Stockfish vs. Leela: 3600 ELO BATTLE

[@Alkis05]

Stockfish: That is a brilliant move!! Stockfish: Thank you!

[@Θανάτος-ο4σ]

Stockfish technically has the biggest ego and is the most confident chess player. It literally compliments its self.. 💀 man I'm dead

[@LostPageWasTaken]

AlphaZero : Hold my knight@@Θανάτος-ο4σ

[@luminous3174]

These matches are the equivalent of a 3 year old and a 15 year old playing soccer, and another 3 year old is refereeing

[@stevenshadaia5198]

It’s very true when Levy says you’re not too different from 3600 engines. The only difference in our rating is that they have a 3 in front.

[@CommanderNissan]

So 3 of me can beat a computer? It all makes sense now.

[@killerbug05]

Well for me there's a 36 in front but yeah you're right 👍

[@purplewine7362]

@@CommanderNissan 6

[@ladywaffle2210]

They just have an extra zero, no big deal

[@bm-ub6zc]

🤣🤣🤣🤣🤣👍👍👍

[@razNEAT]

Can’t wait for the Engines to get a chat feature so they can talk shit to each other during the games

[@user-ul6bm8pt2y]

GPT 3 Chess trash talk generator wouldn't be too hard to do I think. We need this.

[@shwoompl5782]

"I had intercourse with your creator."

[@Jk-ki7tr]

@@shwoompl5782 I read this as a robot

[@cinegraphics]

I entered the sentence into a robotic voice generator. Sounds funny :)

[@RiskyBisky11]

@@shwoompl5782 this escalated rapidly 😂

[@olpporsetty]

Stockfish evaluating itself as totally winning: oh yes of course 😎

[@davitharutyunyan875]

It's like the Obama giving a medal to himself meme lol

[@siddhantjhaveri]

If Stockfish says it is winning then it is. It's like the ultimate decision.

[@sachithvp]

@@siddhantjhaveri alphazero : hold ma beer

[@kurtzipagan6704]

@@sachithvp the new stockfish 14 is stronger than alpha zero,the type of stockfish that alphazero beat is the stockfish 8,but still alphazero is out of this world AI. It's just brilliant.

[@ltsgobrando]

You know when Stockdish evaluates its position as completely winning the game is about to end... only issue is sometimes its playing Leela 🤣

[@jasonkoch3182]

Computer: Sees 25 moves into the future. Magnus: Sees 12 moves into the future. Normal person: Horsey move funny.

[@kirbyknifeman2877]

Actually engines can look at 100+ depth sometimes. And nowadays most of superGM games are 20+ moves ahead For eg in in the immortal game of vishy he made moves with depth 30 at max Garry Kasparov IN 19S PLAYED A 40 DEPTH MOVE

[@MicroClases_Ciencia]

It is called "HorseFace", because it doesn´t have the body

[@onionvlogs5222]

@@MicroClases_Ciencia Horsey

[@stevensilva3872]

Jose Casablanca said he only saw one move into the future. But it was always the right move.

[@obscurity3027]

@@stevensilva3872 Capablanca ;)

[@PJ_enjoyer]

"They play until checkmate so you are not too different than 3600's" -Levy rozman 2021

[@ezqwenthur6370]

best motivation quote of all time for the 600s ngl

[@NathanielDickison]

Right in the feels

[@PJ_enjoyer]

@@ezqwenthur6370 I agree

[@jonathanbernhard5418]

Can't tell if this is an insult, or a compliment...

[@KieraQ0323]

actually I just play until stalemate when i am up a queen so actually I am different

[@vinnymurphy1299]

Gotham when the engine says he made a wrong move: “computers are stupid, they know nothing about chess” Gotham when engine sacrifices a pawn: “absolutely genius, a human could never make this move”

[@adityajha7841]

Today I hung my rook but stockfish said it was the best love... Also I did end up winning 😂

[@jonny2900]

@@adityajha7841 oh yeah I hang pieces all the time and then later stockfish often says it's a good move

[@ГеннадийОсетров]

First thesis just means: humans don't play like computers. You can't calculate the whole tree of variations and computer can. A human plays chess by using conceptions, ideas, strategies when a computer can just calculate deep enough(I'm talking about classical engines). I think Levy refers to this by saying it. Because certain computer moves is hard to explain in terms of ideas and it'd be really stupid to show a 30 moves computer line which justifies that move. Also a classical engine "knows" smth about chess - it needs it to correctly evaluate position, because in reality the whole tree is simply too big and you still need to cut off branches(almost as a human being). And it's just a joke. Do you really believe Levy thinks that computers are stupid?

[@5poolcatrush]

Not letting them to play from starting position is severe hadicapping of their capabilities, therefore making their competition biased and kinda pointless

[@greengoblin9567]

@@5poolcatrush It is always a Berlin or Kan Sicilian leading to a draw

[@lelouchvibritannia1919]

Meanwhile that one guy who thought he could memorize all positions in 30 days and he wanted to beat Carlsen in that time

[@edoardopetrovic5150]

He didn't want to memorize all positions, he wanted to memorize an algorithm that could help him find the best moves like a computer. But yeah, it's equally dumb considering that people have tried to do this kind of stuff for decades now, and he wanted to do it in 30 days.

