I find it frustrating and annoying that a program like Stockfish can beat any human by it's mastery of "the rules." I find it exhilarating that alpha zero can brush aside stockfish like so much chaff by the most OUTRAGEOUS human trait of breaking all "the rules" when they stand in the way of winning.
A theme of these games seems to be stockfish being lured into accepting pawn sacrifices, and then being beaten by a very prolonged attack that defeats its lookahead capability. I wonder if alphazero, playing the other side, would decline those sacrifices...and how much price there would be to pay for that. In other words are the offers really SOUND or are they just tricks that work against traditional engines?
I've been liking this content with Matthew and AlphaZero a lot. I don't play chess at all but enjoy these videos as informative entertainment. Keep it up!
Good lord...no human on Earth is anywhere near the playing strength of Stockfish...let alone Alpha Zero. From what I have seen, Alpha Zero is definitely the strongest chess player on planet Earth.
Stockfish 8 is very outdated -- at least 100 elo weaker than the latest sf10 versions. Until the DeepMind team publishes the results of a fair match between the latest a0 and the latest sf10 (i.e., with comparable hardware, a sufficient sample size of games, proper time controls, proper opening books and endgame tables, etc.), it will be impossible for anyone to really know what the "strongest chess player on planet Earth" really is.
@Cscuile I don't think it matters that is was Stockfish 8. Leela is inferior to Alpha Zero and it can win games against Stockfish 10. I think Alpha would beat any version of Stockfish.
@@Varney_of_London Cscuile made the point that based on the testing, we have every reason to believe Stockfish10 is the strongest engine right now. An engine winning a game against another engine doesn't mean it's better. Stockfish 8 won 6 games against AlphaZero that were just published. It seems like you're a 1200 chess player or less. You don't seem to know much about the game.
@@Varney_of_London Leela is still around 120 elo weaker than Stockfish10. I always find it weird when low-rated, low information folks like yourself speak in such a strange, arrogant way like you know everything
I think that too many people are focusing on the game, which I also follow, as if this were an ordinary player. Since I have significant knowledge, and since I believe that Hawking and Musk were right, I am really anxious by the self-taught nature of this AI. This particular AI (and its more generalized, even more recent variant MuZero) is not the worrisome thing, albeit it has obvious, potential applications in military logistics, military strategy, etc. The really scary part is how fast these were developed after AlphaGO debuted. We are not creeping up on the goal of human-level intelligence. We are likely to shoot past that goal amazingly soon without even realizing it, if things continue progressing as they have. The early, true AIs will also be narrow and not very competent or threatening, even if they become "superhuman" in intelligence. They will also be harmless, idiot savants at first. Upcoming Threat to Humanity. The scary thing is the fact that computer speed (and thereby, probably eventually AI intelligence) doubles about every year, and will likely double faster when super-intelligent AIs start designing chips, working with quantum computers as co-processors, etc. How fast will our AIs progress to such levels that they become indispensable -- while their utility makes hopeless any attempts to regulate them or retroactively impose restrictions on beings that are smarter than their designers? At first, they may have only base functions, like the reptilian portion of our brain. However, when will they act like Nile crocodiles and react to any threat with aggression? Ever gone skinny dipping with Nile crocodiles? I fear that very soon, before we realize it, we will all be doing the equivalent of skinny dipping with Nile crocodiles, because of how fast AIs will develop by the time that the children born today reach their teens or middle age. Like crocodiles that are raised by humans, AIs may like us for a while. I sure hope that lasts. As the announcer in Jeopardy said about a program that was probably not really an advanced AI long ago, I, for one, welcome our future, AI overlords.
"The biggest threat we have to fear from AI is that some people will conclude too early that we have fully understood them." It's a wise man who said this. The problem is not the AI, its the humans. And humans make errors - some are not even recognizable and some are devastating.
@@Pintkonan Blather. Anyone who concludes that he understands an AI is confused or moronic. The ambiguous comment about humans making errors is only relevant if you are referring its applications, e.g., in war.
@@TheKunec It lost many games, only the winning games were published though. And the whole A0 vs SF is a hoax, since SF had bad parameters. SF would not lose to A0 under fair rules.
