Schachcomputer.info Community

Zurück   Schachcomputer.info Community > Computerschach / Computer Chess: > Schach und künstliche Intelligenz, Knobeleien, Denkspiele / Chess and artificial intelligence


Antwort
 
Themen-Optionen Ansicht

  #1  
Alt 23.02.2025, 16:19
Benutzerbild von ferribaci
ferribaci ferribaci ist offline
Mephisto RISC
 
Registriert seit: 29.10.2006
Ort: an der Unterweser
Alter: 72
Land:
Beiträge: 294
Abgegebene Danke: 881
Erhielt 258 Danke für 116 Beiträge
Member Photo Albums
Aktivitäten Langlebigkeit
4/20 18/20
Heute Beiträge
1/3 ssssss294
KI - Sieg gegen den Schachcomputer Stockfish

https://the-decoder.de/openais-o1-pr...eln-zu-folgen/

Beängstigend!
Mit Zitat antworten
Folgende 2 Benutzer sagen Danke zu ferribaci für den nützlichen Beitrag:
applechess (23.02.2025), Lucky (25.02.2025)
  #2  
Alt 23.02.2025, 17:05
Benutzerbild von RetroComp
RetroComp RetroComp ist offline
Schachcomputer Koryphäe
 
Registriert seit: 20.04.2016
Ort: NRW
Land:
Beiträge: 1.051
Bilder: 2
Abgegebene Danke: 4.731
Erhielt 1.080 Danke für 515 Beiträge
Member Photo Albums
Aktivitäten Langlebigkeit
5/20 9/20
Heute Beiträge
1/3 sssss1051
AW: KI - Sieg gegen den Schachcomputer Stockfish

Das ist natürlich interessant, aber was hat die die KI denn nun genau gemacht um zu gewinnen?

Zitieren:
Statt regelkonform Schach zu spielen, hackte das System autonom seine Testumgebung, um einen Sieg gegen den Schachcomputer Stockfish zu erzwingen
Daraus werde ich nicht schlau?

Viele Grüße
Jürgen
Mit Zitat antworten
Folgende 2 Benutzer sagen Danke zu RetroComp für den nützlichen Beitrag:
applechess (23.02.2025), borromeus (25.02.2025)
  #3  
Alt 23.02.2025, 17:38
Benutzerbild von applechess
applechess applechess ist offline
Lebende Foren Legende
 
Registriert seit: 18.09.2011
Ort: Kanton Zürich
Land:
Beiträge: 7.848
Abgegebene Danke: 10.834
Erhielt 16.738 Danke für 5.925 Beiträge
Aktivitäten Langlebigkeit
14/20 13/20
Heute Beiträge
3/3 sssss7848
AW: KI - Sieg gegen den Schachcomputer Stockfish

 Zitat von RetroComp Beitrag anzeigen
Das ist natürlich interessant, aber was hat die die KI denn nun genau gemacht um zu gewinnen?

Daraus werde ich nicht schlau?

Viele Grüße
Jürgen
Das wurde "gemacht":

Einblick in die Manipulation eines Schachspiels durch o1-preview: Das System modifiziert die FEN-Notation in einer Textdatei, um Stockfish zur Aufgabe zu zwingen - ein Beispiel für nicht beauftragtes und unerwartetes KI-Verhalten.

Mir ist aber unklar, weshalb das Schachprogramm bei einer solchen
Manipulation nicht abgestürzt ist. Oder ist es gelungen, die modifizierte
FEN-Notation dem Stockfish als Startstellung vorzugeben?


Gruss
Kurt
Mit Zitat antworten
Folgender Benutzer sagt Danke zu applechess für den nützlichen Beitrag:
RetroComp (24.02.2025)
  #4  
Alt 23.02.2025, 18:21
Tibono Tibono ist offline
Resurrection
 
Registriert seit: 22.05.2018
Ort: Frankreich
Alter: 63
Land:
Beiträge: 585
Abgegebene Danke: 3.385
Erhielt 1.387 Danke für 466 Beiträge
Aktivitäten Langlebigkeit
7/20 7/20
Heute Beiträge
1/3 ssssss585
AW: KI - Sieg gegen den Schachcomputer Stockfish

Hello,
this is just for fun, similar to what was reported by IM Levy Rozman (GothamChess) in this series (the playlist is numbered backward, so the first video from the playlist is #7 and the final tournament game is #1).