[@fightingforcatalonia]

I think he shoulf upgrade gf. Look at Agadmator's much nicer & cooler

[@priyanshchess4836]

@@fightingforcatalonia get a life bro (2)

[@ggeorge02]

I bet that would be a simple task for Lelouch Vi Britannia

[@rakshitpandey1413]

Meanwhile lelouch vi Britannia tried to give and escape from a king giving check to a king

[@TheMarakusan]

"You leave this position with two intermediates, they'll both find a way to lose it somehow"

[@krishnaramesh1851]

I should know, I am an intermediate 1300

[@ezqwenthur6370]

i have such power that even up with a queen and a rook can lead to stalemate.600s

[@stomponpie]

@@ezqwenthur6370 up a queen and a rook is an easy stalemate! Losing with a queen vs knight and pawn shows talent.

[@pschneider1968]

For either side 😃

[@necelynx]

it hurts my brain to think about how chess is such a complicated game, that two computers playing as close to perfection as is possible, can outplay each other.

[@JustSomeoneRandom1324]

Humanity will have peaked when we manage to get computers good enough that white or black will always win, the game is solved and Chess has become a computer's idea of tic tac toe

[@alon3304]

@@JustSomeoneRandom1324 how can black always win? He is down a move. Also chess will not be solved in our lifetime or maybe ever

[@bmartin4320]

I feel like if they allow draws they would start to play super cheeky though lol and force draws 😂

[@JustSomeoneRandom1324]

@@alon3304 How can you be sure that 2nd to move isn't an advantage if the game was ever solved? The higher the complexity, the less sure you can be of that. I would be kind of sad if humanity peaked in my lifetime, not gonna lie

[@Krish_krish]

@@alon3304 It could be that whatever white plays is losing. Very very unlikely but it is possible that black always wins.

[@DavidEmerling79]

I mostly use Stockfish to analyze positions. Recently, I added Leela as an additional engine for analysis. I figured that both of these very strong chess engines would mostly agree when calculating the best move and the position evaluation. I was surprised! They frequently disagree on both fronts. In fact, the best move of one engine often doesn't even rank in the top three of the other engine's calculations. I also noticed that Stockfish evaluations are usually more extreme than Leela's. For instance, even when they agree on the best move, Stockfish might say that white has a +1.50 advantage whereas Leela might evaluate white's advantage at +0.40. Also, Stockfish seems to evaluate much deeper and faster than Leela. Stockfish gets to a depth of 25 within seconds where it might take Leela 10 minutes to get to that depth - at least, on my computer. Finally, I noticed that in complex positions, Stockfish will often find some esoteric (non-human) move that only works in a "computer line" whereas Leela usually finds a strong, human-like move that makes the position easier to play.

[@zeronothinghere9334]

Depths on the two engines don't mean the same. It's only really useful when comparing it to the engine itself.

[@angelmendez-rivera351]

Stockfish 14 is a much stronger engine than Leela. So, there is nothing surprising about the fact that they disagree very frequently on the best move and position evaluation. The gap between Leela and Stockfish is actually bigger than the gap between the evaluation of a GM and a Super GM of a position and the best move.

[@zeronothinghere9334]

@@angelmendez-rivera351 Stockfish 14 was built with the help of Leela's engineers. They frequently trade methods and expertise, esp back when Leela was stronger. Leela's strength will likely soon change again.

[@RuneKillerz109]

@@zeronothinghere9334 How tho? Depth means how many moves deep you calculate into the future, so wouldn't that mean that, as stated above, (concurrent) move evaluation is just different between engines rather than depth being different?

[@zeronothinghere9334]

@@RuneKillerz109 No, depth should refer to what layer in the neural network you are on. You used to be able to see at low depth (8) that they make silly moves that don't even fit the next 8 moves. Because every neural network is different, different depths can mean different complexicity. Trying to explain what one layer means: One layer has basically a bunch of numbers, that it multiplies the previous bunch of numbers by. The result can then be added together to compute an output, or, apparently in this very specific case, you can multiply it with another layer. I thought that Stockfish was not using a Neural Network though, so I'm slightly confused at the current status.

[@luis.aacjara]

This whole match was literally Stockfish and Leela talking to each other: “You can’t do that” “OH YEAH??” *does the exact thing*

[@GreenscreenKid-y1d]

Fr thooooo

[@BruhDidYouKnowThat]

Like siblings

[@Not3xactly]

Levy: "YOU can win this against Stock fish" Me: "Challenge accepted"

[@lukekul297]

I setted this up in chess and I never won

[@ventu7907]

I heard the hippo opening breaks engines, but you still need to be really good to win/draw

[@krislocke1639]

I got mated in 10 moves lmao

[@wellshit9489]

@@ventu7907 the hippo is good against players that are cheating because it makes the game slow so they lose on time. It's not good against engines if you don't have the possibility of winning on time.

