakers that include things like "score with black."
|
|
#
?
Dec 15, 2021 01:00
|
|
|
|
| # ? May 31, 2024 15:15 |
|
|
Sub Rosa posted:I'm a titled correspondence player, You got it, congrats!
|
|
#
?
Dec 15, 2021 01:19
|
|
