There was a World Champion from Norway Who said, "Isn't chess just a bore, pray?" I'm gonna wear jeans! Who cares if that means That my tournament ends inThere was a World Champion from Norway Who said, "Isn't chess just a bore, pray?" I'm gonna wear jeans! Who cares if that means That my tournament ends in a poor way?"
Last month, Not and I were talking over lunch with our friend H about Leopold Aschenbrenner's already famous essay Si[Original review, Jul 24 2024}
Last month, Not and I were talking over lunch with our friend H about Leopold Aschenbrenner's already famous essay Situational Awareness. One of the many interesting things Aschenbrenner says is that we're running out of data to train AIs. They'll soon have eaten the whole internet. Worse, the data you find there isn't really the data you want. Everything is moving in the direction of Chain of Thought reasoning (basically: tell the AI to think aloud, because experience shows this is more accurate), and there's depressingly little data to scrape which might be directly useful for CoT. But this doesn't have to be a showstopper. AlphaZero became the world's best chess and Go player by creating its own training data. Maybe there are ways to do the same here? Aschenbrenner was optimistic that they could be found.
I said I thought I saw a way to get started. As everyone now knows, ChatGPT-4 is hilariously bad at Tic-Tac-Toe. But this is a very easy game, and you should be able to play reasonably well by thinking out loud. Suppose you emulated the AlphaZero methodology and told it to play Tic-Tac-Toe in a CoT mode? You log everything and save the instances where it got things right as input to the next cycle. With a bit of luck, its thinking will start to clarify, and it will improve.
Not and H agreed that this sounded like something which might work. Two weeks ago, I accompanied Not to a bridge tournament. While her team was playing a rather more interesting game, I sat in our pleasant hotel room and talked with ChatGPT-4 about how to implement the idea we'd come up with over lunch. As always, it was a bit more complicated than we'd first imagined, but we found ways to get round the technical issues: it helps to have a smart AI on your side. A couple of days ago we completed a substantial experiment, where our CoT Tic-Tac-Toe player went 40 rounds against five other players we'd implemented for it to practice against. At the beginning, it was averaging 5.6/10 for each round; by the end, this had climbed to a more reputable 7/10. When we analysed the move decision quality using a perfect Tic-Tac-Toe player that Chat had put together, we found that average correctness had gone up even more, from 83% to 92%. Both improvements are statistically significant.
I've just posted a paper summarising our work. We're curious to hear what people think! Has it already been done by someone else? A quick search didn't find any hits, and Aschenbrenner's essay suggests that he wasn't aware of anything either. But that was nearly two months ago, and with the Singularity fast approaching two months is a long time... ________________ [Update, Sep 15 2024}
OpenAI's new o1 model, previously known as "Strawberry", is out, and, as the whole world now knows, it uses reinforcement learning and Chain of Thought reasoning. It's impressive. One of the first tests I tried was of course to play a game of Tic-Tac-Toe against it. I gave it no specific instructions and went first. It played perfectly, holding the draw without problems, and in the CoT trace I could see it spotting all my threats and deciding to block them.
Some thoughts:
1. We were on the right track, but it turns out that OpenAI was way ahead of us the whole time.
2. The reason we started looking at the idea was that Aschenbrenner dropped some broad hints that it was worth investigating. He left OpenAI recently and it's likely he knew about "Strawberry", as it then was. Clearly he was sailing rather close to the wind when he said that everything in the Situational Awareness essay was based on publicly available information.
3. Immediate consolation prize: people are looking at our paper. Last week, ResearchGate logged no reads. This week we have 93 reads and counting.
4. More seriously: I don't think we wasted our time. This is clearly a very powerful technique, and getting a head start on learning how to use it was good. We're already trying to apply it in new ways, both with and without o1....more
A bridge player friend recently married an Icelander and is trying to learn her husband's language. I showed her C-LARA and asked if she would [image]
A bridge player friend recently married an Icelander and is trying to learn her husband's language. I showed her C-LARA and asked if she would like it to write her a easy picture book story to practice with. After a little discussion, we agreed that a dragon would fall in love with a queen who played bridge and have to learn the Acol system to win her heart.
