Stockfish Crushes the FRENCH DEFENSE
ฝัง
- เผยแพร่เมื่อ 20 มิ.ย. 2024
- How does a 3600 SuperComputer destroy the French Defense in less than 40 moves?
0:00 Intro
1:30 Stockfish vs ChessFighter Game
Stockfish 16 vs ChessFighter 3.11
TCEC Bonus Game 2022
1. e4 e6 2. d4 d5 3. Nc3 Nf6 4. e5 Nfd7 5. f4 c5 6. Nf3 Nc6 7. Be3 Be7 8. a3 O-O 9. dxc5 Bxc5 10. Bxc5 Nxc5 11. b4 Nd7 12. Bd3 f6 13. O-O fxe5 14. Bxh7+ Kxh7 15. Ng5+ Kg8 16. Nxe6 Qh4 17. Nxf8 Nxf8 18. Qxd5+ Be6 19. Qe4 Qh5 20. f5 Bf7 21. f6 g6 22. Rad1 a6 23. h3 Rc8 24. Nd5 Ne6 25. Ne7+ Nxe7 26. fxe7 Re8 27. Qxb7 Nd4 28. c3 Ne2+ 29. Kh2 Qh4 30. Qf3 Nf4 31. g3 Qxh3+ 32. Kg1 Bc4 33. gxf4 Qh2+ 34. Kxh2 Ba2 35. fxe5 Bb1 36. Qf8+ Kh7 37. Rf7# 1-0
♙♙♙♙♙♙♙♙
♖♘♗♕♔♗♘♖
NAKHMANSON GAMBIT CHESSABLE COURSE ► ► www.chessable.com/nakhmanson-gambit/course/51578
♟︎♟︎♟︎♟︎♟︎♟︎♟︎♟︎
♜♞♝♛♚♝♞♜
WATCH ME LIVE ►► www.twitch.tv/vampirechicken
Second channel ►► / @vampirechicken
Discord ►► / discord
Lichess ►► lichess.org/@/zolpi
Support my content ►► paypal.me/jonathanschrantz
End Music:
"GOT THE CHESS BLUES?" by Leenhorn: • GOT THE CHESS BLUES?
Would you like to see more computer games analyzed?
Yes
Yes
Yes
Yes
For sure
If an engine knows its force lost, then it could make any move even ridiculous ones. As no matter the move the eval won't change. Stockfish has a better heuristic where it considers the efficiency of winning or losing e.g. number of moves hence putting up the most resilient defense or finding the fastest mate. But some engines work from a pure game theoretic approach without considering such conversion efficiency. NN engines might also depending on training make bizarre choices if they learned ideas like trying to force a stalemate, etc
This is exactly correct.
That's cool
Yes, if an engine knows it has lost. It doesn’t matter if it then chooses to lose quickly or slowly, it still loses.
SF from SF 12 on has neural net tech built-in.
@@dannygjk From my understanding, the primary purpose of NNUE is to cut down the size of the game tree by eliminating obviously bad moves from being searched to high depths. Stockfish still evaluates positions based on a heuristic formula based on piece position, material, etc. I think. At least that was what I took away from reading Stockfish's code for half an hour.
I have heard from other people that the reason that chess engines often throw the game in lost positions is because it is trying to avoid a tablebase loss. The engine just looks at the 7 piece table bases and evaluates them as being worse than giving up the queen, because the engine is trying really hard to stay out of a normal tablebase position. It is much faster to check if an endgame is lost, than for it to calculate a couple of moves to mate. The engine will see a thousands of endgames that say mate in n-moves which it compares it to positions where they have no pieces and it evaluates it as -13. It's a kind of horizon effect that happens in lost positions that developers don't really care about fixing.
Ah this makes a lot of sense, thank you
This definitely sounds like the more plausible of explanations
.... Qh2 the cry to humanity saying I have a conscience and I surrender !
It’s good to know I play like a computer because I saw g3 right away. I mean, I didn’t notice it hung h3 but I definitely would have played g3.
my favorite part of all the "engine crushes engine" French defense games is that apparently the French is still solid but very decisive
As a French player, I’m trying to learn from White’s side but struggle against the Steinetz
I struggle vs the winawer ...alot of theory
2.Qe2 say bye bye to your theory lmao
@@pushingyourownagenda9889 I play the Winawer but I do better against it than I do with it. Also, way to many morons play the Exchange Variation
I found you through Eric Rosen and I have to say your content is phenomenal. Thanks for the great chess lessons!
