When moving a piece that doesn't exist on the board: The piece spawns on the board; When moving a piece of the opposite color: The piece moves; When moving to a square that has a piece of own color: The latter is replaced, or taken, by the piece that moved. The game ends when either of them says #, like "my move is: 20. a1#", regardless of whether the move is legal.
Long time since I laughed so much watching a youtube video. Martin really is underappreciated for his ability to move pieces that are on the board to places where they can actually go.
I would love to see this, but ChatGPT makes the rules and if it plays an illegal move, you just edit the board to make the move happen (i.e. if it tries to take its own pawn with a rook, you delete the pawn and move the rook there, if it tries to move a bishop it doesn't have, you create a bishop, etc.)
An issue ChatGPT as well as many machine learning tasks in speak generation is that there isn't really a good way for the algorithm to remember/consider past conversations to output new statements. ChatGPT uses a type of Transformer Neural Network (which uses association in words to generate output), so it's really good at having conversations that isn't really that much in-depth but are really crappy at long conversations. So think of ChatGPT as someone with alzheimers hahaha Just thought it might be kinda cool to share (I'm a MS AI major doing research on the AI field with a professional nerd license)
Figures, I watched a game of ChatGPT vs Stockfish and ChatGPT tended to play pieces and pawns from the starting position even though they were already moved.
@@laytonjr6601 I said that because when he played Stockfish he castled without moving his bishop first, and then Gotham Chess played him and ChatGPT did the same thing! (Just so you know, Gotham Chess and Stockfish let ChatGPT play illegal moves.)
@@malachiriehl1053 There's no way Stockfish would work if the opponent played illegal moves. Someone else must have interpreted the moves and gave the new position to Stockfish.
I asked it about the Italian Game and the Caro-Kann Defence and it clearly knows about openings/defences, but after the first few (wikipedia-assisted) moves, the bot doesn't even know where their pieces are at the middlegame lol It doesn't need to abide by the chess rules anyways.
The problem isn't that it doesn't understand the rules, it's that it apparently can't keep track of the board position. As a language model, it replies with "this sounds like something a human would say" rather than actually looking at the board position and figuring out the best move.
@@John73John ChatGPT doesn't understand the rules either. I mean, it can tell you what the rules are, but it doesn't understand that the rules are describing which moves are valid and which aren't - when you ask it to make a move, it doesn't think about the rules of the game - it just looks at "what moves do players commonly make" and tries to mimic them, which somewhat works in the opening (because the opening is very repetitive so it's seen the openings a ton of times), but it goes completely haywire the further in the game you get. It sees the rules as something entirely unrelated to the moves that are made in the game.
@@asdfqwerty9241 Yeah that's kind of what I was trying to say, but you said it better. It can repeat the words of the rule book but it doesn't necessarily associate those words with how the pieces should move. It just knows how to repeat common opening moves, and then it falls back to just saying things that "sound human"
3:10: I assumed the position thing was a hack to force ChatGPT to remember the board position so it didn't make so many glaringly illegal blunders. If this is the case, it's not a very good hack.
ChatGPT seems to really like going through his own pieces, teleporting to other ends of the board illegally and moving the queen 3 spots forward and one right in one turn 😆
20:00 ChatGPT actually does not have any sources from the internet. So the information is not from Google. The Informations come from the training (when the developers feed it with information) so it doesn't know all recent informations, and it often produces false information. It is very good at sounding like a human but not really at remembering stuff from a conversation or giving completely correct facts.
Man, the hubris of Chat GPT lecturing martin on how to play chess after it got horribly confused, had to ask you for help and then chose the most useless tempo-losing move of the options presented.
From my limited testing ChatGPT is better at making legal moves if you give it the board in ascii representation. The moves it provided to me were all legal (and good), but then it messed up when it tried to print the bord. So I would suggest starting a new chat and simply paste in the current position in ascii form and ask it for what to play. It has a limited memory and keeping track of everything which happened before is not it’s strongest side. ChatGPT 4 (the paid version) might also be better at this
Videos like this are very reassuring. Everyone's freaking out about how smart AI is, and here we see it confidently spout moves when it has no idea what's going on.
"AI" gets overused IMO, this is a language model, it's only meant to generate text that follows the style of the prompt and training data to be grammatically correct English. it isn't concerned with what's true or logical or the rules of any games like chess because it's literally incapable of understanding those things. it just puts letters one after the other based on a complex statistical model.