In the pdf in page 2, in the table C with TCEC openings and AlphaZero with black it loses to SF by a lot. I want to know what is the speed of the cores of the computer where Stockfish ran and how many cores, also the specs of the computer AlphaZero, how many cores and how many neural units. That will show AlphaZero is just a whole room versus a tiny computer.... TCEC is using a 128 core computer but the most important is not the number of cores but the speed. For example a 128 core with a 2 Ghz speed is inferior than just a 4 core computer with speed of 4 Ghz. Also speed of RAM is crucial too.
@@mogyorospusztai Sorry to burst your bubble but In the new matches Deep Mind did just that they gave SF what people were demanding and AZ still won the matches convincingly. AZ didn't just beat SF 8 in the new matches it also beat SF 9. Also iirc AZ also beat the Dev version. This was done at a normal classic time control. I saw somewhere online that in one experiment even with a big time odds disadvantage AZ still won.
I am impressed by Alpha Zero's play. On the other hand, regarding all the amazement and mystery that is being evoked by commentators after a win, I just wonder about one thing: I think it is a fact many of us are not really used to taking a closer look at comp vs comp chess, so we might be more impressionable to its "general peculiarities" compared to human chess. What I am asking myself in that context is: Aren't all these wondrous moves (like Ng5 here) merely standard for high-level computer chess? I mean, does not even an arbitrary game between Stockfish and Houdini, due the engines's tactical force, contain many moves and concepts that are too deep for the human eye, at least at first glance, and leave it stupefied? What I am trying to say is: Assuming it were rather the norm than the exception that any comp vs comp game contains such wow moves, what exactly - apart from the nature of the A0 engine itself (which would be a bit arguing ad machinam) - does differentiate the shown A0 vs Stockfish (8.0) games from any other game between two high-level engines where one is about 50-60 rating points stronger? In other words: What makes these games special in the end?
I would say that the fact that alphazero is not scared to sacrifice for positional value, and zugzwag while stockfish plays by mostly piece value makes these games especially special.
AZ has shown time and again that many openings which are sneered at by most humans are not that bad. I'm curious how AZ would play many opening lines I'm drooling about being about to run experiments with AZ playing openings like the Orangutan/Sokolsky vs latest SF which is SF 10 I think?
At 3:30 White was supposed to attack the weakside with a2a3. Black cant just blow through the strong side. While black is wasting moves headed for halt, black's weakside would be obliterated.
It seems a parameter for the value of a piece is not a good thing against AlphaZero. If its wins, it really shows how Stockfishs pieces can be worth much less than their standard values. But Stockfish will never find out, since it needs additional parameters for all ways in which pieces can lose value. I like to think that talented players judge positions in a similar way, feeling their way through variations and the overall dynamics in the position, before judging the importance of each piece in that "puzzle".
Exactly top players use their superior intuition and talent to navigate efficiently thru the chess tree. Many chess players think that the top players are looking at variations at super speed and these "normal" players think that's why the top players are so good. They are so wrong.
Would interesting if Mr. Sadler would look some Leela-Chess-Zero games and see if they play is similar. On few games of Leela I can see that at least this Harry boy to pin down king stratagem seems to be in Leela's toolbox. ans do recommend the book "Game Changer" lot read there. Probably wont make u stronger but highlight nicely what sort thing a0 goes for
@@yoloswaggins2161 I don't see how many cores Stockfish had and what is the speed of each core and how many cores AlphaZero had and how many neural units it had. Also how many time per game in this "chosen games". Is perhaps AlphaZero using 745 neural units and Stockfish just 20 cores or so? Even MVL computer has more cores. TCEC is using 128 cores for example.
Very entertaining and thoughtful commentary! Best of all the Alphazero vids. And I agree with him that the Dutch is dodgy but it is a lot of fun, especially for sub-grandmaster players and for chess engines that don't really understand chess, like SF! (The move c5 check by SF, terrible! Few humans would play it, yet I noticed that chess engines seem to gravitate to it, "like the moth to the flame" lol.)
The dutch was something I used to play as black regularly. Lost a lot of games. Glad to see it wasnt my decision that was lacking. Just my gameplay lol
Surprised at SF's naiveté on several crucial junctures, one being the really bad looking 22,,,Be2 -- where 22...Bg2 would have made much much more sense, and that whole "outerworldly" sac feast with threats on h1 would not be possible, or even a distant option. Strong engine to probe this, anyone?