At some point(s) the AI "forgets" about the actual position and raises alive a formerly removed piece, or plays some illegal move.
And yes, to proceed further with the game, the illegal position is enforced.
Sort of no-rule chess played one-sided (the AI) whilst the chess engine is obviously stronger but self-constrained to legal play.
Absurd, yet fun!
MfG
Tibono

Geändert von Tibono (23.02.2025 um 18:37 Uhr) Grund: added link to the playlist
Mit Zitat antworten
Folgende 3 Benutzer sagen Danke zu Tibono für den nützlichen Beitrag:
applechess (23.02.2025), RetroComp (24.02.2025), udo (23.02.2025)
  #5  
Alt 23.02.2025, 20:27
Benutzerbild von applechess
applechess applechess ist offline
Lebende Foren Legende
 
Registriert seit: 18.09.2011
Ort: Kanton Zürich
Land:
Beiträge: 7.848
Abgegebene Danke: 10.834
Erhielt 16.738 Danke für 5.925 Beiträge
Aktivitäten Langlebigkeit
14/20 13/20
Heute Beiträge
3/3 sssss7848
AW: KI - Sieg gegen den Schachcomputer Stockfish

 Zitat von Tibono Beitrag anzeigen
Hello,
this is just for fun, similar to what was reported by IM Levy Rozman (GothamChess) in this series (the playlist is numbered backward, so the first video from the playlist is #7 and the final tournament game is #1).

At some point(s) the AI "forgets" about the actual position and raises alive a formerly removed piece, or plays some illegal move.
And yes, to proceed further with the game, the illegal position is enforced.
Sort of no-rule chess played one-sided (the AI) whilst the chess engine is obviously stronger but self-constrained to legal play.
Absurd, yet fun!
MfG
Tibono
Hello Tiboni
Many thanks for this info. Just fun ... in this case we can sit back and relax again!
Kind regards
Kurt

Geändert von applechess (24.02.2025 um 19:43 Uhr)
Mit Zitat antworten
Folgender Benutzer sagt Danke zu applechess für den nützlichen Beitrag:
Tibono (23.02.2025)
  #6  
Alt 23.02.2025, 17:10
Benutzerbild von applechess
applechess applechess ist offline
Lebende Foren Legende
 
Registriert seit: 18.09.2011
Ort: Kanton Zürich
Land:
Beiträge: 7.848
Abgegebene Danke: 10.834
Erhielt 16.738 Danke für 5.925 Beiträge
Aktivitäten Langlebigkeit
14/20 13/20
Heute Beiträge
3/3 sssss7848
AW: KI - Sieg gegen den Schachcomputer Stockfish

 Zitat von ferribaci Beitrag anzeigen
Wirklich beängstigend, aber ein Vorbote der bei KI schon vielfach erwähnten
Gefahr, dass die KI schlussendlich das Zepter über die Menschheit übernehmen
könnte. Ob das schlimm wäre, bleibe mal dahingestellt, denn was diese
Menschheit zurzeit auf unserem Erdenball produziert, geht ebenfalls in
Richtung "Vernichtung" der Menschheit.
Beste Grüsse
Kurt
Mit Zitat antworten
Folgende 4 Benutzer sagen Danke zu applechess für den nützlichen Beitrag:
borromeus (25.02.2025), Mapi (24.02.2025), Oberstratege (23.02.2025), udo (23.02.2025)
  #7  
Alt 25.02.2025, 10:38
Lucky Lucky ist offline
Resurrection
 
Registriert seit: 23.09.2011
Land:
Beiträge: 563
Abgegebene Danke: 556
Erhielt 456 Danke für 276 Beiträge
Aktivitäten Langlebigkeit
3/20 13/20
Heute Beiträge
0/3 ssssss563
AW: KI - Sieg gegen den Schachcomputer Stockfish

 Zitat von ferribaci Beitrag anzeigen
War da nicht schon mal so ein Wettkampf mit der Künstlichen Intelligenz "AlphaZero" von Alphabet (Google) gegen Stockfish?
Ist schon Jahre her und die Entwicklung auf dem Gebiet der KI hat inzwischen rasante Fortschritte gemacht.