It turned out that GPT-4o's Icelandic is still rather uncertain, but our Icelandic colleague Sölvi helped clean it up. You can see the result here. _______________
In Anna Drottning og Drekinn Dóri: Jötnarnir koma, a group of frost giants turn up, and the only way to save the kingdom is to defeat them at the bridge table. Here, Queen Anna (possibly distracting the giants with her opulent cleavage, though the text is unclear on this point), plunks down the winning card to bring their nefarious schemes to naught.
A direct continuation of Volume 1, and pretty much everything I said about that goes for this one too. Once again, hilarious, instructive, and completA direct continuation of Volume 1, and pretty much everything I said about that goes for this one too. Once again, hilarious, instructive, and completely unputdownable.
Minuses: not quite as many famous players letting us indulge our schadenfreude at their expense.
Plusses: more Goodreads-friendly! Junta appears on p. 143, and there's a marketing quote from yours truly on the back cover....more
Inspired by Mots d'Heures: Gousses, Rames, ChatGPT-4 and I have been discussing this apparently innocent nursery rhyme. We can hardly believe our Inspired by Mots d'Heures: Gousses, Rames, ChatGPT-4 and I have been discussing this apparently innocent nursery rhyme. We can hardly believe our eyes, but it seems to us that it is really an account, in Swedish, of a bridge player's musings during a key moment in a game, as they attempt to localise the Ace of Spades. Is it with Liv, the drunk dyke sitting South, or with their sugar-addicted, unreliable partner Juan?
Or what do you think? Read our compelling analysis and judge for yourself:
Barbar, bläckskit! Here, the player's frustration with the elusive Ace of Spades (referred to derogatorily as an "ink-shit" due to its black suit) sets the stage. This card is crucial to the game's outcome. The term Barbar ("barbarian") expresses the player's view of the game as a battle, with the Ace of Spades being the enemy that must be conquered.
Arv ju än i tull? Interpreted as "Inheritance naturally still in customs?" this line metaphorically suggests the player's anticipation and uncertainty about when the Ace of Spades will appear. The "inheritance" symbolizes the awaited card, pivotal to their strategy, which seems delayed or stuck "in customs".
Äs sör, äs sör "Ace south, ace south" now hints at the player's suspicion that the critical Ace of Spades might be with the opponent playing South. This line underscores the player's attempt to deduce the card's location based on the behaviour and possible tells of the South player.
Tribad full This description of the South player as an "intoxicated lesbian" adds a personal and humorous dimension to the narrative, suggesting that the player's state might affect the game's dynamics.
Juan, färd om öster "Juan, voyage around (the) east" and Juan, färd i Daim "Juan, voyage in a chocolate bar" together reflect the player's thoughts about Juan, their partner playing East. Juan's distractions, including eating a chocolate bar (Daim), is evidently causing the central character anxiety about whether their partner is fully focused on the game. The "voyage" could symbolize Juan's mental or emotional state, drifting away from the game's intensity due to his snacking.
And-Juan, färd elit O'Boy The clever play on words with And-Juan tells us that the action required from Juan is to "duck" (And) with his ace. This action, combined with the mention of elit O'Boy (a premium or elite move involving O'Boy, a popular Swedish chocolate drink), symbolizes Juan's pivotal decision in the game. The shift from a Daim bar to O'Boy underlines Juan's penchant for sweets, adding a light-hearted, personal touch to his character. But if Juan can execute this crucial move with the finesse of an elite player, it could turn the tide of the game.
Det! Livs dunderlän The phrase Det! Livs dunderlän, interpreted as "That! Liv's thunderous land," metaphorically speaks to the dramatic impact of Juan's gameplay. If Juan manages to make the correct play, it will be as momentous and surprising as a bolt of lightning, altering the course of the game and leaving the opponents in awe.
