Chess

[abnaxus]

 

[abnaxus]

Carlsen's gf is appropriately named Larsen

 

[taxalot]

Chess players are not supposed to have cute girlfriends. In fact, they are not supposed to have girlfriends at all.

Leave Chess to computers.

 

[Country_Gravy]

She is serviceable.

 

[Mangoose]

Haven't played in forever, but I remember typically opening as a newbie with King's Gambit or Guioco Piano (or Ruy Lopez). I can't remember when but at some point I decided to use the Smith-Morra against Sicilian. Have a tendency to sacrifice everything for initiative and tempo lol.

(Now imagine gambiting while playing drunk chess, where you take a shot for every piece you lose. Oh and add some kind of urge for a midgame tactical win)

 

[Mangoose]

http://www.chesspersonality.com/

Interesting. Seems kinda accurate. I tend(ed) go for solid positions so I have better venues to openings for attack. If I play for initiative/tempo it's for setting up a "ready" position rather than pushing hard for position (and I find aggressive positional play risky in terms of opening myself up I suppose). Typically for midgame victories (or a midgame victory in itself) so that's probably why positional strategy itself is not the aim. I do lean towards -some- unorthodox moves to stymie enemy plans/avenues of attack/etc rather than in a calculating/grinding manner. I'm positionally-oriented via "acute" actions rather than big-picture positioning I suppose.

However I'm definitely more emotional than calm :raaage:

 

[Mangoose]

If anyone's on redhotpawn, make a game and invite me (username: sabishii) or list your username.

 

[Mangoose]

Playing a few games on redhotpawn to verify the recommended openings :p but it's helping me memorize/understand some openings.

 

[Country_Gravy]

No chess.com players?

 

[luj1]

Still the GOAT game

 

[abnaxus]

 

[VentilatorOfDoom]

https://en.chessbase.com/post/the-future-is-here-alphazero-learns-chess

Fascinating read about how a program that had been given zero chess knowledge, opening related or otherwise, just by using a General Reinforcement Learning Algorithm did learn to play Chess so well in a very short time, that it just beat the strongest chess program to date.
And it did so on significantly weaker hardware.

 

[spectre]

Wasn't stockfish essentially running without its openings database?
Obviously they rigged it a bit to make their wonder boy shine, but it's still pretty impressive as a machine learning effort.
I also read it doesn't really play in a way a "human" would at all, whatever this means, this bit is also quite interesting.

 

[HansDampf]

https://github.com/CYHSM/chess-surprise-analysis

Someone analyzed game 10 with SF, in which A0 made a positional sacrifice.

The world of chess changed considerably after the introduction of AlphaZero to chess. As described here the DeepMind engine won a competition against Stockfish 8 (28 wins - 72 draws - 0 losses). As this library was set out to discover surprising moves in chess games (see below for describtion and examples), I wanted to see if a reanalysis on the games of AlphaZero with the same engine brought some inside into the thought process of Stockfish 8.

Game 10
In game 10 there were probably a lot of surprising moves but one of the most daring moves was move 19 by white (AlphaZero). In this position it decided not to save its knight (Ng4) but to play Re1.

After analysing the game with this library we get this heatmap which shows us the evaluations of Stockfish 8 over each depth for each half-move.

As one can see move 19 was also surprising to Stockfish (half-move number 38) where even for very high depths it still believed his position to be advantageous. The first time Stockfish actually saw a superior position of AlphaZero was at half-move 65.

blue = black (SF) is better
red = white (A0) is better

So, after the sacrifice, it took SF 13 more moves until it suddenly realized that its position was hopeless.

 

[coldcrow]

https://en.chessbase.com/post/the-future-is-here-alphazero-learns-chess

Fascinating read about how a program that had been given zero chess knowledge, opening related or otherwise, just by using a General Reinforcement Learning Algorithm did learn to play Chess so well in a very short time, that it just beat the strongest chess program to date.
And it did so on significantly weaker hardware.

Wrong on 2 accounts. While in the actual match 28k pos vs. 70 mill. doesn't sound impressive, it neglects the absolute bastard 5064 TPU (tensor processing unit: specialized CPUs for neural network calcs, e.g. statistics climbing, monte-carlo-searches, etc.) horde doing the actual work before the match = training the neural networks. When the match was underway instead of having to calculate very far for many candidate moves, it could rely on "knowing" promising variations, wihout having to calculate far, because the actual calculating work was done before.
Also during the match the program ran on TPUs which can efficiently do the tree climbing.
What is new in this approach is the learning algorithm, and the available computing power to brutalize the problem.
So in the end it resembles a famous saying by reti (I think), when he was asked about how many moves deep he was calculating: "One. The best."

EDIT: TPUs are actually not unlike GPUs. They are optimized for quickly crunching through simple, low-precision operations, so are perfect for dealing with large masses of data being computed, which is the case in machine learning.
Also see use of GPUs for bitcoin-mining.

 

[VentilatorOfDoom]

Yeah that info was revealed later, there's a new article about it, the response of the stockfish team. In the article linked by me they made it sound as if that program ran on weaker hardware when this probably isn't even true, plus stockfish was limited to 1min per move which seems... weird.

I know that Reti saying differently:
How many moves do you calculate ahead? One more than my opponent.