I would happily play ten games in a row against Matthew Sadler, just to boost his rating to 2701. He might well be the strongest active player who has never broken the 2700 barrier.
Sir, is it possible to publish the analysis in writing for the upcoming videos as a reference for us, because we may find it difficult to understand the dialect because we are not English
Cant help but be bit worried about usage of AZ kind of Ai in areas where arent exact rules. It will do something suprising what we will just dont get... itll think too ahead for our slow brains .-(
And it's not even alpha one. Only alpha zero. Just the beggining, the pioneer. Alpha Zero is the Bobby Fisher of computer chess. Now let's see Kasparov and Carlsen :)
So... if AZ plays a match vs SF 10 and wins what then? SF 11 will be released and I predict that people will then say, "Try SF 11 against AZ we will see who is best." You see where this is going I think.
"Sf8 made a mistake, but sf10 acknowledge it and would not make it, therefore it will win", while completely disregarding the fact that sf10 could very well do other mistakes. And all of you people don't know a thing about neural network and their learning process, which is pretty much the reason behind the enormous amount of computational power A0 requires. Why can't you simply accept the potential of neural networks like all the goddamn world is doing? If there are individuals who stop the progress are people who think like you, and the sad part is that you don't even realise it
@@dannygjk I am a Leela fan myself but Leela hasn't demonstrated any clear superiority over SF in their battles overall, certainly not to the extent the AlphaZero papers would indicate (if anything SF has proved to be the better engine). It has become a theme that Deepmind achieves something very significant yet it gets immediately overrated by reactionaries. AlphaZero chess: incredible engine with a positional style but not something that would render the "classical" chess engines irrelevant. AlphaStar: The first starcraft AI that could stand its own against pro players, but not "without mechanical advantages through strategy". AlphaFold: Light years ahead of what other programmers were trying with protein folding, but it does not "solve" protein structures.
Can't make a meaningful comparison of the hardware because SF is specialized for one type of hardware while AZ is specialized for another type of hardware. That said SF does about 1,000 times as many nps as AZ does.
All GrandMasters will never beat these AI CHESS Engines. Magnus must have Nothing but Fear up his sleeve to even think of Playing these Engines. No doubt! Humans should stay on Earth Chess, and not Venture into Outer Space Chess.
The way I see it, AZ way of playing will always win against a rules based chess engine. It breaks rules and only cares about long term and final win. It is the AI I fear humanity has no chance of winning in the coming war. Sigh.
I've seen more incredible games of SF 10 demolishing SF 8 than this one. Because it's not put on youtube people don't know it. Until A0 wins a computer chess tournament like TCEC there is not enough evidence that it is stronger than Stockfish.
Well AZ destroyed SF 9 in at least one match so I wouldn't be so confident. SF 10 would do better than SF 9 of course but if I was betting on a AZ vs SF 10 match I would put put my money on AZ.
"Each program was run on the hardware for which it was designed (23): Stockfish and Elmo used 44 central processing unit (CPU) cores (as in the TCEC world championship), whereas AlphaZero and AlphaGo Zero used a single machine with four first-generation TPUs and 44 CPU cores (24)."
@@CP-jp8hh AZ did play SF 9, (it may have been the dev version before it was released under the name SF 9), and won convincingly by a huge margin. I am not confident that SF 10 would stand up well against AZ. SF is facing a machine that apparently has such an excellent "grasp" of chess that in spite of SF having a huge nps advantage AZ "sees" more clearly. The size of the chess tree is so huge that even at SF's speed what it sees is similar to comparing a helium atom to our galaxy. When you look at chess that way it really shouldn't be a surprise that a smarter approach cutting out almost all the inferior variations is more effective.
Do not play a game against a machine..they always win...machine vs machine? ..so this is another story which humanity is no more!...at least i'm glad that i will not be around when it happens.
I find it frustrating and annoying that a program like Stockfish can beat any human by it's mastery of "the rules." I find it exhilarating that alpha zero can brush aside stockfish like so much chaff by the most OUTRAGEOUS human trait of breaking all "the rules" when they stand in the way of winning.
