Chucking all your pieces to loose a game is childplay. Forcing your opponent to checkmate you while being up the maximum amount of materiel, now that is genius
Forcing WorstFish to checkmate you is the impressive part. Farting around with all the queen promotions is fun and all, but capturing their pieces and promoting your pawns is not impressive given that the bot is specifically not trying to stop you.
Losing against WorstFish is actually pretty difficult since it actively walks into 1-move mates, and it's easy to end up in a situation where those are ALL of your moves, and you win.
@@scottwarren4998 that's impossible. Gotham already did a video about WorstFish vs WorstFish. If you don't force them to end the game they will just play on and on into a draw for eternity, both of them blundering mate forever. You can't force WorstFish to win by playing worse than WorstFish. That's like trying to beat Stockfish fair and square. It can't be done.
@@paul_warner Maybe a new worstfisch can lose to the old worstfisch? I don't get why Catius is not trying to play the worst moves, is he too afraid of that since he thinks worstfisch will play worse in a better way than him?
CautiousSeat: masterfully manoeuvres WorstFish into a position where it is forced to play checkmate. WorstFish: I'm gonna do what's called a pro-gamer move *resigns*
This is probably one of the best chess games I've ever seen, a bot that is programmed to lose no matter what it takes being forced to checkmate its opponent... SENSATIONAL
@@purplewine7362 When it plays the worst move, it gives you a chance to punish its blunder. However, if it were designed to lose, it would try to force you to win.
@@purplewine7362 right now the bot would blunder mate in 1 instead of forcing you to checkmate in 3 It has no strategy, for the bot the worst move is the worst evaluation
Throwing is one thing - but forcing an opponent that tries its absolute hardest to lose to actually mate you is just devastating levels of genius. I wouldn’t want someone like this on the other side of my board or opposite me in any interaction really
More specifically, Stockfish evaluates moves assuming that the opponent is going to play the best moves, which is fine when you're trying to be the best because if they make a sub-par move, it will help you. However, when trying to make the worst move, if you're opponent makes a worse move, now, the evaluation can be higher than anticipated.
There's indeed a difference between trying to play the worst moves and trying to lose the game. I was wondering why it wasn't trying to for example give checks with the queen that force you to take it.
Yes, the worse move is conditional on your opponents ability and likely responses. The best move is too. Against a 500 elo, setting up a cheeky discovered attack or long range mate-in-one is very likely to work and give you a larger advantage than the "best move" recommended by stockfish who judges things according to optimal counter play.
@@QuantumHistorian Yep, Stockfish's best move involves a 15 move sequence or more to give you a +0.2 advantage...when you can literally Scholar Mate your opponent :D
worstfish doesn't always make the stockfish minimum move, because stockfish always assumes taht future moves will be optimal. In the opening scenario, f4 cuts them off from giving up their knight on that square, not something stockfish considers
The beauty is not just that he made the worst bot to win by checkmate, it's the sophistication on making the bot beat Black AFTER achieving the _highest_ material advantage possible here. IT'S NOT JUST MAKING THE BOT WIN SOMEHOW, WHICH IS A CHALLENGE ON ITS SELF BUT.... ALSO ACHIEVING THE BEST CONDITION FOR THE BOT TO WIN! UNFATHOMABLE!
It also doesn't make sense if you think about it at all. Like of course you're a winner. Your objective in a game isn't to stop your opponent from getting his or her desired result. It's to get your own desired result. Playing chess against someone trying to lose can be considered a co-op game.
@@byeguyssrymy understanding of that is that when going up against someone trying to win, by checkmating them you will the game but you don't win against them, but if u get checkmated by someone trying to lose, you lose the game but beat them and win agenst them
@@byeguyssrythat's true. Winning equals gaining your objective. It doesn't matter if it helps someone else or not. But, if we say like this that your "win" is not letting other's win on their own goals (envy!!) Then, the quote makes sense. Depends on what your "win" is. You don't win the game, but you win your own accomplishment.
the problem with WorstFish it is that it is trying to find the best move for the opponent, which is different than trying to force the opponent to checkmate you
Yeah, trying to play badly against an opponent trying to win and trying to play badly against an opponent trying to lose require different moves. Also depends on the skill level of the opponent.
yeah, worstfish is not programmed to lose to an opponent trying to lose. it's programmed to lose as quickly as possible against an opponent trying to win. so, the player in this video didnt actually beat the bot, they just played in a way that made the program useless. if worstfish played the worst move assuming the opponent would also play the worst move, then worstfish would never win.
For now, I don't think any proper bots that were trained/optimized with the goal of forcing a loss against an opponent with the same goal exist. There's different strategy to that than is involved in trying to win, so the worst rated move from a bot that evaluates how likely each move is to win is not necessarily the best move when trying to force a loss against an opponent with the same goal. However it is very feasible to train/optimize a new bot in this way. You'd need to invert the loss function or the criterion in the minimax function depending on what bot its based on, and train an evaluation model to evaluate how likely it is for the opponent to be forced to win. Then you'd need to adjust the search tree pruning techniques for considering positional insights that are important for forcing a loss rather than for trying to win.