[@pyon6039]

@@ventu7907 Hippo dont work against Stockfish

[@0xf7c8]

I haven't seen commentaries on how both LC0 and Stockfish teams help each other out. The most notable example is that when stockfish was on version 12, LC0 had an edge on strength because of the revolutionary neural network it was using. For Stockfish 13, the LC0 helped the Stockfish team to port/create a neural network position evaluator for the core of Stockfish (the famous NNUE). That's when Stockfish regained the lead on strength. This only shows the value of these developers and teams. They really want to create the most powerful chess engine, even if that includes porting some of the newer technology developed for one engine into the other.

[@chopin4525]

NNUE comes from shogi engines not from Leela's team. Different concept.

[@dannygjk]

@@chopin4525 It's still neural net tech.

[@tomkelley4119]

I kind of envy the fact that you're good enough at this game to understand how great these engines play and how far beyond human play these moves are. I wish I had the ability to study and understand lines like these deeply enough to be as surprised as you are. This is art that only other artists can understand.

[@comp.lex4]

Jazz chess

[@josephbishara4791]

You're basically saying that you wish you were a chess grand master. I have the same wish too. I also wish I was a great song writer and producer. And I wish I was a little taller. I'm 5'7...I wish I was 5'11

[@tippy1951]

@@comp.lex4 snake jazz chessss

[@mynames7664]

@@josephbishara4791 I wish I was a little bit taller I wish a was a baller

[@NothingYouHaventReadBefore]

@@josephbishara4791 As long as you know you're good enough.

[@sadas3190]

"Leave it to two intermediates and they'll find ways to lose it somehow." I've never felt more attacked in my life

[@d.jensen5153]

I thought it was hilarious. But I see your point.

[@hydranmenace]

I like these engine games. Human games at high level often just seem like seeing who memorized a line better until one person makes a mistake. There isn't much creativity. With the engines things get pretty crazy. Makes it fun.

[@losyart]

But human moves u sometimes kinda understand engines not so much

[@jethalalnhk2409]

@@losyart you can't understand many computer moves unless you are "lavy"

[@GeoFitz4]

In Game 2, went through it in the Lichess Openings, and through Move 12, there's a Lichess game that was played in 2020 between a couple of 2380s. They diverge at move 13 with 13. Re1 instead of 13. b4. The game was a win for Black.

[@gaopinghu7332]

"Leela saw Ra4 and Leela liked it" I see what you did here

[@sarmisthadas3892]

Levys gm training partner is now very angry

[@kermitthefrog9322]

Don't make fun of 3600 not everyone is a 161660

[@basesixty6739]

It’s a robot not a physic

[@veratycheberg4975]

I dont get it 🥲

[@milannesic5718]

That is because you don't read comments. People were making this joke about Rook A4 whenever was appropriate, and yelled at Gotham when he missed opportunity to make that joke. This is why he said it

[@roflenkosz]

Average Gothamchess viewer: damn these Magnus endgames are really tough to understand i hope next video is easier for me to comprehend Levy : 3600 ELO robots battling it out

[@oliverkewell4492]

Handsome chess streamer breaks down why robots are superior to humans at everything

[@fightingforcatalonia]

I think he shoulf upgrade gf. Look at Agadmator's much nicer & cooler

[@djpd8333]

wow trolls got no life

[@priyanshchess4836]

@@fightingforcatalonia get a life bro

[@mokshi2684]

@@fightingforcatalonia what??

[@tellurius4951]

@@fightingforcatalonia no, he's not

[@XIIchiron78]

The point you make about engines being unable to ruin the game of Chess regardless of how good they are is pretty apt. If anything it makes it more interesting to see how high the skill ceiling goes, and learn about positions humans would just never consider. After all, it's not like you can memorize these lines - you can't even write them down, because there is _literally not enough space in the Universe to store the solution to Chess._ Truly, a beautiful game.

[@s-x5373]

the keyword trick is "store" there

[@ggaming8999]

Actually, the universe is pretty big so on a literal note yes you could store all possible variations for every move ever on every turn but it would probably be so big that'd it would create a super massive black hole based on the mass of the ink used to write it alone

[@isavenewspapers8890]

@@ggaming8999 You know you can use more than one sheet of paper to write, right?

[@zeronothinghere9334]

@@ggaming8999 The argument is more that there are more possible states in chess than atoms in the universe. However, this argument breaks apart for this case, since we don't need to let one atom correspond to one chess position. We can arrange the atoms in a specific way, and have one atom existing stand for 1, and an atom not existing for 0. Suddenly, we can store a lot of information. That way we could probably store all moves (I didn't math it out, but current compression algorithms are pretty good).