About the computer throwing off the game at the end, I don't know much about this engine, but designing AIs myself, I could understand a possibility of why it happened. When calculating the lines, sometimes computer judge that all the moves are losing no matter what, and so it loses its preference of moves, as it won't change the result of the game. If that's the case, the engine is obviously coded to play against high rated opponents that won't blunder the game. Otherwise, one would want the engine to fight harder not to lose. Anyways that's just one of my suppositions
What about tablebase theory I posted below? I mean I never heard of engine throwing off because it's +3 or +4 in opponents favor. Normally they do it only if they see mate or pseudo-mate.
I made a simple chess engine recently and I encountered this problem where the engine would throw games if it saw it already lost by force. That queen move was probably the first move in its move list.
@@DonFreeq I believe these engines do not see anymore +3 or +4 evaluations. As the depth tends to get bigger, the evaluations tend do be quickly +50, or already a forced mate in non-drawish position. Especially in these computer tournaments where the computers used are extremly powerful, I wouldn't be surprise to see an engine spot a mate in 50 moves instead of a +3 advantage
It happens when all the losing positions have equal evaluation, in Stockfish case it chooses the 1st move on the list.
I often play like a super computer, like giving the Queen for nothing.
I'm very happy that you have this upload frenzy! Lot's of great content created and released in dazzling speed. Good variation too! I love the ingenious wins against Stockfish in impossible gambits, but seeing coverage of humans and bots is great too!
you deserve so many more subscribers man. I have been watching your videos for a while and just realized I haven't been subscribed, I have subscribed now and was surprised that you didn't have more subs given the quality of your content. Keep it up dude and good things will happen!!
Yes please Jon! Your presentation has improved so much. It's quick and sweet!
Cool video. I love the way you explain things
Does Stockfish beat Stockfish playing the French Defense?
Probably.
That queen sac was probably an last ditch effort to make a draw.
If an engine evaluates forced loss as a constant negative number, then in a completely lost position all moves will be equal. In this case, the engine could either play the first move in its list or a random one. I had this problem with a simple chess engine I made in Java recently.
When I first made game AIs long ago, I ran into this and a corresponding issue of not really trying to win when the game is a lock. I called these twin problems "arrogance" and "despair". The solution, as I'm sure you discovered, is generally to value quicker wins and longer losses higher. When you've got a lock, win as quickly as possible, and when you can't win, hold out as long as possible. I guess there's some value to making moves like the engine here, throwing the queen away, though. People are weird and can somehow just miss really obvious things in unusual situations. I remember even seeing a position where a GM missed mate in 1, possibly because it just wasn't a common pattern.
@@johnmartorana196 I ultimately ended up making the engine choose quick wins and long losses. I also had that problem of "arrogance" too. I really like the terms you used. I used to just say that the engine was "throwing" but "despair" and "arrogance" sound cooler.
Very interesting thread. Thx guys.
Going back to the 1970s, we have observed The Horizon Effect, where if the computer sees a disaster like forced checkmate, which is not unreasonable since seven piece tablebases give the distance to mate, the computer would make a bad move that pushes the forced mate beyond the depth of its current search.
Whst?
WHAT?
I feel like computer games look so much messier than this most of the time. I'm surprised how clean this one looked.
My best guess is that this computer knows he was lost in that rookendgame. And because computer can't resign. It just looked how it good be mated as soon as possible to give and don't be in a longer agony then needed.
“ChessFighter” is an ironic name for an engine that simply throws in the towel when the goin gets tough.
Quite fitting that this happened the day after I came home from France.
looks like that Q h2 gambit has been refuted
Thank you!
I think the rating gap is more relevant here than the rating of the weaker side, as impressive as it may be. Sure it's probably impossible for a human to beat ChessFighter but if you subtract 1000 ELO from both sides, Stockfish is the equivalent of a strong GM playing against a candidate master.
I think the gap grows exponentially, so we can't really understand the difference between 3200 and 3600. It's scary.
Yeah, the way Elo ratings work is 400 rating difference is always 1% probability of upset, assuming same performance as in the past. Doesn't matter if it's 600 vs 1000 or 6000 vs 6400. Elo ratings are based on probabilities based on past performance.
Tha t fish picture is cool
thanks for showing this game bro just what I need Be7 is a tough one
Thanks, good line.
That Botez gambit must have been a bug with that Chess Fighter engine.
Can you share opening moves that are played by computers or correspondence chess which humans don't play?
Nice thumbnail lmao
Nice video
Mouse slip -:)! Interesting analysis thx!
If an engine knows it has lost. It doesn’t matter if it then chooses to lose quickly or slowly, it still loses.