@aramealexanderian2403 There was an old Gotham video where he showed the worst move in chess, in a real game. Someone had mate in 1 on the board, but moved to the wrong square, and the only legal next move was for the other person to capture that piece and checkmate them. Simple case of not seeing a queen across the board lol
@@scottwarren4998 bro he did lose by the bot intentionally and yes he is smater than the bot and even if he plays the worst moves as he did, he bot would just not eat the piece as it did not
@@RishabhSharma10225 No its really not. Don't say if you don't know. Hes said it soo many times thats its levY with a Y but people just love to spell it levi!
WorstFish is officially a GM. To know the worst moves you need to know the good ones Edit: This was a joke. If you don't get it, sorry. Please don't take it seriously.
its actually really stupid because it would simply never happen. the game itself is just a draw as checkmate would never happen. u cannot self mate yourself, your opponent has to checkmate you, so you could never force a self mate. the opponent has to willingly checkmate you, but they wont because thats not their goal. maybe u can change the rules to do something with material@@leonaise7546
Chess is trained with mini-max which assumes your opponent id still playing to win. If you wanted the AI the lose on purpose when the opponent it's also trying to lose, you'd need to invent a new evaluation function. Interesting idea.
This game reminds me of a story of an AI Mancala tournament. The finals were an AI that was a heavy favorite, it had beaten all its opponents by 5, 6, 7 stones, which is a lot. The other AI had only beaten every opponent by 1 stone. After the finals the underdog had won, by 1 stone. The difference between them was the favorite had been programed to figure out the move to get the most stones at the end of it's turn. The winner had been programed to get the most additional stones after both it and its opponent's turn, which was always a difference of 1 stone. Worst fish isn't the worst possible AI because it didn't calculate that if it lost all it's normal pieces it could be forced into a position where it is forced to checkmate its opponent. CautiousSeat was looking to force their opponent to checkmate them.
Worstfish doesn't really try to lose exactly - it's more that it tries to put itself in a position where "even if someone else took over and played perfectly from that point on they would still be losing". When worstfish is evaluating the position, it's assuming that everything after its terrible move is being played optimally (not just the opponent, but also its own moves too), so if it sees that a move can force a checkmate, it will never make that move even if someone actively trying to lose could choose to make that move and then just not follow it up with the checkmate afterwards.
Incredible. I thought I had CautiousSeat figured out. But I didn't notice until the end that they still had every piece and promoted every pawn to queens, and then FORCED WorstFish to checkmate them. Absolutely amazing stuff. This is why we are greater than machines.
I don't see any difference between checkmating someone or forcing someone to checkmate you, especially in the case where you know what moves your opponent will do, everyone could do it, just let him move without taking your piece and avoid stalemates
Why are we greater than machines? That was the point CautiousSeat won the game by not letting the other opponent get its desired result which is to lose.
I’ve been saying this since the first video. Worst fish doesn’t play the worst moves, it plays the worst moves assuming your opponent plays the best moves. You must assume your opponent is trying to lose more than you are to make the worst moves possible
It would still lose cause stockfish would, for example, move the queen into a position that threatens a piece, then on the next move worstfish would move the queen somewhere that it can be taken for free, and then you repeat for every other piece. What might be interesting is Levy vs sub-1000 rated subscribers, but they can redeem a limited number of worstfish moves (3 might be a good place to start) and try to win by strategically using worstfish to sabotage Levy at opportune moments.
Reminds me of the old South Park episode where the boys don’t want to play baseball all summer so they purposely try and lose the game; the problem is that the other team practiced at being bad at baseball. This is also the episode where Randy fights other dads in the stands “I thought that this was America!”
Wow, this is one of the most fascinating games Ive ever seen. I love this kind of chess, watching the best in the world like Magnus is fun, but this sort of creative insanity, or the sheer chaos of low elo games, that can make for even better content imo
being good at chess is difficult. forcing your opponent to win when hes playing the worst moves every time while having the 102 point advantage is hard
There are 8 pawns on each side, so 16 pawns in total. If all these pawns somehow promote to queens, that would be 16 queens. Plus the 2 queens in the beginning So the max is 18 queens, or 9 queens per king which is nuts
The chess bot is programmed wrong , it plays the worst moves expecting the opponent to play the best moves next. Instead it should try to force self mate and lose peaces forcefully (Like giving the opponent only one move to take)
But that's the reason it's programmed right, because it isn't programmed to lose, it's just programmed to play the worst possible move at the given time
@@i_never_had_a_burgerits not the worst move because it selects the worst move in a list of moves under the assumption of perfect play of BOTH parties. This means: 1. the opponent is trying to win and 2. it assumes that it will play perfectly after the move Even if you think that 1. is not important, assumption 2. is. A true worst-move engine should assume that it will try to lose in the future. (currently it is operating on the assumption that it can salvage/save the blunder in the future instead of trying to make it worse)
this is a work of art, but I would like to see a bot that actually tries to force itself to lose, rather than just picking the worst move assuming the other player is trying to win
I thought the goal would be to lose, but then halfway through, I wondered why he took so many pieces, I thought that he was going for a stalemate instead, I only clocked what was really happening when he promoted his 3rd queen.