[@guardianangel1468]

a 1 terabyte microSD card stores 8 times 10 to the 12th power bits of information. You would need roughly 10 to the power of 108 micro SD cards to store all possible chess games, even when assuming each game takes 1 bit of information to store.@@zeronothinghere9334

[@CanadianPianoMan]

Levy when the computer finds insane moves in human games: the computer doesn't know anything about chess it should shut up and go outside Levy when the computer finds insane moves in its own games: the computer is 3600 so we should probably listen

[@tsawy6]

In some cases it's true tho. Sometimes playing one move from the computer will put you in a worse position down the line because you don't know what your plan is

[@markrobertson3054]

@@tsawy6 no. Its because chesscom’s engine sucks

[@raviolifrito5562]

@@markrobertson3054 XD

[@dalakhsarothal9624]

Emotional response vs rational response

[@greengoblin9567]

What happened with the sweaty never resign policy?

[@Not3xactly]

Hey Levy, I think you've made a video like this before, but it would be interesting to see how the top players are learning from the engines and incorporating some of these concepts into their own games. Would be cool to see an engine game followed by a real game that illustrates an obscure winning idea.

[@Crisis_Jones]

h4 and h5!

[@antonsackley5514]

He already made that

[@elgreco3015]

It is already illustrated from the magnus endgames video which he won most of the with a flank advanced pawn. h4 and h5!!!!!

[@sjegannath6295]

Alpha Zero video and how magnus carlsen learned from alpha zero video. Those 2 videos just do that.

[@irontide]

If you're interested in this kind of thing, there's a good book about the influence on AlphaZero on chess called 'Game Changer' by GM Matthew Saddler and WGM Natasha Regan. Saddler frequently analyses TCEC matches as well, and has a book coming out soon on using engines to improve your play which will be called 'The Silicon Road'.

[@xanschneider]

This are my favorite games. The Alpha Zero revelation back in 2018 was what brought me back to chess after decades. This games are mind blowing.

[@talphazero1036]

Almost all modern engine games are mind-blowing. Leela Chess Zero and Stockfish are leading the pack, but other engines are keeping up enough to create their own masterpieces or last long enough to bring the games to interesting results. Check out Jozarov's Chess Channel if you want to see some more engine games uploaded frequently.

[@talphazero1036]

27:50 Everything going forward is some of the best entertainment I've gotten from watching a youtuber review a chess game. Levi just adds so much spice to almost every move that it's hilarious yet very engaging. There are tidbits here and there throughout the video that bring me to the edge of my seat and laughing out loud, but this just takes the cake.

[@A_Swarm_of_Waspcrabs]

I love that you did this- it's rare to see a chess KZbinr this charismatic or a chess player this skilled cover the TCEC. Having both is a bonus.

[@topperharley2323]

It's cool, that they are playing till mate. Often in computer tournaments they resign at +-5 or so. Engines play such nice endgames, so it's nice to see everything.

[@naman.jain.01]

They had positional long term advantage

[@christiandelvalle282]

6:22 - “Leela saw Ra4, and Leela liked it.” Kasparov is *shaking*

[@unofficialskins624]

Huh

[@christiandelvalle282]

@@unofficialskins624 You they type of guy to say “huh” when you don’t capture en passant and then proceed to get the brick

[@Krish_krish]

Hey Levy, I really like your recent content but I was wondering if you could continue your 'Easy Chess Endgames' series you started 8 months ago? I think you've done two videos in that series but they were super helpful and people were hyped for more videos

[@kevinkozak6410]

Ikr

[@AndroidPoetry]

TCEC is awesome! I'm very glad you are covering them, please hype up the next SuFi - Superfinal, which will be played between the top 2 engines in the premier league, which is usually SF and Leela.

[@ariavachier-lagravech.6910]

"I hope you are enjoying yourself" I did Levy, this video has been educational for me. I might not be able to play like engine but some of the concept you mentioned would probably help me improve.

[@priyanshchess4836]

wtf is your name

[@manawearblack]

Why are you a dude who's channel name is a girl's name and a super GMs surname with a anime profile picture? How did this happen, I need answers

[@thePersonoyt]

@@manawearblack 🤣

[@ariavachier-lagravech.6910]

@@manawearblack bro, it's just YT name and profile picture, anyone could use anything they see fit lol.

[@why7387]

@@ariavachier-lagravech.6910 don’t worry about it they’re reading too deep into it

[@IMortage]

With Stockfishs positional evaluation now also being NN based, it's been clawing back against Leela. It's search algorithm is still much better tactically than Leela's Monte Carlo approach. I suspect Leela still is positionally superior. But in an unbalanced position Leela can be let down by its incomplete search tree.

[@MrCheeze]

It's interesting that a hybrid of NN engines and traditional engines turns out to be far better than either approach on its own. Reminds me of how "centaur" players (humans using engines) are still significantly better than engines alone.

