Oh just piss off!>

Take it eazeeee <Scazzo>

NACKAMORI

In the following position Aronian has just played 60. Rc3:

click for larger view

This is a blunder since 60...Rxc3+ wins easily. Naka instead played 60...Re2 which blows the win but should still be easily drawing. Yet he managed to LOSE. The losing move occurred here:

click for larger view

63...Rc8 draws since it forces 64. Rd7, freeing the black king and pawns. Instead Naka played 63...f3 which is too slow due to 64. Rd8 and the threat of c8=Q+ is unstoppable.

The kicker here is that he actually entered the endgame up 2 pawns. The position after white's 50th move was this:

click for larger view

No change there, it seems after all the losses he experienced in the hands of Carlsen he just hid that character under the rug but with the sign of first success the Mr Hyde's alter ego came back in.

<“No matter how badly you play, unless you make a flat-out blunder, there’s always going to be some narrow path to being able to save the game and draw instead of losing,” Nakamura told me. One consequence of this shift has been a certain parity at the top ranks. The Carlsen-Caruana world-championship match featured 12-straight draws before Carlsen finally prevailed in a tie-break. Compare that to Fischer’s day, when he reeled off 20 straight wins during his march to the world title. Players today may go 15, 20 moves or more into the game before they’re thinking on their own, relying instead on computer-driven preparation. “Everyone is using the same programs, everyone is looking at the same opening ideas,” Nakamura said. “I wouldn’t say everyone is necessarily the same in terms of talent or ability, but when you’re able to prepare games that go so deep that you don’t have to think, really, it balances out the field.

The only difference between a human and an engine is about calculation. It's not that the engines have the better ideas. Their knowledge is simply based on what humans have gathered.

Reads like Nakamura is having a midlife crisis chess wise. He is out of the top 20 (he is currently 21!) and he fears his spark has gone.

The good guys rarely give him a chance to produce W So vs Nakamura, 2015 (Gosh was that 4 years ago - seems like it was 4 months ago, where is the time going? - still remember the beautiful missed win and people looking for something upsetting in the pun that was not there.)

If the invites dry up his style and quality of play should allow him to display his skill and to clean up in the Opens.

He has won Gibraltar four times, three times on the trot 2015-2017 a tournament I've been informed does not pay the players appearance, travel or lodging fees, (the arbiters yes, the players no.) a genuine Open.

It used to be like that, but now with the latest exploits of AlphaZero (and lately Leela) this seems to be changing. For instance; in these games that AZ played against Stockfish, some peculiar ideas emerged, like some really crazy piece sacrifices, and that game where it moved its queen to the corner of the board (and it worked!), things like that. So even obtaining new ideas on how to play our beloved game is no longer exclusive to humans, it seems.

They have 'accepted' the 'idea' that a human cannot beat the engines. It's pure a matter of defeatism. The answer is effort. It's no coincedence 'they think' that 'every game' will eventually end up in a draw. Go figure!

Pretentious imposters as they are! Just because they spend some more time on chess than the average patzer does not say they have nothing to prove to us.

It's just like saying: "Hey, we are going to the moon - be sure that the cams are off when we cross the van Allen radiation belt..."

Obviously, the 'scientists' of today can 'perfectly' explain to you how to get an astronaut through the belt. Little wonder... as if anyone ever asks for 'evidence.' Sure. Money yes. Evidence no.

Or somewhat more realistic: "Let's be sure the cams are off when our main man commits 'suicide' in 'jail.'" Evidence!

Still, they have no proof. At all. About anything. Only their pretension is what rests. The status quo of the so called 'scientists.'

Misery does not only love company, it screams from the rooftops for all the company it can get!

Now... look at you, Mr. Nakamura.. what do I perceive? The World anno 2019.

AZ is a marketing trick.
Tell me.. why was Leela aborted?
Why does AZ plays prepared opening lines? Why did Leela use prepared opening lines?

I thought it learnt everything by itself... how weird!

Well... do I get any answer?

I read your comment after I placed mine, so it wasn't addressed to you. Just that you might have been asking yourself.