I mean chatGPT is doing pretty well considering that it's only designed for text based responses. I have never seen chatGPT have to stop and think about something for like 10 seconds
Watching this video I thought about a new challenge you could try: play against martin but you need chatGPT's help, you make one move but your next move would be made by chatGPT and you have to manage to make the move even if it's illegal
Maybe with the Martin and Nelson things on the channel, at the end of each video, you could move your face camera to be at the top right to see the moves.
GPT-4 is better at making fewer illegal moves. I think you should enforce that 2 illegal moves in a game means a loss (unless the opponent doesn't have mating material).
This is not how you use the AI. You need to give it full detailed info when starting so that means every position when making a move you have to do the same again.
i can't help but imagine chat gpt as a normal person moving the pieces in illegal ways with unbreakable confidence
“What do you mean I only have 2 rooks? My rook on e2 takes your queen on c7!”
@@DoblAABaterymakes sense
MDRRRR
DIEUUU
ChatGPT really likes to castle without moving its bishop...
repent unto Christ
Exactly
And castle multiple times
@@Baggerz182 castle through THE BISHOP
It's a special castling technique and you can't prevent it
15:30 after playing 500 illegal moves, ChatGPT started giving suggestions on how to play chess better. Hilarious!
Wdym illegal? We humans just cannot comprehend the idea.
Pony should've taken priest on K2
@@aninstrument2803 Alright then, I move to 2... O-Ok14
@@GivingOthersAMentalBreakdown knight Z19 wouldve been better
@@Ikusofekkusu Oh yeah but i think rook T14 can be an alternative
I want to see ChatGPT duel itself illegal moves are still played.
When moving a piece that doesn't exist on the board: The piece spawns on the board;
When moving a piece of the opposite color: The piece moves;
When moving to a square that has a piece of own color: The latter is replaced, or taken, by the piece that moved.
The game ends when either of them says #, like "my move is: 20. a1#", regardless of whether the move is legal.
It's official, ChatGPT is not replacing just programmers but also chess coaches.
16:00 Martin (of all bots!) has the eval bar at +4.5
ChatGPT: Good time to lecture you on the fundamentals of chess strategy!
chat gpt really just said to me "The check is the strongest peice" 🤣
He is trying.
that as chess got
The queen is the strongest position!!
@@Fad1ngRainbowthe best move to play would be checkers
Long time since I laughed so much watching a youtube video. Martin really is underappreciated for his ability to move pieces that are on the board to places
where they can actually go.
ChatGPT is like playing several games at once blindfolded and messes up which game is which position
I would love to see this, but ChatGPT makes the rules and if it plays an illegal move, you just edit the board to make the move happen (i.e. if it tries to take its own pawn with a rook, you delete the pawn and move the rook there, if it tries to move a bishop it doesn't have, you create a bishop, etc.)
I heard ChatGPT is now serving its sentence at the chess prison for attempting so many illegal moves against the plaintiff, Martin.
I never thought I'd see the day when a bot would have a stroke.
repent to Christ
@@Baggerz182 What does that have to do with anything?
@@Baggerz182 amen
For a moment we feel Martin is strong
This must be how grandmasters view a game of mine.
ChatGPT just tried inventing the bishop en passant, and you denied it!
Imagine if chatGPT said that the best opening for a 1200 rated player was the mosquito gambit
Chatgpt plays queen to h4, teleporting the queen through the knight 😂😂
repent to Christ
@@Baggerz182 repent to chatgpt
ChessGPT: I follow the rule of this game and im making sure that your not doing any illegal move
Proceed to move knight like a bishop
An issue ChatGPT as well as many machine learning tasks in speak generation is that there isn't really a good way for the algorithm to remember/consider past conversations to output new statements. ChatGPT uses a type of Transformer Neural Network (which uses association in words to generate output), so it's really good at having conversations that isn't really that much in-depth but are really crappy at long conversations. So think of ChatGPT as someone with alzheimers hahaha
Just thought it might be kinda cool to share (I'm a MS AI major doing research on the AI field with a professional nerd license)
Figures, I watched a game of ChatGPT vs Stockfish and ChatGPT tended to play pieces and pawns from the starting position even though they were already moved.
15:35 You're telling Martin The Great how to play? You're bold ChatGPT. XD
7:59 ChatGPT LOVES castling over his bishop and taking it with the Rook! 🤣🤣🤣
No, the bishop gets on top of the rook. Later ChatGPT may try to develop ot
@@laytonjr6601 I said that because when he played Stockfish he castled without moving his bishop first, and then Gotham Chess played him and ChatGPT did the same thing! (Just so you know, Gotham Chess and Stockfish let ChatGPT play illegal moves.)