This. Is. Brilliant. I was expecting Black to take the last pawn and checkmate White, achieving a perfect game... but a perfect game is not the desired outcome against a bot that is trying to lose.
I think I figured out at 15:19 the "forced mate" (forcing white to mate black) in about eight to twelve- get a black queen to G4 to "trap" white bishop on the fifth sixth and seventh ranks (and to keep white king from moving), move the knight to e8 to allow the white pawn eventually to move forward, surround the white bishop with queens on the open white squares diagonal to the bishop forcing the bishop to take a queen when there are no other legal moves for white, recapture the bishop leaving white one move, pawn to g7 mate. Thank you for this ELO rated 900 puzzle rush you gave me to try at 15:19 to force mate from that position, obviously I am not good enough to plan the mate from the beginning as CautiousSeat admirably did.
so basically cautiousSeat is creating a so-called "selfmate", which force the opponent to checkmate him as this is the only legal move. This is indeed not that easy!
and actually here comes another task: can you selfmate yourself to the Worstfish and not even let yourself getting any chance to mate Worstfish in 1. (that is to say, however hard Worstfish tries, it can't put itself into mate in 1 by you)
I thought the best way do defeat it was to lose all your pieces by check the WorstFish's king and forcing him take up to a stalemate. But this... This is on a whole different level of madness...
At the start of the video I was trying to figure out how he could possibly force a mate… I was thinking about like trapping the Queen and stuff so that it’s only move is delivering mate, and it was not working. I didn’t even think of trapping the king and making the last available pawn move mate 😅
Black lost the game by getting their king checkmated. Checkmate occurs when a player's king is in check and the player cannot do anything to escape. In the final position, Black's king on h8 is in check from the pawn on g7. Looking at the three ways to respond to check, none of them work: - "Move your king away to a safe square." This is impossible: g7 and h7 are attacked by the white king, and g8 is occupied by a black bishop. - "Block the check." There isn't even any space between the king and the pawn, so this one won't work. - "Capture the piece giving check." None of Black's forces can capture the pawn, so this fails as well. Thus, Black's king is in checkmate, so Black loses the game.
Oh, I see. At the timestamp, the board is on the position right before checkmate. Well, just know that White moved their pawn to g7 on the following move, and that was indeed checkmate.
If you really think about it, if there was a chess variant in which you lose in order to win, this would be a normal game, and in fact WorstFish would be a top engine! Instant edit: I don't actually know for sure this variant doesn't already exist, if it does than we need to match this bot with people!
worstfish would actually be one of the worst engines for that variant because it assumes that it itself will play perfectly in the future (and so does the other player). Its pretty much an engine that can only look into the worst moves at depth 1 and it will self sabotage itself into being unable to strategize because its position evaluation only maximizes "worstness" for one move (and for depth 2+ it actually self-sabotages because it maximizes for the best move again)
@@hypnogri5457 This comment doesn't really make any sense to me. If you take into account how engines work, the search process, and the depth of the expanding tree of moves looked at, it makes sense that if you programmed stockfish to provide the move that made last place in the overall evaluation, it means that this move made it to the last place in the deep search of the 20/30/etc. move depth. The meaning is that it would be the worst move in a tree of evaluations, with the same depth and searching process as the move that came first (the best move). (Basically it's the worst move as far as the evaluation goes, for the next 30 moves (given that the depth is 30), not just for one move)
@@1GMitzy you are correct that its the worst move evaluated by stockfish up to depth X. But stockfish always assumes that it plays perfectly at any depth after the move you are currently trying to evaluate. What it basically does: 1. play a move 2. evaluate the position after the move has been played according to stockfish up to depth X. 3. repeat for all moves 4. pick the move with the worst evaluation according to step 2. (maximize according to the badness-criteria of worst eval)(this is what I call depth 1, I dont actually mean depth, so that was confusing. ) So what I mean is that its only maximizing according to "badness" one move deep. Internally stockfish evaluates the position using the "goodness" criteria of a good eval. edit: it only thinks about maximizing badness for one move and thats the first one. The evaluation up to depth X tries to maximize goodness again. It basically is chosing the move that will make the game as hard as possible to recover from with perfect play from both parties
you can see the flaw of this in this worstfish game I found on lichess: r2qkbnr/pb1p3p/3p4/1pp1KPp1/1n6/8/PPPP1PPP/R1BQ1BNR w kq - 0 10 Worstfish accidentally wins in 9 moves as black because it was unable to see further than one move into the future. It totally missed the tactic from the opponent because it selfsabotaged itself thinking that both are going to play perfectly after each move
@@1GMitzy a loss in 9 moves is proof enough that its a very bad engine in the game of "play-bad"-chess. I assume that an engine made by an amateur will probably be able to beat worstfish at bad chess. I think I saw one in another video that talked about this very topic and that one was made properly
I want to see the best ai that never wins. An ai that will beat a grandmaster’s sanity by beating them in every move and then lose because it blunders on purpose 20 minutes in.
Thats actually hillarious. Worstfish's move is worse than the worst evaluation because the evaluation doesn't know worstfish is planning to sack his knight!!