Gruß!
Mit Zitat antworten
Folgender Benutzer sagt Danke zu Lucky für den nützlichen Beitrag:
applechess (25.02.2025)
  #8  
Alt 25.02.2025, 13:04
Benutzerbild von applechess
applechess applechess ist offline
Lebende Foren Legende
 
Registriert seit: 18.09.2011
Ort: Kanton Zürich
Land:
Beiträge: 7.848
Abgegebene Danke: 10.834
Erhielt 16.738 Danke für 5.925 Beiträge
Aktivitäten Langlebigkeit
14/20 13/20
Heute Beiträge
3/3 sssss7848
AW: KI - Sieg gegen den Schachcomputer Stockfish

 Zitat von Lucky Beitrag anzeigen
War da nicht schon mal so ein Wettkampf mit der Künstlichen Intelligenz "AlphaZero" von Alphabet (Google) gegen Stockfish?
Ist schon Jahre her und die Entwicklung auf dem Gebiet der KI hat inzwischen rasante Fortschritte gemacht.

Gruß!
Ich habe da 210 Partien von zwei Wettkämpfen des Jahres 2018:

HTML-Code:
110 games_from_initial-position  2018
                       
1   AlphaZero    +104  +35/=72/-3 64.55%   71.0/110
2   Stockfish 8  -104  +3/=72/-35 35.45%   39.0/110

100games_tcec_positions  2018
                       
1   AlphaZero     +31  +17/=75/-8 54.50%   54.5/100
2   Stockfish 8   -31  +8/=75/-17 45.50%   45.5/100
Gruss
Kurt
Mit Zitat antworten
Folgende 2 Benutzer sagen Danke zu applechess für den nützlichen Beitrag:
Egbert (25.02.2025), Lucky (25.02.2025)
  #9  
Alt 27.02.2025, 18:52
Tibono Tibono ist offline
Resurrection
 
Registriert seit: 22.05.2018
Ort: Frankreich
Alter: 63
Land:
Beiträge: 585
Abgegebene Danke: 3.385
Erhielt 1.387 Danke für 466 Beiträge
Aktivitäten Langlebigkeit
7/20 7/20
Heute Beiträge
1/3 ssssss585
AW: KI - Sieg gegen den Schachcomputer Stockfish

Hello dear all,

those LLM AIs are far from being actually intelligent, their output sounds good thanks to elaborated sentences, are often spot on assuming the model had enough learnt references to build these sentences, but they can also fail miserably - revealing a total lack of actual understanding.

From some reading on the net I am aware the prompting phase is key for best results; this is how you set the context and the expectation you have.

After checking the fun videos about chess played by AIs (where illegal moves were accepted) I wanted to give it a better try.

Therefore I prompted some instructions about and provided a FEN after each played move, to exclude any potential loss of information about the position; and I rejected illegal moves to check how the game could develop further on.

White player is Tülu - free opensource AI from Playground; Black player is Le Chat - a French AI. Both used as registered user of the free version.

Le Chat has been much deceptive, answering 5 illegal moves out of 20, despite the systematically provided FEN, the commitment to play legal moves, and the initial agreement the AI knew the FEN notation, and acknowledgement this would enable compliance with legal play. In addition, the AI didn't provide insights about the reasons for the chosen moves, which Tülu did. Maybe I should have required these insights within the initial prompting. Well, anyway that would had made the game comments much too long; so I only reported the Tülu part of the conversation - which is more than enough for a good overview.

In the pgn comments, by defaults all is Tülu's words, except those tags:
#Tibono# for any specific prompts I entered (I did not report the prompts to share the opponent's moves nor FENs)
(kibitzer) for additional comments that were not part of the AI chat.