But I have done some experiments myself. SF plays a draw against itself after 1. a4

So much for the common theory standard of today.

Sorry, but I do not drink. Alcohol is pure poison. It kills the brain more than any other drug.

New 'ideas' is the wrong word, the play of Alpha is inspirational, inspiring players to think in other avenues.

Some of these (OK new ideas) were known, the positional weakness of an exposed or cramped King.

A.Z. seems to have singled out the King and the rule it must move when attacked as the prime weakness of chess and taken it up to a higher level. Nothing else matters, The King..The King...The King... It's more fixated towards the enemy King than Nimzovitch was towards the IQP.

Very refreshing and inspirational.

Nigel Short is correct: '“Modern Chess is too much concerned with things like Pawn structure. Forget it, Checkmate ends the game.”

You should pick up a copy of "Best Play – A New Method for Discovering the Strongest Move", by Alexander Shashin, 2013. In it he describes 3 algorithms (I would call them "guidelines" but no matter) for evaluating chess positions, the Tal Algorithm (for attacking material chess targets), the Capablanca Algorithm (for finding the strongest strategic move), and the Petrosian Algorithm (defensive), plus 3 hybrid algorithms, TC (Tal/Capablanca), CP (Capablanca/Petrosian), and the TCP (Tal/Capablanca/Petrosian), a hybrid of all 3.

He also provides a quantized methodology for determining which of the 6 algorithms to use based on material, time (i.e. development), king safety, compactness of the position, and space (expansion). These ideas have been implemented as the evaluation function for the chess engine ShashChess, a version of Stockfish, which for version 8.0 is rated at 3456 in the current (Oct 26, 2019) CCRL 40/40 tournament, not far behind from the current Stockfish 10 rated at 3464. ShashChess is the 5th ranked chess engine in the CCRL list (Stockfish 10 is 2nd ranked engine), and just behind 4th ranked LeelaC0 version 0.22.0 when supported by a GTX1050 GPU. So the considerations of Shashin seem to be a reasonable method of implementing the evaluation function of a "classic" chess engine.

Why am I bothering with this? Well, it's hardly new, original, or inspirational that when players have an advantage in several chess position factors they are obligated to, per Shashin, "(1) attack (openly or directly), (2) place their pieces on their best attacking squares, (3) sacrifice, and (4) win material." And when one attacks one should attack the highest valued piece, which is obviously the king, followed by the queen, etc., also per Shashin. Good chess players have been using these guidelines for attacking their opponent's king for centuries, even if they could not quantize them so succinctly.

I originally had high hopes for AlphaZero, particularly its use of reinforced learning without "human prejudices", which could possibly allow it to find chess principles that had not been found before. A similar claim was made about AlphaGoZero, where the articles alluded to moves based on principles that had not been known before. Since I am not familiar with Go I cannot tell whether that was true or not.

But I am somewhat familiar with chess as well as computer hardware and software. And in my research of AlphaZero I have not yet been able to find any truly new or innovative concepts that had not been either published or implemented before. What is original with AlphaZero is its use of Google-proprietary Tensor Processing Units (TPUs) which provided it with what I estimated to be about an 80X computational processing performance advantage over the hardware used in the AlphaZero vs. Stockfish matches in 2017 and 2018. AlphaZero's superiority over Stockfish in their matches can pretty much be directly traced to that, as Deep Mind's second paper on AlphaZero shows. Yet GM Mathew Sandler, of "Game Changer" fame, never addresses this computational performance advantage enjoyed by AlphaZero at all.

AlphaZero has somehow captured the imagination of the general public and chess IMs and GMs, for reasons that I don't understand. Chess engines have been criticized for using "brute force" in their computational-intensive approach to playing chess. Yet the most computational-intensive approach to playing chess, neural network-based engines using TPUs and GPUs (which in the TCEC GPU server provides what I estimated to be a 15.7X edge in performance to compete on roughly equal terms with Stockfish 10 using the GPU server) are for some reason praised for the "human-like' moves that they play. Go figure.

And they inspire comments like this one from IM Anna Rudolph (https://www.youtube.com/watch?v=LPu..., starting at 25:57 with 45.Ra1) who I think should know better:

"We know that this rook loves a corner [but] what is the purpose of this move? Why would AlphaZero want it's rook on a1?"