Chucking all your pieces to loose a game is childplay. Forcing your opponent to checkmate you while being up the maximum amount of materiel, now that is genius
when ur oponnent is a grinder,that robot is a bugatti
Lose
I think most people like to tight a game
Forcing WorstFish to checkmate you is the impressive part. Farting around with all the queen promotions is fun and all, but capturing their pieces and promoting your pawns is not impressive given that the bot is specifically not trying to stop you.
it's lose, not loose, learn how to spell.
Losing against the Worstfish should be a new speedrun category
you meant winning?
Losing against WorstFish is actually pretty difficult since it actively walks into 1-move mates, and it's easy to end up in a situation where those are ALL of your moves, and you win.
It’s easy just resign
I’m getting a world record!
Should be fairly easy (discounting resigning which is trivial), basically delete your own pieces through zugzwang and force it to mate you.
In my opinion this is literally the most impressive chess game of all time.
Right? The amount of time taken to perfect that and the chess knowledge and experience to do that is insane
Why doesn't cautious try to play the worst possible moves where Cautius ALSO LOSES THE GAME? If catius succeeds with that, he is smarter than the bot.
Think about it. wouldn't it be harder to lose to worstfisch with 3 queens instead of 8 queens?
@@scottwarren4998 that's impossible. Gotham already did a video about WorstFish vs WorstFish. If you don't force them to end the game they will just play on and on into a draw for eternity, both of them blundering mate forever. You can't force WorstFish to win by playing worse than WorstFish. That's like trying to beat Stockfish fair and square. It can't be done.
@@paul_warner Maybe a new worstfisch can lose to the old worstfisch?
I don't get why Catius is not trying to play the worst moves, is he too afraid of that since he thinks worstfisch will play worse in a better way than him?
Everyone saying “oh why is worst fish a grandmaster??” but it you think about it, you have to know what the best move is to be able to play the worst…
Well said. If you _truly_ play the worst move every time, your understanding of the game means you're the best player in the world.
...to play the worst (assuming perfect play from then on, as opposed to knowing your opponent)
See also: To get a zero on a test without cheating, you have to know all of the answers.
That's some fallacy right there 😂.. @@elderlycatpatriot
@@benjaminoechsli1941 So true.
CautiousSeat: masterfully manoeuvres WorstFish into a position where it is forced to play checkmate.
WorstFish: I'm gonna do what's called a pro-gamer move
*resigns*
gg
😮
CautiousSeat: forces him to unresign
literally the best comment I have read in all my life. bravo
even with 1.3 k likes
it is still an underrated comment
This is probably one of the best chess games I've ever seen, a bot that is programmed to lose no matter what it takes being forced to checkmate its opponent... SENSATIONAL
It's not programmed to lose , it's programmed to make the worst move
@@gokaytaspnar1355 isn't that the same thing?
@@purplewine7362 When it plays the worst move, it gives you a chance to punish its blunder. However, if it were designed to lose, it would try to force you to win.
@@gokaytaspnar1355 yes but i'm saying the worst move should force you to win. so maybe what this bot plays isn't technically the worst move?
@@purplewine7362 right now the bot would blunder mate in 1 instead of forcing you to checkmate in 3
It has no strategy, for the bot the worst move is the worst evaluation
Throwing is one thing - but forcing an opponent that tries its absolute hardest to lose to actually mate you is just devastating levels of genius. I wouldn’t want someone like this on the other side of my board or opposite me in any interaction really
Yeah man
That means they're _good,_ snce they know what to play.
Incredible. Worstfish's mistake was in not considering that his opponent may play terribly, which would actually change the moves he would do.
More specifically, Stockfish evaluates moves assuming that the opponent is going to play the best moves, which is fine when you're trying to be the best because if they make a sub-par move, it will help you. However, when trying to make the worst move, if you're opponent makes a worse move, now, the evaluation can be higher than anticipated.
There's indeed a difference between trying to play the worst moves and trying to lose the game. I was wondering why it wasn't trying to for example give checks with the queen that force you to take it.
Yes, the worse move is conditional on your opponents ability and likely responses. The best move is too. Against a 500 elo, setting up a cheeky discovered attack or long range mate-in-one is very likely to work and give you a larger advantage than the "best move" recommended by stockfish who judges things according to optimal counter play.
@@QuantumHistorian Yep, Stockfish's best move involves a 15 move sequence or more to give you a +0.2 advantage...when you can literally Scholar Mate your opponent :D
@@justinbrentwood1299 Also it assumes that after this move it itself will make good moves.
worstfish doesn't always make the stockfish minimum move, because stockfish always assumes taht future moves will be optimal. In the opening scenario, f4 cuts them off from giving up their knight on that square, not something stockfish considers
The beauty is not just that he made the worst bot to win by checkmate, it's the sophistication on making the bot beat Black AFTER achieving the _highest_ material advantage possible here.
IT'S NOT JUST MAKING THE BOT WIN SOMEHOW, WHICH IS A CHALLENGE ON ITS SELF
BUT....
ALSO ACHIEVING THE BEST CONDITION FOR THE BOT TO WIN!
UNFATHOMABLE!
"Are you actually a winner if you help the other person get their desired result?"
That quote isn't talked about enough😂😂😂
It also doesn't make sense if you think about it at all.