This game blows my mind!!
This tatics, this level of playing chess is from another dimension :)
White: blocks Ng5 with h2-h4
Black: plays Ng5 anyway
White: ....
White: #facepalm
it's like ! hum i'd play Ng5 IF you play h4 ! x') this game is so crazy... ^^
Dear Matthew,
You are an Amazing Commentator.
Thank You
A theme of these games seems to be stockfish being lured into accepting pawn sacrifices, and then being beaten by a very prolonged attack that defeats its lookahead capability. I wonder if alphazero, playing the other side, would decline those sacrifices...and how much price there would be to pay for that. In other words are the offers really SOUND or are they just tricks that work against traditional engines?
I've been liking this content with Matthew and AlphaZero a lot. I don't play chess at all but enjoy these videos as informative entertainment. Keep it up!
Yes Matthew has a very entertaining style, as well as insights into the game!
one of the best reviews ever
Good lord...no human on Earth is anywhere near the playing strength of Stockfish...let alone Alpha Zero. From what I have seen, Alpha Zero is definitely the strongest chess player on planet Earth.
Stockfish 8 is very outdated -- at least 100 elo weaker than the latest sf10 versions. Until the DeepMind team publishes the results of a fair match between the latest a0 and the latest sf10 (i.e., with comparable hardware, a sufficient sample size of games, proper time controls, proper opening books and endgame tables, etc.), it will be impossible for anyone to really know what the "strongest chess player on planet Earth" really is.
@Cscuile I don't think it matters that is was Stockfish 8. Leela is inferior to Alpha Zero and it can win games against Stockfish 10. I think Alpha would beat any version of Stockfish.
@@vickmackey24 And yet a lesser AI from Deep Mind (Leela) can beat Stockfish 10. I don't think it matters what you put up against Alpha Zero.
@@Varney_of_London Cscuile made the point that based on the testing, we have every reason to believe Stockfish10 is the strongest engine right now. An engine winning a game against another engine doesn't mean it's better. Stockfish 8 won 6 games against AlphaZero that were just published. It seems like you're a 1200 chess player or less. You don't seem to know much about the game.
@@Varney_of_London Leela is still around 120 elo weaker than Stockfish10. I always find it weird when low-rated, low information folks like yourself speak in such a strange, arrogant way like you know everything
Thank you for this excellent commentary. Please do more!
THX for todays games, hard lesson :)
That was simply amazing.
I think that too many people are focusing on the game, which I also follow, as if this were an ordinary player. Since I have significant knowledge, and since I believe that Hawking and Musk were right, I am really anxious by the self-taught nature of this AI.
This particular AI (and its more generalized, even more recent variant MuZero) is not the worrisome thing, albeit it has obvious, potential applications in military logistics, military strategy, etc. The really scary part is how fast these were developed after AlphaGO debuted.
We are not creeping up on the goal of human-level intelligence. We are likely to shoot past that goal amazingly soon without even realizing it, if things continue progressing as they have.
The early, true AIs will also be narrow and not very competent or threatening, even if they become "superhuman" in intelligence. They will also be harmless, idiot savants at first.
Upcoming Threat to Humanity.
The scary thing is the fact that computer speed (and thereby, probably eventually AI intelligence) doubles about every year, and will likely double faster when super-intelligent AIs start designing chips, working with quantum computers as co-processors, etc. How fast will our AIs progress to such levels that they become indispensable -- while their utility makes hopeless any attempts to regulate them or retroactively impose restrictions on beings that are smarter than their designers?
At first, they may have only base functions, like the reptilian portion of our brain. However, when will they act like Nile crocodiles and react to any threat with aggression? Ever gone skinny dipping with Nile crocodiles?
I fear that very soon, before we realize it, we will all be doing the equivalent of skinny dipping with Nile crocodiles, because of how fast AIs will develop by the time that the children born today reach their teens or middle age. Like crocodiles that are raised by humans, AIs may like us for a while. I sure hope that lasts. As the announcer in Jeopardy said about a program that was probably not really an advanced AI long ago, I, for one, welcome our future, AI overlords.