 

[abnaxus]

not a bad channel

 

[Burning Bridges]

https://en.chessbase.com/post/the-future-is-here-alphazero-learns-chess

Fascinating read about how a program that had been given zero chess knowledge, opening related or otherwise, just by using a General Reinforcement Learning Algorithm did learn to play Chess so well in a very short time, that it just beat the strongest chess program to date.
And it did so on significantly weaker hardware.

I always find it interesting when all Chess knowledge is thrown overboard, and the Chess world with it's mostly just memorized strength begins to shake.

A very nice example is the alleged "Short vs Fischer" which probably was Short vs "Some Grandmaster Troll with a Computer"

Fischer's sequence 3. Kf7 4. Ke6 5. Kd6 6. Kc7 is simply hilarious, when you think that the opponent was the almost World Chess Champion Nigel Short, and that he lost the game



If you want some of that, simply open b7-b6 https://www.amazon.com/Play-b6-Dynamic-Hypermodern-Opening/dp/1857444108 I also move my king nowadays on the second rank, whenever possible, and so far it always worked.

 

[coldcrow]

This is ofc pretty much BS, if you just analyze it for a bit. It's no wonder that AlphaZero's openings overlap with some of the most successfully employed by the human elite. And these hypermodern openings lead to slightly inferior for black in the best case. They tend to work if the opponent play stereotypical as they pose some unique problems and in blitz/rapid games.

 

[Burning Bridges]

They tend to work if the opponent play stereotypical as they pose some unique problems and in blitz/rapid games.

But this seems to be the case. Chess is by a large part a game of memory, and it's hilarious when you start eg a King attack or move back your knights and the opponent has no countermove.

 

[abnaxus]

A human can make some troll moves in the beginning like in those supposed Fischer games and still easily win against any other human by switching to an engine.

 

[abnaxus]

Svidler is awesome

 

[Mustawd]

Haven't played in forever, but I remember typically opening as a newbie with King's Gambit or Guioco Piano (or Ruy Lopez). I can't remember when but at some point I decided to use the Smith-Morra against Sicilian. Have a tendency to sacrifice everything for initiative and tempo lol.

Guioco Piano was my shit.

I kinda lost interest after they decided it was a good idea to play Blitz chess to decide the outcome of the world championship in case there was a draw. What utter garbage.

 

[Mangoose]

Haven't played in forever, but I remember typically opening as a newbie with King's Gambit or Guioco Piano (or Ruy Lopez). I can't remember when but at some point I decided to use the Smith-Morra against Sicilian. Have a tendency to sacrifice everything for initiative and tempo lol.

Guioco Piano was my shit.

I kinda lost interest after they decided it was a good idea to play Blitz chess to decide the outcome of the world championship in case there was a draw. What utter garbage.

what the FUCK?!!?

 

[Mustawd]

https://www.chess.com/news/view/tiebreak-to-decide-world-cup-final-after-4th-draw

The final of the FIDE World Cup between Ding Liren and Levon Aronian will be decided in a tiebreak on Wednesday as the fourth game also ended in a draw. It was again Aronian who had the better chances.

After winning their dramatic semifinals with which they reached the Candidates', Ding and Aronian slowed down a bit. All of their games in the World Cup final ended in draws, and so we'll see at least two rapid games tomorrow, and possibly blitz and Armageddon. The winner will earn $120,000 whereas the loser takes home $80,000.

https://www.chess.com/forum/view/ge...shouldn-t-define-a-classic-world-championship

I think it's a result of the problems with past formats. When the 24 game match format was in vogue, where the Champ retained the title on a tie, there were complaints that since GM's playing for a draw can very often achieve it, the player who got an early lead simply started playing for draws. So they changed to the winner must win 6 games in 1978 and Karpov-Korchnoi played 32 games, with the much older Korchnoi exhausted and at a disadvantage at the end.

Today's format, including rapid games, is perhaps the only way to force more wins into the match while keeping to a reasonable schedule.

https://worldchess.com/2016/11/30/guide-to-the-world-championship-tiebreakers/

Spoiler

The World Chess Championship ended regulation play after 12 games with the match score tied at 6 points apiece for Magnus Carlsen, the reigning titleholder from Norway, and Sergey Karjakin, the challenger from Russia. To break the tie, the two competitors will play a series of games at faster time controls.

The first tiebreaker will be a series of four rapid chess games in which each player will have 25 minutes to start and have 10 seconds added to their time after each more. If, after the four games, the score is still tied, then the players will proceed to the second tiebreaker.

The second tiebreaker will be two blitz games, in which each player starts with five minutes and has three seconds added to his time after each move.

If the score is still tied after the two blitz games, they will then play another set of two blitz games. If the score continues to be tied, they will play blitz games up until they have played a total of five sets of two.

If the score is still tied, they will play an Armageddon game for the title. The players will draw lots to see who has the choice of color. In the Armageddon game, one player will have White and start with five minutes, the player with Black will have only four. Neither player will receive more time unless the game goes 60 moves, at which point each will begin to receive three seconds after each move. The player with White will have to win in order to win the title; the player with Black only needs to draw in order to be declared the winner.

A full guide to the match regulations can be found here. The part concerning tiebreakers starts on page 3 under section 3.7.1.a.