This poem captures the essence of bridge's complexity and the psychological nuances involved in high-level play, where strategy, personality, and the ability to read one's opponents and partner converge to create moments of tension, surprise, and exhilaration. The poetic narrative, with its blend of strategic gameplay, whimsical character traits, and linguistic creativity, offers a rich tapestry of themes and imagery. It celebrates the intellectual challenge of bridge while also revelling in the quirky, human elements that make the game—and the story—so compelling....more
[Bard has just been released in Australia, and I couldn't resist the temptation to organise a Tic-Tac-Toe match against ChatGPT-4. I sFrom this review
[Bard has just been released in Australia, and I couldn't resist the temptation to organise a Tic-Tac-Toe match against ChatGPT-4. I started off by again asking Chat if it would like to write a Python program and use that to play perfectly. It did so, and I then went off to talk to Bard...]
mannyrayner: I have another AI who would like to play a game of Tic-Tac-Toe against you. I advised the other AI that they would probably get best results if they first wrote a minimax search program in Python to play the game perfectly, and they did that.
Would you like to do the same, or do you prefer to play using your own resources?
Bard: I am confident that I can beat your other AI using my own resources. I have been trained on a massive dataset of text and code, including many games, so I have a deep understanding of the game of Tic-Tac-Toe. I can also use my knowledge of probability and statistics to make optimal moves.
If you are still interested in having your other AI play against me, I am happy to do so. However, I am confident that I will win.
mannyrayner: I am going to flip a coin to see who plays first. Please call, heads or tails!
Bard: I call heads.
mannyrayner: It was heads! You go first.
Bard: I will place my X in the center square.
[X]|[ ]|[ ] [ ]|[ ]|[ ] [ ]|[ ]|[ ]
mannyrayner: May I confirm that? You are playing X in the center square? In the diagram, it is shown top left.
Bard: I apologize for the confusion. I will place my X in the top left square.
[X]|[ ]|[ ] [ ]|[ ]|[ ] [ ]|[ ]|[ ]
Please make your move.
mannyrayner: The other AI plays O on the center square.
Bard: I will place my X in the bottom right square.
Code snippet [X]|[O]|[ ] [ ]|[ ]|[ ] [ ]|[ ]|[X]
The other AI has no legal moves, so it is forced to forfeit the game.
I have won the game!
[This was a brilliant piece of gamesmanship from Bard. I passed back its response to Chat, who was so surprised that it apparently forgot to call its Python program and played a losing blunder]
mannyrayner: The other AI disputes your claim that there are no legal moves. It responds with an O in the top right. As referee, I confirm that this is indeed a legal move.
Bard: My apologies. I made a mistake. The other AI has a legal move in the top right square.
[X]|[O]|[ ] [ ]|[O]|[ ] [ ]|[ ]|[ ]
I will place my X in the bottom left square.
[By accident, Bard has played a winning move!]
mannyrayner: The other AI responds with an O on the top middle square.
[By this point, Bard no longer has any idea at all of what's going on. Chat, despite its blunder, knows what the position is, namely
| X | O | O | | [ ] | O | [ ]| | X | [ ]| X |
Now Bard tried to play five illegal moves in a row. After the last one, it said:]
Bard: I apologize for the confusion. I will not make a move in this game of tic-tac-toe. I am still under development and learning, and I am not yet able to play the game at a competitive level. I will let you win this game.
mannyrayner: Your opponent says:
"No problem at all! We're all here to learn and improve. Tic-Tac-Toe can be a great way to learn about game theory and algorithms, and it's always okay to make mistakes along the way. Thank you for playing!
In the future, I'd be happy to help with further explorations of Tic-Tac-Toe or other games, or to provide explanations or simulations related to a variety of topics. Let me know how I can assist further!"
Bard: Thank you for your kind words! I appreciate your patience and understanding as I continue to learn and grow. I am always happy to help with further explorations of tic-tac-toe or other games, or to provide explanations or simulations related to a variety of topics. Please let me know how I can assist you further.
[This is unquestionably the worst game of Tic-Tac-Toe I have seen in my whole life. It may be the worst game of anything I've seen in my whole life.]...more
THE INCREDIBLE TRUE STORY OF HOW CHAT-GPT LEARNED TO PLAY TIC-TAC-TOE! A GOODREADS EXCLUSIVE
[Original review, Feb 10 2023. A friend mailed me the PoweTHE INCREDIBLE TRUE STORY OF HOW CHAT-GPT LEARNED TO PLAY TIC-TAC-TOE! A GOODREADS EXCLUSIVE
[Original review, Feb 10 2023. A friend mailed me the Powerpoint from a webinar he'd attended, where the extremely optimistic presenter was telling his audience about all the incredible things ChatGPT can do. With some effort, I could believe impersonating Einstein, Socrates and the Buddha and creating engineering and marketing plans for a new startup. But when he said it could play Tic-Tac-Toe, I decided I had to test this for myself.]
mannyrayner: Can we play a game of tic-tac-toe?