Whenever you analyze stockfish playing a popular opening I'll be here. Stockfish is my favorite opponent.
At this level of immense omniscient awareness of the game, does the winner go to whichever side has white and goes first?
No. He said at the beginning of the video that 9-% of the positions end with a draw.
I always get this position as black, but from this Alekhine move order!
e4 Nf6 Nc3 d5 e5 Nfd7 f4 c5 d4 e6 Be3 Nc6
and black scores very well (53% at Lichess 2000+)
There are many lines to deviate for white, often with a notable advantage for black. This is what you get for not playing e4 Nf6 e5!
Yes, I also play the Alekhine and Nc3 is very common. There are some nice tactics in some lines, like the one where a white knight is pinned on d5 (with the white Queen on d4.
@@brasileirosim5961 exactly :) Also after e4 Nf6 Nc3 d5 e5 Nfd7 (f4/Nf3) c5 it's always possible to respond to Nxd5 with Nxe5 with the counterattack on white's d5 knight.
I'd say that opening deserves a video, on the other hand I don't want people to know how they should play against this. xD
Nevvva be a chickkkkkkkken when ya loooose 🎶
I kept trying to understand why you keep saying that Stockfish is "up a pawn." ChessFighter had a rook and two minor pieces against Stockfish's two rooks. Sure, Stockfish had more pawns, but I never really got how understand how you were calculating the material imbalance. When I hear the phrase "up a pawn," I assume that there is a complete material balance with the exception that one player has one more pawn than the other. That was not the case at any time in this game.
I think ChessFighter didn't actually go "crazy." It probably saw a force mate for Stockfish and what chess engines usually do is try to force its opponent to make the longest possible forced mate sequence, even if that means making seemingly ludicrous moves.
After reading previous comments I think that the case may be that if the computer see something like a forced mate or maybe do not see any way to win it evaluates all moves as plus or minus infinity so the chosen move do not matter at all
I have even an impression that Stockfish make moves that accelerate its defeat when it realises that a situation is hopless :p
Engine just gave a spite check
"never play f6"_ Gold, Fine
I think what happened was the engine calculated a line that ends up in a 7 piece tablebase loss and opted against it (tablebase positions aren't played out and are either drawn or resigned immediately).
It comes down to the engine design, moves are probably categorized into winning, neutral, and losing. The engine is most likely programmed to pick the top winning move and if none are available a move that keeps the evaluation neutral. Once no moves are winning or neutral the engine may have no directives on how to select losing moves, or the move that most prolongs the game.
very nice
where did u find dev16 ?
Weird. But a great upload thx!!
Stockfish decided it won't defend it's title next year due to the competition format.
Interesting the a3 move : forgetting the plan with long castle for White. Maybe it's a clue that Black can play the a6 move anyway and not the "waiting" move Be7. Usually White has a lot to do along the g1-a7 diagonal. As a french French player a6 and then maybe Qb6 or b5, depending on how White is defending the d4 pawn.
I also found that comment about a3 interesting.
I'm a long time French player that learned the opening all on my own decades ago and never though a3 was that significant a move. There's definitely an idea behind the move but my initial superficial feeling is that it's a suspect move. Traditional thought is that you shouldn't be wasting moves where you're weakest... Not only are you using up valuable tempi but you're creating weak squares. Of course a3 is arguing that grabbing some space and possibly some critical squares Black would normal count on compensates but that's not supposed to work out well. Of course, chess engines today are turning traditional rules inside out so the idea should be investigated. But if I saw that in a game I would be unconcerned and especially if a3 was followed up with b4 (expending a second tempo) because that would mean White's attack on the K-side would be that much slower.
Was …f5 better than …f6? blocking the diagonal. Or was it already too late for Black at this point? I am curious if the whole line with …Be7 is suspicious in view of a3…..
What if after ....Bc4 for black attacking the rook on f1, white would play Rd8 ?
I thought stockfish 15 was rated about 3782. So is this an earlier stockfish?
First of all reading doesn't hurt ;p Secondly you are reffering to CCRL blitz ratings where here you have TCEC ratings. Thirdly these are most likely live ratings cuz chessfigther has 3224 and it was ranked 3225 after season. SF 16 had no rating as this version had been just introduced and didn't play a season yet.
Yes when you are almost winning the engine goes nuts seen it many times before.
When a3 happens, I move a5 to avoid queenside expansion
New gambit for white in the Caro Kann: -0.2!