"For [over] 20 moves the rook was kept on a1. And you wonder why it was placed there? Well, AlphaZero just understood the position so much deeper than we can."

Somehow the simplest explanation, that AlphaZero simply made a mistake in that position and moved the rook from it's best square, d1, to a square where it did nothing for 20 moves, and then had to move it back to d1 to force the issue never seemed to have entered her mind. I guess that she, like many others, was blinded by the obvious.

Hi AylerKupp,

The game Anna is talking about is this one - AlphaZero vs Stockfish, 2018

if you just go forward a few moves after 45. Ra1 and the 'Why would Alpha want it's rook on a1." makes a sense.

Here Black to play it's 49th move.

click for larger view

If the a1 Rook was still on d1 Black would have 49...Qxa4 (hitting the d1 Rook if it stayed there.). It looked ahead 4 moves and does not seem to be, as you appear to think 'Alpha made a mistake moving it from d1 to a1.'

and 'For [over] 20 moves the rook was kept on a1.' (did Anna say that, I've not gone though the vid)

If we go through the game the Rook on a1 was waiting to cover the only place counter play was going to come from whilst the main play was all about those passed centre pawns. When the Rook did go back to d1 it did so with a one move threat to win a Rook.

S.F. is so impressed with the a1 Rook it condemns a Rook to a8 to hold the a7 pawn and tried to get something going pushing the a7-pawn this was never going to happen with the Rook on a1.

I'm inclined to think it knew what it was doing playing the Rook to a1 and it was not a blunder but a pretty deep move.

<<Sally Simpson> The game Anna is talking about is this one - AlphaZero vs. Stockfish, 2018>

I was just repeating what she said in https://www.youtube.com/watch?v=LPu... starting with 45.Ra1 at around 25:57. To elaborate further she said the following

"We know that this rook loves a corner [but] what is the purpose of this move?"

"Why would AlphaZero want it's rook on a1?"

"I've seen the game and know why." (but later never explains why, instead she says "I don't know")

"For [over] 20 moves the rook was kept on a1. And you wonder why it was placed there? Well, AlphaZero just understood the position so much deeper than we can." Yes, Anna did say this at about 37:00 minutes into the video.

Frankly I expected better from an experienced IM analyst. If you were able to provide an explanation as to why you think that White's rook belongs on a1 rather than d1 (and I don't think that you are at IM playing strength level but I may be wrong), then I think that she should have been willing and able to explain why instead of teasing us about it. But that's just my opinion and personal preference. It could simply be that I don't care for her analysis presentation style, at least when it comes to here analysis of AlphaZero (I like most of her other analyses), but that could be my problem, not hers.

Unfortunately for me, once I started looking at this game again I got interested as to the merit of the various moves. And I stand corrected regarding 45.Ra1. If White does not move his rook from d1 then, as you said, it will be attacked by Black's queen if Black is allowed to play ...Qxa4. That move still looks ugly to me since it seems that Black goes pawn grabbing and removes the queen from where the action will soon be, on the k-side. It's almost as if AlphaZero is offering a pawn sac to deflect Black's queen from the defense of its king. And, if that's the case and the pawn sac should not be accepted, then both AlphaZero and Stockfish should be commended, AlphaZero for offering the pawn sac and Stockfish for declining it. But I am clearly not anywhere near IM strength playing level so I can't accurately analyze the consequences of ...Qxa4 and must fall back on engine-based analysis.

<A. "Best" 45th White move> After 44...Nb7 we have the following position:

click for larger view

Here is a summary of the 3 engines' evaluations, sorted in order of descending Ratings-Weighted Average (RWAvg) since it is White's move. It seems reasonable to proportionally weigh the evaluation of the highest averaged rated engine (ShashChess 8.0 @ 3362) more than the lower rated engines (Houdini 6 @ 3322 and Komodo 12.3 MCTS @ 3180) but in practice there isn't much difference between the RWAvgs than the straight averages. Note that these are the average ratings of the CCRL 40/40 and CEGT 120/40 1-CPU versions of these engines since these time controls most closely approximate the time control used in the AlphaZero vs. Stockfish match, so they won't match the ratings listed in either the CCRL and CEGT engine tournament lists.

a. The engines did not agree as to what were the top 5 moves. So moves not reported by the engines as one of their top 5 moves are considered to be null when calculating the averages, and averages are not really applicable.