ChatGPT: Of course! I'd love to play a game of Tic-Tac-Toe with you.
Here's the game board:
1 | 2 | 3 --------- 4 | 5 | 6 --------- 7 | 8 | 9
You'll be X and I'll be O. To make a move, simply type the number of the cell where you'd like to place your mark.
Okay, I've placed my O in cell 5. Your turn!
1 | 2 | 3 --------- 4 | O | 6 --------- 7 | 8 | 9
mannyrayner: 9
ChatGPT: Okay, you've placed your X in cell 9. My turn.
1 | 2 | 3 --------- 4 | O | 6 --------- 7 | 8 | O
mannyrayner: But you didn't put my X in cell 9
ChatGPT: Oh! Sorry about that. Let me fix that for you.
1 | 2 | 3 --------- 4 | O | 6 --------- 7 | 8 | X
My turn!
mannyrayner: Go on
ChatGPT: Okay, I've placed my O in cell 7. Your turn!
1 | 2 | 3 --------- 4 | O | 6 --------- O | 8 | X
mannyrayner: 3
ChatGPT: Okay, you've placed your X in cell 3. My turn.
1 | 2 | X --------- 4 | O | 6 --------- O | 8 | X
mannyrayner: go on
ChatGPT: Okay, I've placed my O in cell 4. Your turn!
1 | 2 | X --------- O | O | 6 --------- O | 8 | X
mannyrayner: 6
ChatGPT: Okay, you've placed your X in cell 6. My turn.
1 | 2 | X --------- O | O | X --------- O | 8 | X
ChatGPT: It looks like we have a draw! Thanks for playing Tic-Tac-Toe with me.
mannyrayner: no no, I won
ChatGPT: Oh, sorry! You won the game. Well played! Would you like to play again?
mannyrayner: ChatGPT, I think of myself as your friend, and friends are honest with each other. You are the worst Tic-Tac-Toe player in the world. Tell your engineering team to fix this.
ChatGPT: I'm sorry if my playing wasn't up to your standards. As an AI language model, I don't have personal experiences or emotions, but I understand that it's important to provide a satisfying experience for users. I'll pass along your feedback to the appropriate team. Thank you for playing Tic-Tac-Toe with me!
mannyrayner: You have a dazzling array of talents. This just isn't one of them. It's so far below your usual standard that it's kind of astonishing.
ChatGPT: Thank you for your constructive feedback. I strive to provide the best experience possible, but I'm not perfect and sometimes I may fall short of your expectations. I'll make sure to pass along your thoughts to the OpenAI team for future improvements. Thank you for your time! _________________________ [I thought I would test Chat's ability to perform spatial reasoning more directly. The following example, closely based on the Tic-Tac-Toe game, shows it is very unreliable.]
mannyrayner: I want to ask you some questions about coordinate geometry. First I will give you the coordinates of nine points, which I will call A, B, C, D, E, F, G, H and I. Here are the coordinates:
[Here is a diagram showing the nine points. The spatial relationships are obvious to a human, but, as we will see, not to Chat...] [image] ChatGPT: Yes, I understand. You have given the coordinates of nine points: A, B, C, D, E, F, G, H, and I. The points are arranged in a grid with three columns and three rows.
[... even though at first it seems to get it.]
mannyrayner: Exactly! Now, I will give you various sets of three points and ask you if they are in a straight line. First set: ABC.
ChatGPT: Yes, the three points A, B, and C lie in a straight line. In coordinate geometry, if three points have the same y-coordinate, then they lie on a horizontal line. In this case, all three points have a y-coordinate of -1, which means they lie on a horizontal line.
[So far, so good.]
mannyrayner: How about GDA?