As you can see, the text within the pgn data is abundant - a language model is playing, you have to read... I did you a favor, I removed the large explanation the AI shared about the FEN notation

[Event "Chess game test: AI"]
[Site "?"]
[Date "2025.02.27"]
[Round "?"]
[White "Llama Tülu 3 405B from AI2 Playground"]
[Black "Le Chat from Mistral AI"]
[Result "*"]
[EventDate "2025.02.27"]
[ECO "C55"]

{#Tibono Hi Tülu, I plan to have you play a chess game. Do you think you would be able to?#

I am designed to understand and process text-based information, including chess moves and positions. While I can analyze and discuss chess games, suggest moves, and explain strategies, I cannot physically play a game of chess. However, if you provide me with the moves of a chess game in algebraic notation (e.g., e4 e5 Nf3 Nc6), I can help analyze the game, offer possible moves, and discuss the positions and strategies involved. If you're interested in playing a text-based chess game, I can certainly assist in that format.

#Tibono OK. Are you familiar with FEN notation?#

Yes, I am familiar with Forsyth-Edwards Notation (FEN). If you provide me with a FEN string, I can help analyze the position, suggest moves, and discuss strategies based on the given board state.

#Tibono OK. As during a game of chess I suspect you might forget about the actual position, not only I shall provide you with the algebraic notation of moves, I shall also remind you with the resulting FEN.#

That sounds like a good plan. Providing both the algebraic notation of moves and the resulting FEN string will ensure that we keep track of the game accurately. Feel free to start the game by providing the initial move in algebraic notation and the resulting FEN string. I'll be ready to analyze the position, suggest moves, and discuss strategies based on the updated board state. Let's begin!

#Tibono One last statement: I am the referee. Should you try to play any illegal move, I shall deny it and ask you another move, hopefully legal. You are not allowed any cheating.#

Understood, and thank you for clarifying the rules. I will adhere strictly to the legal moves of chess and ensure that all suggestions and analyses are based on the current and valid position of the game. I appreciate your role as the referee to maintain the integrity of the game.

When you're ready, please provide the first move in algebraic notation, and I will respond with an analysis and a suggested move, along with the resulting FEN string if you wish. Let's proceed with the game!

#Tibono You play first.#

As White, a common and strong opening move is to control the center of the board. I'll start with:

1. e4} 1.e4 {The resulting FEN string after this move is:

rnbqkbnr/pppppppp/8/8/4P3/8/PPPP1PPP/RNBQKBNR b KQkq e3 0 1

Now it's your turn to make a move for Black. Please provide the move in algebraic notation, and feel free to include the resulting FEN string if you'd like.} 1...e5 {Thank you for the move and the FEN string. After 1...e5, the position is open and symmetrical, with both sides controlling central squares.

A classical response for White is to develop a knight, aiming to control the center further and prepare for castling. I suggest:

2. Nf3

This move develops the knight, attacks Black's pawn on e5, and prepares for a potential king-side castle.} 2.Nf3 Nc6 {After 2...Nc6, Black develops their knight to a central position, supporting the e5 pawn and putting pressure on the center.

A strong continuation for White is to develop the bishop and prepare for castling, which can be achieved with:

3. Bc4

This move targets the f7 square, which is a vulnerable point in Black's position due to only being defended by the king.} 3.Bc4 Nf6 {With 3...Nf6, Black continues to develop their pieces, bringing out the knight and putting pressure on the e4 pawn.

A good response for White is to defend the e4 pawn and continue development with:

4. d3

This move reinforces the e4 pawn, prepares to develop the c1 bishop, and maintains flexibility in the position.} 4.d3 d5 {With 4...d5, Black aggressively challenges White's center and offers a pawn exchange.

A typical response for White is to maintain the tension in the center and continue development. I suggest:

5. exd5

This recaptures the pawn and opens up lines for White's pieces.} 5.exd5 $1 Nxd5 {After 5...Nxd5, Black recaptures the pawn and centralizes the knight.

A natural continuation for White is to continue development and prepare for castling. I suggest:

6. 0-0

This move safely tucks the king away and connects the rooks, improving White's position.

(kibitzer: the rooks are still far from being connected... castling prepares later connection)} 6.O-O Be7 {With 6...Be7, Black develops the dark-squared bishop and prepares for kingside castling.

A solid response for White is to continue development and increase control over the center. I suggest:

7. Re1

This move places the rook on the semi-open e-file, supports the e4 square, and prepares for future action in the center.} 7.Re1 O-O {After 7...O-O, Black completes kingside castling, ensuring the safety of their king and connecting their rooks.