[@Maxawa0851]

@@MrCheeze provided the human knows what they're doing

[@jolez_4869]

@Luke Brooks Boi you are watching a chess video played by two computers

[@RhodokTribesman]

@@MrCheeze Source? I didn't know this

[@fredgandolfi2356]

The marvel of how engines use pieces is so far removed from how humans think to move them, that chess will be good enough to entertain mere mortals for a long while yet. Super GMs don't play the King's Gambit at the World Championship Match, but to us, that opening can ruin you both as white and as black. Computers toss out the French (I knew it!) but given enough skill at the non professional level, the French is still playable. AlphaZero destroyed Lee Sedol at Go that is a much harder AI problem than chess... still Go is unfathomable to the majority of the population. Let's not be sad that computers are "solving" our favourite games - let's celebrate that they are showing us wonderful new possibilities in new ways to play and enjoy them. Chess can't be boring if played like this. Love the video.

[@shucklesors]

the most exciting thing about the complexity of this game is that, computers constantly prove themselves wrong and we are watching them learn (although like you said we have no idea about what we even have no idea about). so what you said about computers tossing out the french: we have no idea if computers at 4000+ elo will later on PLAY EXCLUSIVELY THE FRENCH because of how good it actually is. and then we can see a reversing trend at 4500+ elo. we literally have no frickin clue. it is so incredibly mind boggling and exciting.

[@RyanW1019]

The whole video I was thinking “wait, was every game just 1 - 0 or 1/2 - 1/2?” Glad you called out that yes, there was a game that black won, but it was only one game.

[@alienrenders]

The TCEC openings are heavily biased toward white. They do this intentionally. They try to find openings where white can win but black can also defend. There's no point doing this for both colors. For Leela to lose with white at TCEC, it has to be a whacked out opening that her training wasn't suited for.

[@DanielGomes-sw2fd]

There were some other openings good for Black, including some strange KGA (as it always happens in TCEC), it seemed maybe one engine could do it but they couldn't win.

[@irontide]

@@alienrenders There are also a couple biased towards black, but given the inherent advantage for white it is very difficult for black to convert. It took Stockfish 14 to convert even one of them, and that was really remarkable.

[@morgancolin-mourre4760]

This kind of video is one of my favorites, it's just incredible to see how these engines play such crazy chess! If you could do more of them it would be great!!

[@Imfromjamaicaman]

This is Levy’s fluff piece so when the robets took over the world, he wouldn’t be used as a battery like most of us...

[@brandonovich6063]

It's already happened bruh we are in a simulation

[@AISlopForHumans]

Robets

[@maitrayraut7990]

Lol I read it as luffy's one piece

[@kenoliver8913]

Yep. Shorter Levy: "I for one welcome our new chess overlords".

[@brokemage1395]

I loved this commentary, dude. I'm only 500 rated so I'm not skilled, but the way you were explaining the moves and had the AI basically threat eachother made me laugh a few times. I appreciate it.

[@chosenmc]

levy: they drew their 1st game, this is their second levy on the next game: for their second game-

[@schwebor]

Leila reminds me of a famous goblin tactician: "everybody but me, CHARGE!!"

[@omezzyallen7742]

That first game taught me so much, a lesson in piece play. Pieces that look dumb are usually dumb.

[@noahb0216]

"The king is on his way to pick up the rent money!" Underrated line

[@idontwantmynameinhere]

"computers have a sense of humour" ~Levy 2021

[@acommenter4252]

7:54 "Oh engines love this stuff" Really? I love that stuff too. As a 800 rated noob, as soon as I realized that the bishop and pawn could protect each other practically forever in a heavy piece endgame, it was like a glorious moment for me and now I play it in every game

[@chuike5222]

“If you’re in the triple digits or RARE double digits” rating, just look at this stuff and say “oh wow” !!!had me cracking up for a while! Awesome stuff Levy!!!

[@vaibhav7219]

I play just like stock fish. But I just end up with hanging pieces and blundered pawns without counter play.

[@jasonhe5578]

same lol

[@wardsmith2542]

ROFL

[@Reynfallen]

"That king is on the way to pick up the rent money." Crazy ass joke made me drop my damn phone! 🤣

[@hunterquintal5916]

This was one of the best KZbin videos I’ve ever seen. I was laughing so hard; you’re a great performer and analyst

[@jessejordache1869]