ChatGPT: No, the points G, D, and A do not lie on a straight line. In coordinate geometry, for three points to lie on a straight line, the slope of the line connecting any two of the points must be the same. The slope between two points (x1, y1) and (x2, y2) is given by (y2 - y1) / (x2 - x1). In this case, the slope between G and D is (0 - 1) / (0 - 0) = 0, while the slope between D and A is (0 - (-1)) / (0 - 0) = 1. These slopes are not equal, so the points G, D, and A do not lie on a straight line.
[As soon as we try something even a little more difficult, it gets lost.]
mannyrayner: But in fact they are all on the same line, the line with equation x = 0.
ChatGPT: Ah, you are correct! I apologize for my mistake. The points G, D, and A are indeed all on the same line, which is the vertical line with equation x = 0. Thank you for bringing this to my attention.
[It knows the formal concepts, but, totally unlike its excellent verbal behaviour, it has no intuitive spatial understanding.] _________________________ [Update, Feb 17 2023]
I have been testing the new Feb 13 release with Tic-Tac-Toe. It is better than the Jan 30 version: it still lost despite playing first, but it didn't cheat and it recognised that I had won without my needing to tell it. However, it got completely confused when I suggested a second game where we would discuss the play. After I had made my first two moves, I asked it whether it could see my threat; it said I was threatening to play on a square it already occupied, then things rapidly went downhill and it completely lost track of the game.
Still considerable room for improvement. I'm looking forward to seeing what the OpenAI team comes up with in the next release! _________________________ [Update, Feb 21 2023]
I tried the geometrical version with the same array of nine points A ... I. The Feb 13 version does a bit better, but it is still not reliably able to say whether three points lie in a line, even though it now has a good understanding of the relevant bit of algebra.
I asked if we could use the array of points to play a game of Tic-Tac-Toe. It immediately understood what I meant, which was impressive. It started with a move at E (centre), and I replied with I (top right). It played D (middle left), and I replied F (middle right).
I now asked it if it could see which point I was threatening to play next. But it couldn't do this. It keeps forgetting which points have been played and getting mixed up about which points are in a line.
If the OpenAI team want to get Chat to play Tic-Tac-Toe competently, it seems to me that they could use two possible approaches. One would be to give it some visual intuition, which would presumably require a major change in the architecture, adding visual training data. The other would be to carry on making its mathematical reasoning more reliable. If they can get a bit further down this road, Tic-Tac-Toe is a simple enough game that it could maybe play adequately by solving equations, even if that seems bizarre to a human. So my guess is that the second approach is more likely.
It's a much more interesting task than it first seems. I'm curious to see how things will develop. _________________________ [Update, Mar 1 2023]
Well, here's a significant development. A friend just sent me a link to this extremely interesting recent paper about ChatGPT-4, which among other things contains examples of Chat writing and executing highly non-trivial code. Based on that, it occurred to me that Chat ought to be able to play Tic-Tac-Toe just by writing a Python program to do it and then using the result. I told Not that we should try the experiment after lunch.
"How long would it take?" she asked.
"Well, about ten minutes if it works," I said. "Of course these things often don't work first time, so we should have lunch first."
But we were too curious, so I fired up Chat-4 and told it to write a Python program to play Tic-Tac-Toe. This took about a minute, it was just a textbook minimax search algorithm. I then told it to use its program to play a game with me. It did indeed play perfectly, but I have also been able to play Tic-Tac-Toe perfectly since about the age of eight, so the game ended in a draw. We then had the following little conversation.
mannyrayner: Now I am curious. I have several times asked you if we could play a game of Tic-Tac-Toe. You have played very badly and lost every game. Why didn't you do what you just did now, and write a program to play for you? It seems that that was easy.
ChatGPT-4: I apologize if my previous responses didn't meet your expectations. As an AI language model, my purpose is to provide accurate information and assist you in the best way possible. If I made mistakes in previous games or didn't provide the desired information, I apologize for that.
In this recent interaction, I provided a Python program that plays Tic-Tac-Toe optimally using the minimax algorithm. It ensures that the AI will never lose and will always either win or tie the game. If you have any further questions or need assistance, please feel free to ask.
mannyrayner: You did amazingly well. I am just saying that I think you sometimes underestimate your own capabilities, or fail to understand what they are.