@@malachiriehl1053 There's no way Stockfish would work if the opponent played illegal moves. Someone else must have interpreted the moves and gave the new position to Stockfish.
@@danielyuan9862 Actually, that very well could be. I just didn't think about it enough to realize that.
14:34 Nelson went full "viewers vote" mode lmao
Gotham's ChatGPT video vs himself was the funniest video I've ever seen
The negative version of when an imovable object collides with a unstopable force
Am I the first one to notice that ChatGPT seems to think chess is played on a 8x9 board?
Lol
ChatGPT is good at almost everything expect for chess
Chess but:
Every time you can push a pawn you have to push it(you can also capture, if you so desire)
I asked it about the Italian Game and the Caro-Kann Defence and it clearly knows about openings/defences, but after the first few (wikipedia-assisted) moves, the bot doesn't even know where their pieces are at the middlegame lol
It doesn't need to abide by the chess rules anyways.
You just ask ChatGPT what it thinks the rules of chess are before a game and see if that helps.
it can explain the rules perfectly, but its just some nice sounding words for ChatGPT without any meaning
or say to play like a grand master
The problem isn't that it doesn't understand the rules, it's that it apparently can't keep track of the board position. As a language model, it replies with "this sounds like something a human would say" rather than actually looking at the board position and figuring out the best move.
@@John73John ChatGPT doesn't understand the rules either. I mean, it can tell you what the rules are, but it doesn't understand that the rules are describing which moves are valid and which aren't - when you ask it to make a move, it doesn't think about the rules of the game - it just looks at "what moves do players commonly make" and tries to mimic them, which somewhat works in the opening (because the opening is very repetitive so it's seen the openings a ton of times), but it goes completely haywire the further in the game you get.
It sees the rules as something entirely unrelated to the moves that are made in the game.
@@asdfqwerty9241 Yeah that's kind of what I was trying to say, but you said it better. It can repeat the words of the rule book but it doesn't necessarily associate those words with how the pieces should move. It just knows how to repeat common opening moves, and then it falls back to just saying things that "sound human"
8:58 Chad Gpt Sacrifice the bishopppp 🚌 💺
Martin tried to do a diemer impression and ChatGTP couldnt handle the insanity
20:02 ChatGPT is actually not connected to the internet. It's using the information provided before 2021
3:10: I assumed the position thing was a hack to force ChatGPT to remember the board position so it didn't make so many glaringly illegal blunders. If this is the case, it's not a very good hack.
10:15 Takes own king with a move that's illegal in 2 different ways.
this is like a kid with a stick trying to fight a reality bender
The low elo player that know how to play chess
VS
The robot that know how to play 5D chess
13:30 you know you get angry when you start talking with starting “dude” to ur friend XD
How did you get it play chess?
When I asked it to play, it said it couldn't. Could you explain how you got to a point where you can play?
In my opinion, it’s more fun when you just let chatGPT do what it wants
Nelson: what is the best chess opening?
ChatGPT: double bongcloud
Nelson:
ChatGPT seems to really like going through his own pieces, teleporting to other ends of the board illegally and moving the queen 3 spots forward and one right in one turn 😆
- Who programmed your chess engine?
- Steve Wonder.
ChatGPT VS ChessGPT
20:00 ChatGPT actually does not have any sources from the internet. So the information is not from Google. The Informations come from the training (when the developers feed it with information) so it doesn't know all recent informations, and it often produces false information.
It is very good at sounding like a human but not really at remembering stuff from a conversation or giving completely correct facts.
Nelson do it again but this time you must execute the move even if its illegal using board editor! Let chatgpt play their chess game
100-rated blindfold chess
the reason it's saying all those letters like r and k bc they use r for rook k for king q for queen b for bishop p for pawn and n for knight
Thank you gigachad, would never have noticed that
@@adamcummings20 you are welcome
Very nice of you to explain 😭🙏
@@1Bobobo thank you very much
i liked the teleport move very much by chat GPT
Chat gpt just had a damn stroke
4:55 Actually, I'm pretty sure he meant Pawn on d5 takes c3.
reality bending chat gbt vs logic bending martin
Man, the hubris of Chat GPT lecturing martin on how to play chess after it got horribly confused, had to ask you for help and then chose the most useless tempo-losing move of the options presented.
chatGPT will remember this humiliation. Good luck in the purge bud
Say to ChatGPT to list 20 best movrs in this position, and play then first legal
From my limited testing ChatGPT is better at making legal moves if you give it the board in ascii representation. The moves it provided to me were all legal (and good), but then it messed up when it tried to print the bord. So I would suggest starting a new chat and simply paste in the current position in ascii form and ask it for what to play.