"The biggest threat we have to fear from AI is that some people will conclude too early that we have fully understood them."
It's a wise man who said this. The problem is not the AI, its the humans. And humans make errors - some are not even recognizable and some are devastating.
@@Pintkonan Blather. Anyone who concludes that he understands an AI is confused or moronic. The ambiguous comment about humans making errors is only relevant if you are referring its applications, e.g., in war.
10:40 "and black won a few moves later.." I'm kinda pissed u did not show that...
I wish I had the enthusiasm of this guy for chess !
Would be interesting to analyze a game that Alphazero loses, just to get the "contours" of its behavior.
it never lost one... as far as I know
@@TheKunec It lost many games, only the winning games were published though. And the whole A0 vs SF is a hoax, since SF had bad parameters. SF would not lose to A0 under fair rules.
In the pdf in page 2, in the table C with TCEC openings and AlphaZero with black it loses to SF by a lot. I want to know what is the speed of the cores of the computer where Stockfish ran and how many cores, also the specs of the computer AlphaZero, how many cores and how many neural units. That will show AlphaZero is just a whole room versus a tiny computer.... TCEC is using a 128 core computer but the most important is not the number of cores but the speed. For example a 128 core with a 2 Ghz speed is inferior than just a 4 core computer with speed of 4 Ghz. Also speed of RAM is crucial too.
@@mogyorospusztai they played 3 hour games with 15 second increments. Opening tables included. How is that bad parameters?
@@mogyorospusztai Sorry to burst your bubble but In the new matches Deep Mind did just that they gave SF what people were demanding and AZ still won the matches convincingly. AZ didn't just beat SF 8 in the new matches it also beat SF 9. Also iirc AZ also beat the Dev version. This was done at a normal classic time control. I saw somewhere online that in one experiment even with a big time odds disadvantage AZ still won.
Sir your explaining is profound can you please continue making these kind of videos.
I am impressed by Alpha Zero's play.
On the other hand, regarding all the amazement and mystery that is being evoked by commentators after a win, I just wonder about one thing: I think it is a fact many of us are not really used to taking a closer look at comp vs comp chess, so we might be more impressionable to its "general peculiarities" compared to human chess.
What I am asking myself in that context is: Aren't all these wondrous moves (like Ng5 here) merely standard for high-level computer chess?
I mean, does not even an arbitrary game between Stockfish and Houdini, due the engines's tactical force, contain many moves and concepts that are too deep for the human eye, at least at first glance, and leave it stupefied?
What I am trying to say is: Assuming it were rather the norm than the exception that any comp vs comp game contains such wow moves, what exactly - apart from the nature of the A0 engine itself (which would be a bit arguing ad machinam) - does differentiate the shown A0 vs Stockfish (8.0) games from any other game between two high-level engines where one is about 50-60 rating points stronger?
In other words: What makes these games special in the end?
I would say that the fact that alphazero is not scared to sacrifice for positional value, and zugzwag while stockfish plays by mostly piece value makes these games especially special.
AlphaZero made Dutch great again ;)
AZ has shown time and again that many openings which are sneered at by most humans are not that bad.
I'm curious how AZ would play many opening lines I'm drooling about being about to run experiments with AZ playing openings like the Orangutan/Sokolsky vs latest SF which is SF 10 I think?
A good game generally has one "you'd have to be off your rocker to even consider that" moves. An exceptional game, two. I lost count in this one!
He can hear what moves opponent makes in his head.
At 3:30 White was supposed to attack the weakside with a2a3. Black cant just blow through the strong side. While black is wasting moves headed for halt, black's weakside would be obliterated.
Is that current theory?
@@dannygjk No. It's what I see.
It seems a parameter for the value of a piece is not a good thing against AlphaZero. If its wins, it really shows how Stockfishs pieces can be worth much less than their standard values. But Stockfish will never find out, since it needs additional parameters for all ways in which pieces can lose value. I like to think that talented players judge positions in a similar way, feeling their way through variations and the overall dynamics in the position, before judging the importance of each piece in that "puzzle".
Exactly top players use their superior intuition and talent to navigate efficiently thru the chess tree.
Many chess players think that the top players are looking at variations at super speed and these "normal" players think that's why the top players are so good. They are so wrong.