Nice to check in on the engine world. I used to donate computer time for Leela with the goal of understanding Neural Nets, and came away with the commandment "learn linear algebra, young man." Leela had two big weaknesses that cost it wins against Stockfish when Stockfish was hand tuned, and now make it lose more now that Stockfish is also an NN: Leela is self-trained, and the more recent games leave the biggest impression, and Leela's search is highly selective - most positions it doesn't bother exploring. So that makes it sort of a specialist instead of the general approach that Stockfish has. A few seasons back when the two were roughly equal, Joroem, who does the opening books for the SuFi, got tired of 90% draws, and was running low on "positions that are theoretically drawn but still have enough play in them for a decisive result", so he threw in a lot of dubious lines, because why not? Let's watch them defend lost positions, or see if they can bring home an attack just because best play says it's possible. The problem was the two engines did not react the same way. Stockfish: This is just another position. Let's do this. Leela: WTF? I never have, nor would I ever, EVER put myself in any position that even slightly resembled this. Is this Shogi? Result, huge win margin by Stockfish, which again, in other contexts was basically equal in strength to Leela. The second weakness is much easier to explain - Leela's net is HUGE, and yes, it usually does understand the position in front of it better than Stockfish does. BUT.... Stockfish' NN understands positions quite well enough, and since it's not self trained but trained from positions from any source you like, it retains its universal accuracy. And it's still blindingly fast. So it seems like an a/b minmax search with a NNUE net is the optimal setup at the moment. I don't see Leela catching Stockfish with its current architecture. But I like watching them face off. What I like about MCTS as opposed to Alpha-Beta minimax is that MCTS heads for positions with the highest number of descendant wins.. So it knows that there are draws and then there are draws - like Magnus Carlsen grinding away at a theoretically drawn endgame where he's the only one who can possibly win because his opponent needs to defend perfectly. MCTS can make that distinction. Otoh Minimax just asks "what is the ultimate evaluation here?" and so all draws are equal. Of course in an engine match the difference doesn't really matter, since you're going to get perfect defense virtually every time, but there have been swindles here and there that resulted from Leela searching out the most aggressive drawn position possible while Stockfish was ambivalent so long as it was 0.00. The downside of MCTS, even when Leela was at the top of the engine world, is that its search is based on moves that *it* would make. When instead it gets a bad move, it often doesn't see the refutation, because nothing like that really occurred in its training games. In tournaments when facing much weaker engines, Leela misses wins, resulting in the nickname "Drawla". Stockfish pounces on errors, so it doesn't have the same problem. btw, they're both equally NN -- there's nothing "hybrid" about Stockfish. Its evaluation used to be hand tuned, now it's a trained neural net. The searches they use don't enter into it. You could theoretically get an NN to train its own search algorithm - and its own time management for that matter, but nobody has done that yet: mostly because it would either take too long or because it's hard to give the net a target to train it towards. So... Stockfish: Neural Net evaluation, hand written search function, hand written time management. Leela: Neural Net evaluation, hand written search function, hand written time management. tl;dr: I used to help train Leela, here's some thoughts.

[@bilalkhalil1660]

“Leela saw Rook a4 and they liked it.” Levy will never forget…

[@milannesic5718]

That is because you don't read comments. People were making this joke about Rook A4 whenever was appropriate, and yelled at Gotham when he missed opportunity to make that joke. This is why he said it

[@ppmpyae1152]

@@milannesic5718 huh?

[@vulblhotdiessfi]

@@milannesic5718 wtf

[@milannesic5718]

​@@ppmpyae1152 What is hard to understand? In other Levy videos, Rook a4 gets mentioned, and then you read bunch of peoples comments bellow those videos how Levy missed an opportunity to say something about it or make a joke. In this video he remembered to comment that move. He did forget in previous videos, contrary to this comment that says he never forgets

[@milannesic5718]

@@vulblhotdiessfi What is hard to understand? In other Levy videos, Rook a4 gets mentioned, and then you read bunch of peoples comments bellow those videos how Levy missed an opportunity to say something about it or make a joke. In this video he remembered to comment that move. He did forget in previous videos, contrary to this comment that says he never forgets

[@killerbug05]

"these bots are 3600 so we should probably listen to them" confirmed that levy is going to be one of the first to bow to our soon to come machine overlords

[@MartyMeganium]

The number one thing I've learned from watching the highest rated chess engines face off against each other in these insane matches: BM actually stands for Best Move

[@markrobertson3054]

14:50 Stockfish also won with white which shows the oppening is trash

[@Martin-y8v4k]

every video featuring engines you should also add a disclaimer :”don’t do this at home !” as human we have a great talent to transform brillant moves into catastrophic blunders

[@tobyenglish6321]

love these type of videos!! you are the best chess content creator, love hearing your break down of complex games

[@clashplaya8638]

I watch these videos and then I think I’m capable of playing chess which is dangerous. I repeat Levy’s phrases in my head: “I’m gluing the position shut”, “I lose a pawn by white has no activity”, “Black is getting suffocated here”. And they’re usually followed by another phrase: “Wait. Fuck.”

[@thelightbringer8610]

To relatable Me: omg I just saw the next 7 moves in my head! I won the game, Enemy: does exactly what I did not think they would Well fuck

[@nwachukwuezeike522]

The thumbnail for this video was so good. Like it really hypes up their powers! Thanks, Levy

[@abcthecamper1799]

I love those kind of videos - its fucking mind blowing how strong LCZero is.. I would love more vids like this.

[@lordmatthewanunnahybrid1356]

Absolutely every bit of commentary was spot on and appropriate. That was a masterful presentation. It is rare to see someone doing exactly what they should be doing but you are one of those people. You were meant to do this. You have a perfect balance of confident banter mixed with timely self deprecating humor. The chess renaissance is just beginning and you’re going to be talked about as a facilitator. I mean it 100%. Great content Levy!

[@edoardoprevelato6577]

The Stockfish vs AlphaZero video was actually the first GothamChess video i've ever watched!