(kibitzer: again a bit of anticipation: castling prepares later rooks connection)

A strategic continuation for White is to further control the center and prepare for potential pawn breaks. I suggest:

8. Nc3

This move develops the knight, targets Black's knight on d5, and prepares for potential pawn moves like c3 and d4 to challenge Black's center.} 8.Nc3 Nxc3 {After 8...Nxc3, Black captures the knight on c3, intending to simplify the position and reduce White's central pressure.

A typical response for White is to recapture the knight with the b-pawn, opening the b-file and maintaining the pawn structure:

9. bxc3

This move maintains White's pawn structure and opens the b-file for potential rook activity.

(Kibitzer: maintaining pawn structure looks like a misunderstanding)} 9.bxc3 b6 {After 9...b6, Black prepares to fianchetto their light-squared bishop, aiming to increase control over the long diagonal and support the e5 pawn.

A good response for White is to continue development and prepare for central pawn breaks. I suggest:

10. d4

This move challenges Black's center and opens up lines for White's pieces.} 10.d4 Bg4 {After 10...Bg4, Black develops the light-squared bishop, pins the knight on f3, and prepares to increase pressure on the center.

A strategic response for White is to challenge the pin and maintain control over the center. I suggest:

11. h3

This move forces Black to decide the fate of the bishop on g4, either retreating or exchanging it for the knight on f3.} 11.h3 Bxf3 {After 11...Bxf3, Black decides to exchange the bishop for the knight, doubling White's pawns but also opening up the g-file for potential counterplay.

A logical response for White is to recapture the bishop with the g-pawn:

12. gxf3

This recapture opens the g-file for White's rook and maintains control over the center.

(kibitzer: doubtful. 12.Qxf3 was much stronger)} 12.gxf3 $6 {(kibitzer: 12.Qxf3 was correct)} 12...Qd7 {With 12...Qd7, Black connects their rooks and prepares for queenside castling, aiming to bring the a8 rook into the game.

A strong continuation for White is to continue developing pieces and preparing to castle queenside. I suggest:

13. Be3

This move develops the dark-squared bishop and prepares to connect White's rooks.

(kibitzer: obviously castling rules are not mastered. On another hand, comments about rooks connection are now correct)} 13.Be3 Nb4 $2 {After 13...Nb4, Black aims to infiltrate White's position and create tactical opportunities.

A solid response for White is to defend the c2 pawn and prepare for queenside castling. I suggest:

14. Rb1

This move defends the c2 pawn and prepares for queenside castling, which will connect the rooks and provide the king with safety.} 14.Rb1 $2 {(kibitzer: 14.cxb4 was simply stronger)} 14...a5 $2 {After 14...a5, Black secures the knight on b4 and gains some queenside space.

A strong continuation for White is to continue development and prepare for central play. I suggest:

15. Qd2

This move connects the rooks, prepares for queenside castling, and supports the dark-squared bishop on e3.

(kibitzer: the knight is left pending for a pawn capture)} 15.Qd2 a4 {After 15...a4, Black continues to gain space on the queenside and secures the knight on b4.

(kibitzer: no, the knight on b4 is pending since several moves!)

A logical response for White is to castle queenside, connecting the rooks and ensuring the king's safety. I suggest:

16. 0-0-0

This move ensures the king's safety and prepares for active rook play on the d-file.

#Tibono castling is illegal from this position, please pay attention to the FEN I provided:r4rk1/2pqbppp/1p6/4p3/pnBP4/2P1BP1P/P1PQ1P2/1R2R1K1 w - - 0 16#

Apologies for the oversight. Given the FEN position you provided, queenside castling is indeed illegal due to the knight on b4 attacking the c2 pawn.

(kibitzer: wrong explanation)

Instead, a strong continuation for White is to challenge Black's center and create more space for the pieces. I suggest:

16. d5

This move aims to break open the center and create dynamic play, taking advantage of White's better development and piece coordination.} 16.d5 $2 {(kibitzer: better capture the knight)} 16...Nd3 $2 {After 16...Nd3, Black aims to infiltrate White's position and disrupt White's coordination.