Like of course you're a winner. Your objective in a game isn't to stop your opponent from getting his or her desired result. It's to get your own desired result. Playing chess against someone trying to lose can be considered a co-op game.
@@byeguyssrymy understanding of that is that when going up against someone trying to win, by checkmating them you will the game but you don't win against them, but if u get checkmated by someone trying to lose, you lose the game but beat them and win agenst them
@@byeguyssrythat's true. Winning equals gaining your objective. It doesn't matter if it helps someone else or not.
But, if we say like this that your "win" is not letting other's win on their own goals (envy!!) Then, the quote makes sense. Depends on what your "win" is. You don't win the game, but you win your own accomplishment.
Ah yes. Sweet paradoxes.
@@LordNNero I think the real sense behind trying to loose rather than win here is that it is a considerably more interesting thing to do.
64 squares, 100 queens. 36 queens noclipped alongside the kings, nice
they just noclipped into another dimension
Noclip? Geometry dash reference?
@@ignDartI was about to say that…
the king is just built different
@@ignDartw reference
this kind of feels more impressive than any other chess achievement I've seen
3:55 "how do you beat something that is trying to lose?" you make them win
42
@@dantheman4205what is 42?
I love how they absolutely didn't need to get nine queens, they just did it for the big number on the side. Like a victory lap.
As the queen, I can confirm 100 of us fit on 64 squares (we had bunk beds).
🤣
As the king, well, it's gonna be a very fun night.
@@Tetracontakaitetragon_💀💀💀_
@@gracamaria508damn that's wavy ngl
@NotAWake3.14 did u kill the 90 other queens like a squid game cauz now there cause now there only 10 queens 😂😂
*Gotham getting all philosophical in this video.*
"What does it truly mean to win?"
Theres no meaning
the problem with WorstFish it is that it is trying to find the best move for the opponent, which is different than trying to force the opponent to checkmate you
Yeah, trying to play badly against an opponent trying to win and trying to play badly against an opponent trying to lose require different moves. Also depends on the skill level of the opponent.
I would love to compare that worstfish to a chess bot that plays the worst possible move considering the opponent also plays the worst possible move
@@jacobD643 wouldn't be hard to flip that around in its minimax algorithm.
yeah, worstfish is not programmed to lose to an opponent trying to lose. it's programmed to lose as quickly as possible against an opponent trying to win. so, the player in this video didnt actually beat the bot, they just played in a way that made the program useless. if worstfish played the worst move assuming the opponent would also play the worst move, then worstfish would never win.
@@bobjoe8182But assuming that, worstfish should open with shortest mate possible which is 2 move mate
It's official.
Humans ARE better than robots at being worse at chess!
even siri will still lose?!
the A.I. is trained to beat a human who tries to win
train it to lose human who tries tto lose and we'll se
For now, I don't think any proper bots that were trained/optimized with the goal of forcing a loss against an opponent with the same goal exist. There's different strategy to that than is involved in trying to win, so the worst rated move from a bot that evaluates how likely each move is to win is not necessarily the best move when trying to force a loss against an opponent with the same goal.
However it is very feasible to train/optimize a new bot in this way. You'd need to invert the loss function or the criterion in the minimax function depending on what bot its based on, and train an evaluation model to evaluate how likely it is for the opponent to be forced to win. Then you'd need to adjust the search tree pruning techniques for considering positional insights that are important for forcing a loss rather than for trying to win.
13:00 nah bros playing Tetris with queens 💀
rarest things in chess
1-en passant mate
2-castle mate
3-forced move for you opponent that is actually mate in one
Is the 3rd one even possible??
@@user-pt2el1tt1r as you see in the video, it is, but against humans no.
@@aramealexanderian2403 it's happened in a real game before. gotham has another video on it
@aramealexanderian2403 There was an old Gotham video where he showed the worst move in chess, in a real game. Someone had mate in 1 on the board, but moved to the wrong square, and the only legal next move was for the other person to capture that piece and checkmate them. Simple case of not seeing a queen across the board lol
42000 minutes into this video, I can confirm that this is indeed a GothamChess classic
29 days ahead.
Nice.
I'm the 1000th like
@@scottwarren4998 bro he did lose by the bot intentionally and yes he is smater than the bot and even if he plays the worst moves as he did, he bot would just not eat the piece as it did not
Why doesn't cautious try to play the worst possible moves where Cautius ALSO LOSES THE GAME? If catius succeeds with that, he is smarter than the bot.
@@vaibhavvishnoi7342 Catius didnt play the worst moves. he got queens and shit, thats not allowed.
Levy is such a great teacher. He teaches us how to play the worst move possible.
You cannot stop an avalanche with a horse, but you can create it.
- Sun Tsu
Finally, a worthy opponent for Martin
“You cannot stop an avalanche with a horse” -Levi 2023
*LEVY
@@gamesafoot No its Levi roseman
@@RishabhSharma10225 No its really not. Don't say if you don't know. Hes said it soo many times thats its levY with a Y but people just love to spell it levi!
Which video was this😂
@@ThanastyGoat65536it’s probably a joke
WorstFish is officially a GM. To know the worst moves you need to know the good ones
Edit: This was a joke. If you don't get it, sorry. Please don't take it seriously.