@@dannygjk Well top players are calculating more moves in the same time than amateurs after all.
Well put
Yaaaa! A Leningrad Dutch and Black wins! This is one of my favorite openings.
Would interesting if Mr. Sadler would look some Leela-Chess-Zero games and see if they play is similar. On few games of Leela I can see that at least this Harry boy to pin down king stratagem seems to be in Leela's toolbox.
ans do recommend the book "Game Changer" lot read there. Probably wont make u stronger but highlight nicely what sort thing a0 goes for
Anand plays such type of move ....watch anand vs aronian tata steel tournament...you'll amazed...
Wait what? Published in december 2018? But that's right now, can anyone link the publication?
@@bluuuemoooon Yeah that's a year old right? A pre-print isn't exactly the same as a publication.
@@yoloswaggins2161 It's the new one. deepmind.com/research/alphago/alphazero-resources/
@@bluuuemoooon aha this one science.sciencemag.org/content/362/6419/1140.full.pdf
@@yoloswaggins2161 I don't see how many cores Stockfish had and what is the speed of each core and how many cores AlphaZero had and how many neural units it had. Also how many time per game in this "chosen games". Is perhaps AlphaZero using 745 neural units and Stockfish just 20 cores or so? Even MVL computer has more cores. TCEC is using 128 cores for example.
@@jmbbao wtf is a neural unit bro. It says in the paper sf had 44 cores and a0 had 4 tpus (as far as i can tell).
Very entertaining and thoughtful commentary! Best of all the Alphazero vids. And I agree with him that the Dutch is dodgy but it is a lot of fun, especially for sub-grandmaster players and for chess engines that don't really understand chess, like SF! (The move c5 check by SF, terrible! Few humans would play it, yet I noticed that chess engines seem to gravitate to it, "like the moth to the flame" lol.)
Truly incredible!
The dutch was something I used to play as black regularly. Lost a lot of games. Glad to see it wasnt my decision that was lacking. Just my gameplay lol
Yes I played the Dutch now and then for a change just for fun but I never played it in a tourney.
AlphaZero is fantastic but I'm more amazed by this man's ears.
U sir, made my day.
what a poor comment from someone hiding behind an F
Matthew Sadler is all class. You'd do yourself no harm by looking a bit closer at that.
Irrelevant.
I am most amazed by this man's amazing and very enthousiastic commentary.
Matthew's grin @4:43 LOL
More please 👍👍
Surprised at SF's naiveté on several crucial junctures, one being the really bad looking 22,,,Be2 -- where 22...Bg2 would have made much much more sense, and that whole "outerworldly" sac feast with threats on h1 would not be possible, or even a distant option.
Strong engine to probe this, anyone?
Good game, good video. Thumbs up!
I would happily play ten games in a row against Matthew Sadler, just to boost his rating to 2701. He might well be the strongest active player who has never broken the 2700 barrier.
Sir, is it possible to publish the analysis in writing for the upcoming videos as a reference for us, because we may find it difficult to understand the dialect because we are not English
Will alpha zero still wins? Stockfish 12 nnue is released
Where can I purchase alpha zero?
every_tribe I watch this 10 times. Matthew 's comments is very clear.
Absolutely beautiful
Did you see alphazero talking?
Excellent game, excellent explanation! Are these games published anywhere? In a replayable format, to be studied?
He published a book with Natasha Regan with a long title. It begins with “Game Changer...”
Can alphazero mate stockfish without queen in the beginning of their game?
Nah, alpha isn't that good
Cant help but be bit worried about usage of AZ kind of Ai in areas where arent exact rules. It will do something suprising what we will just dont get... itll think too ahead for our slow brains .-(
Alpha zero should play stockfish at full strength
...or maybe you could look at TCEC 15 when Leela chess zero beat SF 10 (yes ten).
Carlson vs alpha. Z.
Please.
And it's not even alpha one. Only alpha zero. Just the beggining, the pioneer.
Alpha Zero is the Bobby Fisher of computer chess. Now let's see Kasparov and Carlsen :)
Can i learn to play like alphazero by watching these videos? Stockfish will for me too
I love it: "absolute disaster"
So... if AZ plays a match vs SF 10 and wins what then? SF 11 will be released and I predict that people will then say, "Try SF 11 against AZ we will see who is best." You see where this is going I think.