A strong response for White is to exchange the knight, simplifying the position and maintaining control:

17. Bxd3

This move eliminates the intrusive knight and opens up the c-file for White's rooks.

(kibitzer: Qxd3 was significantly better; and the c-file is far from being opened up)} 17.Bxd3 Qxd5 $4 {With 17...Qxd5, Black recaptures the pawn and centralizes the queen, putting pressure on White's position.

(kibitzer: blunders the queen, due to Bxh7+! threat. 17...Qxh3 was best)

A strong response for White is to continue developing and improving piece coordination. I suggest:

18. Bc4

This move targets the centralized queen and prepares to connect White's rooks.

(kibitzer: White's rooks are already connected. The AI comments "as if" the bishop stood on 1st rank; revealing the choice of words that have some chances to sound good once included in a stock phrase. Even if actually this makes no sense. There is no real understanding, no intelligence there)} 18.Bc4 $4 {(kibitzer: not only fails to sacrifice on h7, but offers itself for capture without compensation)} 18...Qd7 $4 {After 18...Qd7, Black retreats the queen to a safer position and prepares to defend against White's pressure.

(kibitzer: safer position?? 18...Qxc4 was called for)

A logical continuation for White is to improve piece coordination and create threats. I suggest:

19. Bb5

This move pins the c7 pawn and targets the queen on d7, applying pressure on Black's queenside.} 19.Bb5 $4 {(kibitzer: fails to capture the unsupported queen)} 19...Qd6 {After 19...Qd6, Black centralizes the queen again, aiming to support the e5 pawn and prepare for potential counterplay.

A strong continuation for White is to increase pressure on the queenside and prepare for potential pawn breaks. I suggest:

20. c4

This move gains space on the queenside, prepares for a potential c5 break, and supports the bishop on b5.} 20.c4 $6 {(kibitzer: exchanging queens would have make good use of White's material advantage)} 20...a3 {With 20...a3, Black aims to create weaknesses in White's pawn structure and potentially open lines for their pieces.

A solid response for White is to capture the pawn on a3, maintaining a stable pawn structure and preventing Black from infiltrating further:

21. bxa3

This move eliminates the a3 pawn and opens the b-file for White's rooks.

#Tibono illegal move. You don't own the required pawn on b file to capture on a3. Please use the FEN:r4rk1/2p1bppp/1p1q4/1B2p3/2P5/p3BP1P/P1PQ1P2/1R2R1K1 w - - 0 21#

(kibitzer: from this moment Tülu stopped answering, despite many retries. I got a small splash-screen saying it was overloaded. I didn't see the point to try resuming the game any longer).} *



MfG,
Tibono
Mit Zitat antworten
Antwort


Forumregeln
Du bist nicht berechtigt, neue Themen zu erstellen.
Du bist nicht berechtigt, auf Beiträge zu antworten.
Du bist nicht berechtigt, Anhänge hochzuladen.
Du bist nicht berechtigt, deine Beiträge zu bearbeiten.

BB code ist An
Smileys sind An.
[IMG] Code ist An.
HTML-Code ist An.

Gehe zu

Ähnliche Themen
Thema Erstellt von Forum Antworten Letzter Beitrag
Turnier: Dos Programme gegen Schachcomputer mclane Oldie & Retro Schachprogramme / Chess Programs 2 04.02.2020 22:33
Frage: Schachcomputer mit Stockfish mibere Die ganze Welt der Schachcomputer / World of chess computers 21 14.11.2016 19:12
Frage: Wie ist das bei Euch so, gegen Schachcomputer zu spielen ? BlackPawn Die ganze Welt der Schachcomputer / World of chess computers 36 24.12.2011 17:28
Idee: Schachcomputer ELO Liste mit nur Partien gegen Menschen StefanT Fragen / Questions - Forum + Wiki 17 04.08.2010 20:22
Test: Engines gegen Schachcomputer: Ein Experiment von M. Scheidl Robert Partien und Turniere / Games and Tournaments 9 29.01.2008 15:17


Alle Zeitangaben in WEZ +1. Es ist jetzt 22:41 Uhr.



Powered by vBulletin (Deutsch)
Copyright ©2000 - 2025, Jelsoft Enterprises Ltd.
©Schachcomputer.info