I never knew i was born as a grandmaster. 💀
indeed
@@Hello_darkworldoof
The missile knows where it is, because it knows where it isnt.
to know ALL the good ones
I really wanna see two grandmasters play the game this way
It's just two GMs trying to force self-mate. I'd love to see that
Selfmate World Championships has at least the same prowess as the fisher random WC.
This should be an official game mode. The loser is the winner
its actually really stupid because it would simply never happen. the game itself is just a draw as checkmate would never happen. u cannot self mate yourself, your opponent has to checkmate you, so you could never force a self mate. the opponent has to willingly checkmate you, but they wont because thats not their goal. maybe u can change the rules to do something with material@@leonaise7546
whoever loses wins
Chess is trained with mini-max which assumes your opponent id still playing to win. If you wanted the AI the lose on purpose when the opponent it's also trying to lose, you'd need to invent a new evaluation function. Interesting idea.
Worst fish is the smartest bot because it knows all the bad moves
It most certainly should be celebrated. Thanks for bringing it to our attention. FIDE should award CautiousSeat an Honoury GM Title!
*BM = Bad Master
This game reminds me of a story of an AI Mancala tournament. The finals were an AI that was a heavy favorite, it had beaten all its opponents by 5, 6, 7 stones, which is a lot. The other AI had only beaten every opponent by 1 stone. After the finals the underdog had won, by 1 stone. The difference between them was the favorite had been programed to figure out the move to get the most stones at the end of it's turn. The winner had been programed to get the most additional stones after both it and its opponent's turn, which was always a difference of 1 stone.
Worst fish isn't the worst possible AI because it didn't calculate that if it lost all it's normal pieces it could be forced into a position where it is forced to checkmate its opponent. CautiousSeat was looking to force their opponent to checkmate them.
Worstfish doesn't really try to lose exactly - it's more that it tries to put itself in a position where "even if someone else took over and played perfectly from that point on they would still be losing". When worstfish is evaluating the position, it's assuming that everything after its terrible move is being played optimally (not just the opponent, but also its own moves too), so if it sees that a move can force a checkmate, it will never make that move even if someone actively trying to lose could choose to make that move and then just not follow it up with the checkmate afterwards.
Incredible. I thought I had CautiousSeat figured out. But I didn't notice until the end that they still had every piece and promoted every pawn to queens, and then FORCED WorstFish to checkmate them. Absolutely amazing stuff. This is why we are greater than machines.
We are not greater then AI.
@@gsas3012 We absolutely are...for now...
I don't see any difference between checkmating someone or forcing someone to checkmate you, especially in the case where you know what moves your opponent will do, everyone could do it, just let him move without taking your piece and avoid stalemates
Why are we greater than machines? That was the point CautiousSeat won the game by not letting the other opponent get its desired result which is to lose.
@@gsas3012we still are in open environments (chess is a closed environment) for now
Was not expecting that result...
Congrats to the lad for beating Worstfish's coding and forcing it to win
I’ve been saying this since the first video. Worst fish doesn’t play the worst moves, it plays the worst moves assuming your opponent plays the best moves. You must assume your opponent is trying to lose more than you are to make the worst moves possible
This is actually harder than winning in chess lmao. Also, big props to levy still making the game so exciting even tho the goal is to lose lmao
It would be interesting to see a game where stockfish and worstfish alternate moves and play against a human
B
Human would win very easily.
Human could probably scholar's mate this bot
It would still lose cause stockfish would, for example, move the queen into a position that threatens a piece, then on the next move worstfish would move the queen somewhere that it can be taken for free, and then you repeat for every other piece. What might be interesting is Levy vs sub-1000 rated subscribers, but they can redeem a limited number of worstfish moves (3 might be a good place to start) and try to win by strategically using worstfish to sabotage Levy at opportune moments.
Good cop bad cop
I love how Levy is constantly saying stuff like "Now it's over!", like it hasn't been "over" for the last 25 moves 😂
It's over
I will destroy the 69 likes just as my 69 among us levels were changed to 70
Humans proving their dominance in the field of losing chess games once again
I'm like "what th f### is Levi on about??? why is he so quiet and why is he- OOOOOOOOOOOOOOOOOOOOOOOOOHHHHHHHHHHHHHHHHHHH"
Reminds me of the old South Park episode where the boys don’t want to play baseball all summer so they purposely try and lose the game; the problem is that the other team practiced at being bad at baseball. This is also the episode where Randy fights other dads in the stands “I thought that this was America!”
I love that episode lol
I didn't hear no bell
Fun fact: the guy who lost was actually Gotham in disguise
Maybe
Sometimes the worst move isn’t the worst move if your opponent isn’t trying to win
In Combinatorial Game Theory, this is known as "misère" play. There's a whole theory behind how to do it in a variety of games.
Sometimes the only winning move is not to play.
"If one is designed to lose, would it not be winning to lose yourself?" -Practically Gotham
and eminem too
Wow, this is one of the most fascinating games Ive ever seen. I love this kind of chess, watching the best in the world like Magnus is fun, but this sort of creative insanity, or the sheer chaos of low elo games, that can make for even better content imo
This reminds me of that Guess the Elo game, where the Gotham sub forced his opponent to checkmate him as mate was the only legal move.