1.e4 c6 2.d4 d5 3.Nc3 dxe4 4.Nge2
Also if you think the Scandinavian is boring to play against, you can play the Dunst-Perrenet Gambit via this move order:
1. e4 d5 2. d3 dxe4 3.Nc3 exd3
you should’ve analyzed the line where the 3200 rated computer did not take E5 with f6 pawn?
I'm french, but i don't like to play french defense lol 🤣😂
GOOD
Chess Finder realized he was losing so he decided to go for the line that gives him the most chance of a draw by stalemate or such
I wonder if the engine could crack the rubinstein defense of the french
I hit resign when they do that
Why not . . . Nf4 rather than ...Nd7 ? Seem way more active for black.
Nooo I love the French defense
There was a game LC0- Baron in WCSC in this variation with similar B*h7 sacrifice. There, black even had a bishop on e7 covering g5 but still Lc0 manged to win with another beautiful exchange sac .Was one of the most beautiful games i'd ever seen. Take a look. I saw it on the silicon road channel
link- th-cam.com/video/gY2-1v9toYM/w-d-xo.html
And yes,u should continue doing computer games.
Maybe most current high level games end in draws because we live an age of ~3000 rated engines and players.
Perhaps, as time goes on and computing power increases exponentially (and engines have ratings of 10,000 or more), we will enter an age where White wins every game!
Your speculation is unlikely.
Draws have almost always dominated highest levels of chess whether it's man or machine. Reason is because mistakes have to happen for one side to win. There is no known forcing line that White or Black can play that leads to a win.
Because mistakes are a lot rarer in highest levels of chess (or at least identified as mistakes), games are often played where mistakes significant enough to tip the balance are rarely seen. Only special talents like a Tal or Fischer have been able to break the mold which makes them that much more special. It's far more likely and practical to rise to the highest levels by eliminating mistakes from their games rather than depend on some extraordinary talent that can take a game by force.
As usual it turns out that most engines just don’t really understand how to play the French, and end up making elementary positional mistakes or just greedily go pawn grabbing - although at least Stockfish understands that that 9…Bxc5 was a mistake, and 9…Nxc5 keeping dark squared bishops on the board would have been fine for black.
9:30 - Chessfighter calculates that every move is a losing move which makes all moves equally bad/good in it´s mind. So it´s pretty much a random move which would be crazy if it played against a human. Versus Stockfish it don´t matter.
I followed this game with Stockfish and from time to time black makes moves that Stockfish really doesn't like, saying they lose about half a pawn.
For example:
Black taking with bishop on c6.
Black pawn to f6.
Black pawn taking on e5.
Hard to say whether these evaluations translate into real material loss for black, but if black is making moves that aren't the best, after enough such moves it's not hard to see how Stockfish can get a win against it.
Edit: After having seen the whole game, it's obvious that Stockfish can read deeper than ChessFighter, and as a result wins material and the game.
Edit #2: 9:29 About ChessFighter "going crazy" here, the thing is that StockFish already evaluates this position as -10 for black. ChessFigher knows it's going to lose. What's the "correct" play when every move leads to a loss? Is it to draw the game out for a maximal number of moves? Is it making moves that will have the highest likelihood of the opponent blundering? When every move is a loss, there is really no correct move.
Without any book and the engine Chessfighter does NOT play e5 or d5 or even c5????????
Sacrificing the queen proly bought it more time somehow
winawer with 7 ....0-0 and 13...Bd7i instead of Qf7?! is very sound. also, inthis game, why does black take c5 with the bishop and not the knight? I never like to lose the black squared bishop in this type of game
The French defense is petty much busted at computer-chess level, as it is too passive. I'm surprised they didn't use an opening book. Usually when engines play without book they play the opening rather poorly, but here they actually played quite well. Black's ...fxe5 was quite easy to refute, though.
Are you sure? This example surely shouldn't count as one engine is much stronger than the other. Do you have any examples of a fairly weaker engine playing vs the French and winning for instance? That may show it is indeed busted. If so I'd love to see a png or something of that.
@@1001011011010 In the Top Chess Engine Championship white wins most of the time in the French as far as I can remember.
Is it possible that chessfighter was programmed to, when it evaluates that it is sufficiently losing, play the worst move instead of the best move to finish the game faster? Like a pseudo-resign
Some engines does that if they see mate, but I haven't heard of engine doing it when "sufficiently losing".
My question is here: where they allowed to use endgame tablebase? If yes than the answer may be that chessfighter didn't see a way to avoid 6-men or 7-men tablebase position that is losing and therefore it would be as it had seen mate incoming (no difference in its perspective). If not... well... I have no other idea than it was used by Chessfighter anyway :P
I don't buy "sufficiently losing" explanation. However it is not season so maaaaaybe but odds are really slim in my opinion.