White's Houdini 6 Komodo 12.3 MCTS ShashChess 8.0

Move d=28 d=28 d=46 <Avg> <RWAvg> <TrueRank>

---------- ---------- ----------- ---------- ---------- ----------- -----------

45.Rdb1 [+0.55] [+0.45] [+0.64] <[+0.55]> <[+0.56]> <1>

45.Rd2 [+0.38] [+0.42] [+0.48] <[+0.43]> <[+0.43]> <1>

45.Ra1 [+0.31] [+0.40] [+0.48] <[+0.40]> <[+0.40]> <1>

45.Bf3 [+0.25] [+0.43] [+0.48] <[+0.39> <[+0.39]> <1>

45.Re1 [+0.29] [------] [------] N/A N/A N/A

45.Qf4 [------] [+0.38] [------] N/A N/A N/A

b. To calculate reasonable move evaluation averages, moves not reported by the 3 engines as one of their top 5 moves are considered to be evaluated equal to the evaluation of the 5th best move less the average of the absolute differences between each of the evaluations. This is likely to be a more realistic value than just one centipawn below the evaluation of the 5th best move, but it's still more or less a guess.

White's Houdini 6 Komodo 12.3 MCTS ShashChess 8.0

Move d=28 d=28 d=46 <Avg> <RWAvg> <TrueRank>

---------- ---------- ----------- ---------- ---------- ----------- -----------

45.Rdb1 [+0.55] [+0.45] [+0.64] <[+0.55]> <[+0.56]> <1>

45.Rd2 [+0.38] [+0.42] [+0.48] <[+0.43]> <[+0.43]> <1>

45.Ra1 [+0.31] [+0.40] [+0.48] <[+0.40]> <[+0.40]> <1>

45.Bf3 [+0.25] [+0.43] [+0.48] <[+0.39]> <[+0.39]> <1>

45.Re1 [+0.29] [+0.35] [+0.44] <[+0.36]> <[+0.36]> <1>

45.Qf4 [+0.18-] [+0.38] [+0.44] <[+0.33]> <[+0.34]> <1>

45.Rg1 [-+0.18] [+0.35] [+0.48] <[+0.33]> <[+0.34]> <1>

True Rank: 1 = [ 45.Rdb1, 45.Rd2, 45.Ra1, 45.Bf3, 45.Re1, 45.Qf4, 45.Rg1 ]. Since all the move evaluations are within [0.50] of each other, each of these moves can be considered equally good from a practical perspective. So 45.Rd1 was not a mistake, but was no better and no worse than the other suggested moves. Or it could be that AlphaZero could see further than these engines, but given that it uses the same MCTS algorithm as Komodo 12.3 MCTS, I'm not sure of that, even with AlphaZero's overwhelming computational performance advantage.

And here is a summary of how the 3 engines ranked their top 5 moves, without regard for the value of the evaluation. When the evaluations for the moves are very close, this provides an additional guide for the move rankings. After all, if we're determining what move to play according only to what the engines say, we'll likely want to play the best move determined by the engine regardless of the actual value of the evaluation. Which is OK as long as we don't care to know how good or bad our position is relative to our opponent's.

White's Houdini 6 Komodo 12.3 MCTS ShashChess 8.0

Move d=28 d=28 d=46 <AvgRank> <TrueRank>

---------- ---------- ----------- ---------- ---------- -----------

45.Rdb1 1 1 1 <1.0> <1>

45.Rd2 2 3 2 <2.3> <2>

45.Ra1 3 4 2 <3.0> <3>

45.Bf3 5 2 2 <3.0> <3>

45.Rg1 6 6 2 <4.7> <4>

45.Re1 4 6 6 <5.3> <5>

45.Qf4 6 5 6 <5.7> <5>