Do you know how to find the video? Would be interesting to watch.
This is insane… I hope this gets taught to beginners as a shining example of human ingenuity
This was beautiful. Truly remarkable. All props to cautiousseat for this stunning display. He must be absolutely god tier at puzzles.
You know, I loved that "worst vs worst" match and would enjoy it as a regular feature.
being good at chess is difficult. forcing your opponent to win when hes playing the worst moves every time while having the 102 point advantage is hard
"Ok so we can't be better that bots in chess, any solutions?"
"Let's try beeing worse"
Saw the checkmate from a mile away. But what I didn't see was the absolute maniacal move of Nf5+, gaining a 102 point advantage
I'm extremely impressed.
I thought he would force a capture or two.
But to force an opponent trying it's best to lose to checkmate you? Amazing!
He didn't just beat worst fish, he humiliated worst fish.
Hm. Inspired genius, perhaps? The only way to force Worstfish to "lose" is to force it to win. That's mental.
We need worstfish vs worstfish now
haven’t seen the video but theres only 64 squares and 8 pawns im highly skeptical of the 100 queen claim Mr.Rozman
There are 8 pawns on each side, so 16 pawns in total. If all these pawns somehow promote to queens, that would be 16 queens. Plus the 2 queens in the beginning
So the max is 18 queens, or 9 queens per king which is nuts
bunk beds exist
@@eliasmazhukin2009 Average Muslim enjoyer with 72 queens
The chess bot is programmed wrong , it plays the worst moves expecting the opponent to play the best moves next.
Instead it should try to force self mate and lose peaces forcefully (Like giving the opponent only one move to take)
Perfectly said.
But that's the reason it's programmed right, because it isn't programmed to lose, it's just programmed to play the worst possible move at the given time
@Abbas but it DOESN'T play the worst possible move at any given point. That's the point.
Example: Force opponent to take queen.
@@jaideepshekhar4621 yeah, not forcing opponent to take queen was a mistake even I noticed but I think rest was fine
@@i_never_had_a_burgerits not the worst move because it selects the worst move in a list of moves under the assumption of perfect play of BOTH parties. This means:
1. the opponent is trying to win
and
2. it assumes that it will play perfectly after the move
Even if you think that 1. is not important, assumption 2. is. A true worst-move engine should assume that it will try to lose in the future. (currently it is operating on the assumption that it can salvage/save the blunder in the future instead of trying to make it worse)
"you can't defeat an opponent who allows himself to be defeated" - some random dude
I guess he was wrong
“Wow, he lost against the worst opponent, he must be bad.”
worst fish vs worst fish
Damn💀
The world is gonna end 😂👍
I think the most impressive part is how they played so brilliantly against something trying to lose!
When a blunder becomes a brilliant move
this is a work of art, but I would like to see a bot that actually tries to force itself to lose, rather than just picking the worst move assuming the other player is trying to win
the worst thing was not playing en passant twice
i loved the part when the queens combined 🗣🔥🔥💀
the L word got spicy af
WorstFish was playing a different game. And CautiousSeat beat it at its own game.
Such a great video! Thank you for making this!
I thought the goal would be to lose, but then halfway through, I wondered why he took so many pieces, I thought that he was going for a stalemate instead, I only clocked what was really happening when he promoted his 3rd queen.
In my opinion the worst bot would be one that tries to forcefully checkmate itsself.
the way to defeat it is to resign
If this isn't at the how to win chess playlist I'm gonna cry 😭
The stare today was on the spot, gave me shivers down my spine. Love it Levy!
This. Is. Brilliant. I was expecting Black to take the last pawn and checkmate White, achieving a perfect game... but a perfect game is not the desired outcome against a bot that is trying to lose.
The fact that worstfish isnt called unstockfish angers me. Also the fact that worstfish has a grandmaster title wth
I think I figured out at 15:19 the "forced mate" (forcing white to mate black) in about eight to twelve- get a black queen to G4 to "trap" white bishop on the fifth sixth and seventh ranks (and to keep white king from moving), move the knight to e8 to allow the white pawn eventually to move forward, surround the white bishop with queens on the open white squares diagonal to the bishop forcing the bishop to take a queen when there are no other legal moves for white, recapture the bishop leaving white one move, pawn to g7 mate. Thank you for this ELO rated 900 puzzle rush you gave me to try at 15:19 to force mate from that position, obviously I am not good enough to plan the mate from the beginning as CautiousSeat admirably did.
so basically cautiousSeat is creating a so-called "selfmate", which force the opponent to checkmate him as this is the only legal move. This is indeed not that easy!
and actually here comes another task: can you selfmate yourself to the Worstfish and not even let yourself getting any chance to mate Worstfish in 1. (that is to say, however hard Worstfish tries, it can't put itself into mate in 1 by you)
@@wonderwind2716 That's actually pretty easy to achieve, since WorstFish doesn't think like that.
I thought the best way do defeat it was to lose all your pieces by check the WorstFish's king and forcing him take up to a stalemate.
But this... This is on a whole different level of madness...
At the start of the video I was trying to figure out how he could possibly force a mate… I was thinking about like trapping the Queen and stuff so that it’s only move is delivering mate, and it was not working. I didn’t even think of trapping the king and making the last available pawn move mate 😅
Humanly, you gotta be sooo good at chess to play the worst possible game.