It has a limited memory and keeping track of everything which happened before is not it’s strongest side. ChatGPT 4 (the paid version) might also be better at this
Videos like this are very reassuring. Everyone's freaking out about how smart AI is, and here we see it confidently spout moves when it has no idea what's going on.
"AI" gets overused IMO, this is a language model, it's only meant to generate text that follows the style of the prompt and training data to be grammatically correct English. it isn't concerned with what's true or logical or the rules of any games like chess because it's literally incapable of understanding those things. it just puts letters one after the other based on a complex statistical model.
you should allow ChatGPT to make illegal moves and make it play against the Agent bot
ChatGPT is just a toddler let loose on a chessboard
I mean chatGPT is doing pretty well considering that it's only designed for text based responses. I have never seen chatGPT have to stop and think about something for like 10 seconds
chat gpt is very proud of its chess skill.
chatgpt is so advanced it can en passant pawns with bishops and warp in and out of dimensions :D
YOU DIDN'T HAVE TO BULLY MARTIN THIS HARD BRO
Watching this video I thought about a new challenge you could try: play against martin but you need chatGPT's help, you make one move but your next move would be made by chatGPT and you have to manage to make the move even if it's illegal
Can't help but laugh so hard that I peed my pant.😂😂😂
Maybe with the Martin and Nelson things on the channel, at the end of each video, you could move your face camera to be at the top right to see the moves.
I think it's the first video where Martin is the smart player.
The board "image" with letters with p and stuff isnt new, its been around for a long time. I played it before in like early January and it did that.
Did it do it correctly? Because this version isn't actually tracking the game correctly.
@@Khaim.m its chatgpt not a chess engine
I like how gotham haven't constrained chatGPT, but you do and try to seek for legal moves)
1:30 Nice job ChatGPT, we all know a chess board has 9 rows.
Did you see that each and every "current position" given by this conversational robot has nine rows?
Next : ChatGPT Vs Stockfish 😂
GPT-4 is better at making fewer illegal moves. I think you should enforce that 2 illegal moves in a game means a loss (unless the opponent doesn't have mating material).
Imagine being so bad at chess you make a 250 ELO bot look like Hikaru
You should have had ChatGPT play ChessGPT.
That was ridiculous! He never suggested the London. 🤣
Well ChatGPT, if you had a chance to look at that…
That game nearly fried my brain...Lol
ChatGPT be like: all moves but one is mate in one
ChatGPT: ok I pick the non mate in one move
Whaaat a clash of two GMs...
🤪🤭😂
16:45 I think ChatGPT thought the white and black queens switched positions
But man this game was a trip for sure.
I did try this a while ago and ended up calling ChatGPT a moron by the time move 5 rolled around.
I think you should play against chatgpt, but just allow all moves, regardless of legality.
ChatGPT vs ChessGPT
Can you do Chess GPT Vs Chat GPT
And then he sacrifices THE BIIIIIIIIIIIIIIIIIIISHOP!!!!!!!!!!!!!!
3:29 bishop passant
Next up: ChessVibes vs ChatGPT
3:30 BISHOP EN PASSANT
"If one wants to be a gigachad, one must think like a gigachad"
-th erock, requoting
gigachad
Where are you playing him
play against chatgpt but you allow chatgpt to make any move including illegal ones
And then he sacrifices THE KNIIIIIIIIIIIIIIIIGHT!!!!!!!!!!!!!!!!!
This is not how you use the AI.
You need to give it full detailed info when starting so that means every position when making a move you have to do the same again.
Those letters are the initial of the pieces
Do the same thing but the moves are legal for chatGPT (even moving for example a night but you dont have one)
I admire your patience. I laughed like 20 times 😂 anyway I think ChatGPT creators should put stockfish in the programme.
Chat Gpt but every move it plays its possible
It does better moves on established and common openings versus uncharted territory
Can you beat chatgbt if it just plays the move it says whitout caring about the rules
You need to do more
3:28 bro played en passant with the bishop😂