The solution is very easy: let A0 participate in a CCCC or TCEC tournament in real time, publicly streamed as usual.
"Sf8 made a mistake, but sf10 acknowledge it and would not make it, therefore it will win", while completely disregarding the fact that sf10 could very well do other mistakes. And all of you people don't know a thing about neural network and their learning process, which is pretty much the reason behind the enormous amount of computational power A0 requires. Why can't you simply accept the potential of neural networks like all the goddamn world is doing? If there are individuals who stop the progress are people who think like you, and the sad part is that you don't even realise it
@@Coskunn Lc0 is doing that.
@@dannygjk I am a Leela fan myself but Leela hasn't demonstrated any clear superiority over SF in their battles overall, certainly not to the extent the AlphaZero papers would indicate (if anything SF has proved to be the better engine).
It has become a theme that Deepmind achieves something very significant yet it gets immediately overrated by reactionaries. AlphaZero chess: incredible engine with a positional style but not something that would render the "classical" chess engines irrelevant. AlphaStar: The first starcraft AI that could stand its own against pro players, but not "without mechanical advantages through strategy". AlphaFold: Light years ahead of what other programmers were trying with protein folding, but it does not "solve" protein structures.
@@Coskunn To our knowledge AZ hasn't played all the SF versions that Leela has played.
What about the hardwares? People say that the stockfish hardware is too poor even not able to cope with alpha zero. Can you clarify? appreciated
Can't make a meaningful comparison of the hardware because SF is specialized for one type of hardware while AZ is specialized for another type of hardware. That said SF does about 1,000 times as many nps as AZ does.
Who is better Magnus Carlsen or AlphaZero ?
clearly Magnus is better. AlphaZero is zero when power is gone. :)
@@roydondsa I don't know if MidnightFox was serious with that question:)
lmao AZ is roughly 800 Elo above Carlsen (ballpark figure).
Analysis was too fast to enjoy liitle details
The pace is totally under your control.
Oh my god
Is still chess champion AI??
All GrandMasters will never beat these AI CHESS Engines. Magnus must have Nothing but Fear up his sleeve to even think of Playing these Engines. No doubt! Humans should stay on Earth Chess, and not Venture into Outer Space Chess.
The way I see it, AZ way of playing will always win against a rules based chess engine. It breaks rules and only cares about long term and final win. It is the AI I fear humanity has no chance of winning in the coming war. Sigh.
And here I present to you, two persons who have absolutely no idea what they're talking about!
I've seen more incredible games of SF 10 demolishing SF 8 than this one. Because it's not put on youtube people don't know it. Until A0 wins a computer chess tournament like TCEC there is not enough evidence that it is stronger than Stockfish.
Well AZ destroyed SF 9 in at least one match so I wouldn't be so confident. SF 10 would do better than SF 9 of course but if I was betting on a AZ vs SF 10 match I would put put my money on AZ.
Was Stockfish running on a potatoe again?
"Each program was run on
the hardware for which it was designed (23): Stockfish and Elmo used 44 central processing
unit (CPU) cores (as in the TCEC world championship), whereas AlphaZero and AlphaGo Zero
used a single machine with four first-generation TPUs and 44 CPU cores (24)."
Cscuile well when the games were played, sf10 was bot available
@Cscuile And Stockfish 9.
@@CP-jp8hh AZ did play SF 9, (it may have been the dev version before it was released under the name SF 9), and won convincingly by a huge margin. I am not confident that SF 10 would stand up well against AZ. SF is facing a machine that apparently has such an excellent "grasp" of chess that in spite of SF having a huge nps advantage AZ "sees" more clearly. The size of the chess tree is so huge that even at SF's speed what it sees is similar to comparing a helium atom to our galaxy. When you look at chess that way it really shouldn't be a surprise that a smarter approach cutting out almost all the inferior variations is more effective.
Dan Quayle: Noe
Do not play a game against a machine..they always win...machine vs machine? ..so this is another story which humanity is no more!...at least i'm glad that i will not be around when it happens.
lol why stockfish 8 and not the 10? Unfair