When you do at least half the moves the bad chess bot does you know somethings wrong in the brain.
I's amazing how a bot that´s programed to loose actually knows chess so much better than any human
So basically, AI lost to a human, because this was a battle to get checkmated, not win
Someone who truly deserves honourary GM title 🎉🎩
Worstfish is so good at chess, he always selectively plays the worst moves.
Levy, pair this bot with you, but every 5 moves it changed to stockfish
Stockfish: Mate in 1
Worstfish: Self-mate in 1
The stare today was impeccable 🔥🔥
Yes 0.0000005 seconds longer than the last one 😂
I am afraid he is losing his mind to chess.
Who are the people who keep liking these "stare" comments? Do they still find them funny?
4:00 worstfish blundered en passant
Gotham doesn’t even know that point advantage isn’t everything, position and development matter too lmao
He almost had him there. Maybe if he practices a little more he'll be able to beat the AI
That stare stared into the depths of my soul. 9/10
16:50 - Wait... how did he lost the game? /genq
He played in such a way that forced the bot to checkmate him
Black lost the game by getting their king checkmated. Checkmate occurs when a player's king is in check and the player cannot do anything to escape. In the final position, Black's king on h8 is in check from the pawn on g7. Looking at the three ways to respond to check, none of them work:
- "Move your king away to a safe square." This is impossible: g7 and h7 are attacked by the white king, and g8 is occupied by a black bishop.
- "Block the check." There isn't even any space between the king and the pawn, so this one won't work.
- "Capture the piece giving check." None of Black's forces can capture the pawn, so this fails as well.
Thus, Black's king is in checkmate, so Black loses the game.
Oh, I see. At the timestamp, the board is on the position right before checkmate. Well, just know that White moved their pawn to g7 on the following move, and that was indeed checkmate.
@@isavenewspapers8890Yeah, I was confused by that, too 😅
16:19 +103
And lose
he said the highest possible material advantage and lose
If you can lose to worst fish, you're a grandmaster automatically
Worstfish: I shall lose easily
Cautious: you underestimate my sheer ability to not win on purpose.
This is really really impressive
If you really think about it, if there was a chess variant in which you lose in order to win, this would be a normal game, and in fact WorstFish would be a top engine!
Instant edit: I don't actually know for sure this variant doesn't already exist, if it does than we need to match this bot with people!
worstfish would actually be one of the worst engines for that variant because it assumes that it itself will play perfectly in the future (and so does the other player).
Its pretty much an engine that can only look into the worst moves at depth 1 and it will self sabotage itself into being unable to strategize because its position evaluation only maximizes "worstness" for one move (and for depth 2+ it actually self-sabotages because it maximizes for the best move again)
@@hypnogri5457 This comment doesn't really make any sense to me.
If you take into account how engines work, the search process, and the depth of the expanding tree of moves looked at, it makes sense that if you programmed stockfish to provide the move that made last place in the overall evaluation, it means that this move made it to the last place in the deep search of the 20/30/etc. move depth.
The meaning is that it would be the worst move in a tree of evaluations, with the same depth and searching process as the move that came first (the best move).
(Basically it's the worst move as far as the evaluation goes, for the next 30 moves (given that the depth is 30), not just for one move)
@@1GMitzy you are correct that its the worst move evaluated by stockfish up to depth X. But stockfish always assumes that it plays perfectly at any depth after the move you are currently trying to evaluate. What it basically does:
1. play a move
2. evaluate the position after the move has been played according to stockfish up to depth X.
3. repeat for all moves
4. pick the move with the worst evaluation according to step 2. (maximize according to the badness-criteria of worst eval)(this is what I call depth 1, I dont actually mean depth, so that was confusing. )
So what I mean is that its only maximizing according to "badness" one move deep. Internally stockfish evaluates the position using the "goodness" criteria of a good eval.
edit: it only thinks about maximizing badness for one move and thats the first one. The evaluation up to depth X tries to maximize goodness again. It basically is chosing the move that will make the game as hard as possible to recover from with perfect play from both parties
you can see the flaw of this in this worstfish game I found on lichess:
r2qkbnr/pb1p3p/3p4/1pp1KPp1/1n6/8/PPPP1PPP/R1BQ1BNR w kq - 0 10
Worstfish accidentally wins in 9 moves as black because it was unable to see further than one move into the future. It totally missed the tactic from the opponent because it selfsabotaged itself thinking that both are going to play perfectly after each move
@@1GMitzy a loss in 9 moves is proof enough that its a very bad engine in the game of "play-bad"-chess. I assume that an engine made by an amateur will probably be able to beat worstfish at bad chess. I think I saw one in another video that talked about this very topic and that one was made properly
This video is a masterpiece, a piece of art , indeed. Ellegant narration of a thrilling accomplishment.
I want to see the best ai that never wins. An ai that will beat a grandmaster’s sanity by beating them in every move and then lose because it blunders on purpose 20 minutes in.
Thats actually hillarious. Worstfish's move is worse than the worst evaluation because the evaluation doesn't know worstfish is planning to sack his knight!!