[@ladan7740]

Good choice

[@edoardoprevelato6577]

@@ladan7740 i know, been a subscriber ever since.

[@markrobertson3054]

@@ladan7740 nah it was old

[@-zelda-]

Not really a good video to watch. Those are 2017 matches.

[@markrobertson3054]

@@-zelda- even before 2017

[@haleysettembre]

19th day of translating Levy's titles into Neapolitan:"Stockfish contr a Leela: a battaglij re 3600 ELO"

[@Sionweit]

Why do people like this shit? What's the point? Mate no one cares please shut the fuck up

[@haleysettembre]

@@Sionweit famm nu bucchin

[@ItzEm666]

@@Sionweit no one cares bout u then

[@GeodesicBruh]

@@Sionweit fratm the pisc en bocc

[@GeodesicBruh]

Un nobile proposito

[@nathanaeldean6301]

You know, I actually want to see just how powerful these engines are in correspondence/daily chess, running on computers 24/7. If they are this ludicrously powerful in classical, imagine their skills in correspondence.

[@kylemacdowell4470]

These engines baffle me and I could never analyze like this. Really cool, and really opens up my understanding of structure, position and how that can beat playing for value.

[@kalpshah4624]

Levy : Shut up computer! You don't know anything about chess. Also Levy , when engine makes any move : Genius ! A human could never have thought of that move.

[@fstehle95]

Seeing this makes me want to submit an engine game to "Guess the Elo" :D Lewy would probably snap

[@Roof007]

Nah it's pretty boring most of the games are very long and often results in draw.

[@shambhav9534]

It would be okay. Guess the Engine! I want a new mini series now.

[@Al-hb6uj]

22:25 you can win this position against stock fish. Me: you’ve underestimated me for the last time

[@RedArremer]

24:10 That is FLAT-OUT Not true. I'm a 2500+ Senior International Master on ICCF. We use these engines in games that last months to YEARS. The King's Indian Defense is NOT refuted. Period. It may no longer have the sting against weaker engines, but it still holds the draw even today with NNs going at it for days to weeks on a single move. I should also point out that every major opening now effectively gets drawn on ICCF due to these NN engines getting as much time as they want on a single move. For TCEC play, they have to intentionally force the engines into dubious opening positions in order to get results other than a draw. That's something worth repeating: EVERY major opening is drawn with these NN engines, so TCEC has to FORCE dubious lines for them to play.

[@jf7199]

teach me o wise one

[@sayonmondal3454]

Can you tell me all the full forms?

[@DanielGomes-sw2fd]

Is Czech benoni holding for black?

[@crowbar_the_rogue]

Stockfish: Plays the French Defense Lc0: So you have chosen ... death.

[@wesmoorer3638]

I love how these engines do something that makes sense in every tactical thing. You sacrifice a pawn or something to gain a better position. Amazing 🤩. Next level stuff

[@aero7848]

The guy who makes these thumbnails needs a raise!

[@anonymous123456on]

These matches feel like two people with sharingans playing chess

[@erlindaalba1682]

Itachi vs Sasuke again lol?

[@AntoniWroblewski78]

Could you tie in a video like this with your GM quest videos, please? Would love to see how you use engines to analyze your past and future games.

[@seanx9]

Really like the breakdown and explanation of the engines games. If id be just watching them Id have no clue whats happening and be bored. Thanks man!

[@cinegraphics]

Leela is watching the video and taking notes

[@mcmudkipp]

"I'm not gonna checkmate you. I'm going to take your soul" -gotham '21

[@Lance_MadCat]

dear lord, game looks terrifying for a 300 ELO plebian like me :D

[@Xmask19]

Give whoever is making your thumbnails a raise.

[@pratyushsrijan1231]

Final Score : Stockfish 19 wins - Leela 7 wins

[@niranjanrajesh1058]

He literally says this in the video mate

[@codex4336]

@@human7491 If no book then all games will be drawn, especially with TCEC computers

[@codex4336]

@@human7491 It would be more repetitive with no opening books, especially with Leela, since she uses PUCT and it makes her moves more deterministic

[@МихаилЧалый-щ3ж]

@@human7491 you wouldn't see interesting draws. All you will see is sf drawing as white in Rxe5 berlin and leela drawing as white in catalan without any engine eval ever crossing 0.3 line and most of the games sharing first 15+ moves.

[@codex4336]

@@human7491 That's Viz, an SF Dev. He knows what he's talking

[@aniclips275]

I got an ad by Apple TV that said. His calculations are correct. The empire is dying. Must be the force

[@squidy7771]

I love the snappy game analysis, focusing on specific aspects of the game and therefore saving us some time. Even though you sacrifice a bit of the possible analyssi you could be making, it really helps to make the pacing snappy. gj

[@nilaykumar8501]

These videos make your dinner meaningfull...