I don't have a lot of experience programming chess engines, but I don't think it's deliberate.
It is more likely that this is simply the first line it calculated (given that it is the shortest) and simply could not improve on it. A mate is a mate...
I couldn't find that particular game at TCEC site so I would appreciate insight on whether Chessfighter was allowed to use tablebase :)
White isn't up a clean pawn though, black has the 2 pieces for a rook which is typically some compensation but the compensation doesn't make up for black's shattered kingside and weak pawns that will fall
To pseudo-quote Han Solo, "Good against computers? That's one thing. Good against the living? That's something else" 🤣
French is a pain in the ass to go after. I've played it for 40 years on black and I still can't stand when somebody plays it against me. You be like "I got this guy" and five moves later after you got jacked by a rhetorical 2x4 to the face you're like "Dude, wh...where? How?"
That's very cool. Probably the queen blunder is simply a bug. Or an easter egg the coders put in for losing situations.
Let's be humanly kind and finish the game without the computer AND without the egregious queen blunder.😇
Computers are the " Smart Idiots " proven again 😂
The one time i beat Fritz 8, Fritz also played a completely useless move 😂
Funny how stockfish 14 nnue still says the French defense is the best reply to e4
"Crushes French defense" Yah, the computer playing the French Defense gave up. Wow. CRUSHING.
"Stockfish Crushes the FRENCH DEFENSE" isn't such an original clickbait for 6-year olds, is it?
My app does the same thing when it feels its position is weak. It starts sacrificing pieces presumably to try for a draw. Very unsporting and antichess tactics imo
Hi Jonathan, had an idea for a new concept that would probably work great for a chess channel. It's a play on the old music game "Name That Tune".
Premise is you and your guest (hopefully Eric Rosen is a repeat contestant & would love to finally see you crossover with Agadmator) take a look at a losing position (from an engine perspective) and make a friendly "wager" who could beat someone from each others subscriber pools starting with the losing position.
If that wasn't a clear explanation... I.e. You and your youtube guest take a fairly brief look at a historical losing position (the lesser well known & more abstract the better) and auction who is the bravest facing the others subs...such as you start with you could beat their 1200 sub (from the losing position), then they up you with "I could beat your 1500 sub" etc...
If you really wanne know if SF crushed the French defence, let it play french against it self. Prob still would be a draw I suspect.
But don't forget, SF can beat any engine with French defense !!!
Chess isn't a science.
Lol, chess fighter plays the French and surrenders...
Engines don't have ratings without regard for the hardware they run on. Put a 30 year old program on a modern supercomputer and it'll play hella strong. Put Stockfish on a PC XT and it will not be so strong.
According to strong engines the French Defense is inferior.
I always considered the French Defense to be very weak and passive.
I once thought that too, but then I began realizing that I blunder against the French more in any other opening. It's passive, but it doesn't mean it can't bite hard.
I don't think passive is the right word. It's solid. But it has a very clear plan of breaking down the center, much like the Caro. Not everyone's cup of tea, but the masters still trot it out. It's my highest percentage winning opening as black.
Perhaps passive, but very irritating for white players who like open games...
"I'm not one to criticise a 3200-rated chess engine"
Oh yes, you are :)
Day 13 of asking for a video on my funny English lines
Against a 3200 elo engine, SF would win in 20 moves about every other game.
doesn't count. Might as well pit a 800 elo player vs a 400 nd base the title off that. The opening is crushed if and only if SF14 consistantly beats SF14.
Your services will soon be requested as prep will now be against stockfish. Someone asked how do you beat stockfish? I know you've got some answers! I hope you're enjoying the chess drama today. Yikes!
Never play f6
I'm so confused that the presenter does not notices that black has more pieces? Up one pawn is not winning when you have one rook for two minor pieces. It is all about the positional advantage.
At that moment white had a rook and two pawns in exchange for a knight and bishop which is 1 points difference, kinda threw me for a bit too
You can learn French in high school and college. It's a clever way to communicate with the people you really want to hang out with.
e6 is mostly a compromise, though. Let's be honest about it.
unless you live in western Canada where you don't want to be caught dead speaking French
Hate the french. Always thought it was dodgy with fight square bishop stuck for 10 moves 🤣 and stockfish confirms it now 😂🤣
Too much "looks like" analysis. As black you need to know what "should " be played, or what "must" be played, and have authentic historical examples for any serious analysis, rather than just another video.