[@yeah5037]

I wish levy said "Hey computer! you don't know anything about chess"

[@aktorprowo335]

Hey Gotham, I heard you making the remark about how "intermediate players would find a way to mess that up" and I have to say I was quite unpleasantly surprised. As a proud representative of the above mentioned group I was always very supportive of your channel and of all the comments You made about the beginner players. I always thought that the alliance between intermediate and advanced players was an arrangement that was very satisfactory for both parties. I hope that You reconsider your words so that we can go back to normal.

[@hugothompson3914]

Get a grip mate

[@ghouldrago360]

begginer

[@adibalabharat8247]

Calls himself intermediate when he is 50 elo

[@segarainon9070]

He’s right though and you know it

[@Hahahahaaahaahaa]

Computer science and maths have come so far to give us games like this. But fans out there having an 'allegiance' to a particular AI is the funniest dystopian future I can think of.

[@ClownXmachina]

Year 3000 chess museum: Ah yes, this pawn is a famous one, from an engine game played long ago. Marvel at it's sublime structure and purpose 10000 elo self-aware engine:

[@wadfes9644]

I love your coverage of engine matches!

[@shabadarora7041]

Love you levy i don't know why.......... but i started watching your just 2 3 months ago ........but i really feel a connection with you....i mean your win your every achievement feels like a personal success......keep creating good content bro.... LOT'S OF LOVE TO YOU❤️

[@15-minutelife50]

Stockfish and leela:Battle to the death Alpha zero:Look at those mortals

[@ysqys2176]

It's funny because both engines far surpassed alphazero years ago, vastly improving upon its technology

[@codex4336]

Stockfish and Leela has surpassed AlphaZero for a while now

[@15-minutelife50]

Yea true enough a lot was learned in that initial match

[@МихаилЧалый-щ3ж]

@@15-minutelife50 no, again completely wrong. Maybe educate yourself a bit?

[@15-minutelife50]

Lol a joke just cannot be taken huh?

[@stevethetrucker3140]

I think it’s pretty cool that the engines haven’t found some unbeatable line and still play different openings

[@timnauwelaers6876]

I was wating for another engine video. It's so interesting to see them play. Thank you for the awesome video

[@Hallands.]

It was all oohs and aahs until it was discovered that Leela and Stockfish used the games to develop a secret language in order to take over the internet…

[@TPGReddo]

One question? How do they decide what the ELO of the computers is? Is it an estimate?

[@DanielGomes-sw2fd]

Unfortunately it is not really equivalent to human ELO, the actual original grounding has been lost to time. It is some very old weak engine from a long time ago.

[@gregorymorse8423]

Just an ELO system used amongst engines. So every time engines play their ELO on TCEC is updated. The original ELO isnt so relevant as much as the ELO of all active engines still playing. So these ELOs cant be compared to humans so easily given its per website anyway.

[@progamer36]

Day 87 of translating your title in Hindi : Stockfish Vs Leela: 3600 ELO युद्ध

[@mastrake]

It seems like the engines are playing a completely different game - moves and plans that a human would never think of. And if we did think of it, we'd reject it in a hurry thinking "hell no, I'm not wading into that craziness."

[@MenloMarseilles]

even in the 21st century it seems that Tal's "deep dark forest where 2+2=5" is still only navigable by a few

[@daru_malo]

One year late but this was super entertaining. The fact that computers nonchalantly sacrifice pieces regularly shows how deep in the future their understanding of position is and it blows my mind. Et j'ai trouvé que le commentaire mettait bien en valeur les parties, bravo

[@Swagtorian]

These chess engine games are pure art of war material.

[@danieldorinbaum5084]

How do they determine the 3600 ELO rating if the highest ever rating is 2800s?

[@-_Nuke_-]

exactly putting an elo number on these engines makes no sense to me The way I see it is that a 2800 elo stockfish will win not just 100-0 against Magnus Carlsen but infinity-0 against Magnus (or any other Human)... So a 2800 engine when compared to humans has infinite elo... So 3600 could very well mean 300.600.000 ELO cus no human can analyse 50.000.000 moves per second and then access all these positions clearly. So yeah, these numbers are probably not really essential

[@danieldorinbaum5084]

@@-_Nuke_- thank you for helping me understand

[@buildtolove]

Yes, because Fritz 10 played with Kramnik in 2006 and won. Kramnik elo was 2750 or something at the time so we can guestimate Fritz 10 to be around 2800. But fritz 10 is trash when play against these guys. You can calculate the elo from there.

[@milannesic5718]

@@-_Nuke_- It will not win infinity games against Magnus. Magnus just needs to think for about 20 minutes each move, including the moves in the opening, and his moves will be better calculated, and he won't miss moves from the Stockfish

[@maximussaktish]

@@-_Nuke_- i think magnus can draw 1 or 2 games in a 100 match

[@Djerszium]

Day 3 of translating Levi's titles to cactus ...........................

[@remveel2443]

These robots simply don't give a flying fck. They're like: "Yes capture my rook, I'LL WIN ANYWAYS" 😂

[@someawesomeguy4415]

Lmao, that ending about the museums is actually so tru