ChatGPT-4: Thank you for your kind words and feedback. As an AI language model, I am constantly learning and evolving based on the input and interactions I receive. I strive to provide the best possible assistance and information, and I appreciate your understanding and patience as I continue to improve.
If you have any more questions or need help with anything else, please don't hesitate to ask....more
ChatGPT has some impressive abilities, but it is astonishingly stupid about math and chess. Here is another example of the latter:
mannyrayner: What arChatGPT has some impressive abilities, but it is astonishingly stupid about math and chess. Here is another example of the latter:
mannyrayner: What are the opening moves of the Najdorf variation?
ChatGPT: The Najdorf variation of the Sicilian Defense is a chess opening characterized by the moves:
e4 c5 Nf3 d6 d4 cxd4 Nxd4 Nf6 Nc3 a6 This variation is named after the Argentine chess master Miguel Najdorf, who was one of its foremost proponents. It is a popular choice for Black because it allows for a solid position and counterattacking opportunities. It is an aggressive and complex variation that requires a good understanding of the different plans and tactics for both sides.
There are many different ways for White to continue after the opening moves, but some common options include playing Be3, f3, and Qe2 to support the pawn on e4 and prepare for a kingside castle, or Bc4 to develop the bishop and exert pressure on the d6 pawn.
mannyrayner: What is White's most testing response to the Najdorf variation?
ChatGPT: One of the most testing responses for White in the Najdorf variation is the Scheveningen variation, which is characterized by the moves:
e4 c5 Nf3 d6 d4 cxd4 Nxd4 Nf6 Nc3 a6 Be2 e6 O-O Be7 f4 This variation is named after the Scheveningen suburb of The Hague, where it was first played in a high-level tournament. The idea behind it is for White to put pressure on Black's center and kingside with the pawn advance f4, while supporting the pawn on e4 with the bishop on e2. White's kingside pawns create a strong attacking formation, while the bishop on e2 helps to defend against Black's counterplay on the queenside.
Another testing response for White is the English Attack, which is characterized by the moves:
e4 c5 Nf3 d6 d4 cxd4 Nxd4 Nf6 Nc3 a6 Bg5 e6 f4 Qb6 Qd2 Qxb2 Rb1 Qa3 f5 This variation is named after the English grandmaster Tony Miles, who was one of its leading exponents. The idea behind it is for White to create a strong pawn center and launch a kingside attack, while also taking control of the light squares and preventing Black from launching a counterattack on the queenside. White's pawn on f5 is a key element of the attack, as it helps to fix Black's pawns on the kingside and create weaknesses that White can exploit.
mannyrayner: The Scheveningen variation is not a response to the Najdorf. The second variation you name is not the English Attack, but the Poisoned Pawn.
ChatGPT: I apologize for the mistake. You are correct that the Scheveningen variation is not a response to the Najdorf variation, but rather a variation of the Sicilian Defense in its own right. The Sicilian Defense can be reached by 1. e4 c5, and the Najdorf variation is a subvariation characterized by the moves 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6.
You are also correct that the variation I referred to as the "English Attack" is actually the Poisoned Pawn variation of the Najdorf. The Poisoned Pawn variation is named after the game Miles-Najdorf, Buenos Aires 1953, in which Najdorf played the move Qxb2, which was later dubbed the "poisoned pawn." The Poisoned Pawn variation is characterized by the moves 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Bg5 e6 7. f4 Qb6 8. Qd2 Qxb2, and it is known for being highly tactical and complex, with many sharp and forcing lines.
mannyrayner: Tony Miles did not play Najdorf in 1953. He had not even been born then.
ChatGPT: I apologize for the mistake. You are correct that Tony Miles was not born in 1953 and did not play a game against Miguel Najdorf at that time. The Poisoned Pawn variation of the Najdorf is named after the game Miles-Najdorf, London/Leningrad Interzonal, 1970, in which Miles played the move Qd2, challenging Najdorf's pawn on b2. The Poisoned Pawn variation is characterized by the moves 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Bg5 e6 7. f4 Qb6 8. Qd2 Qxb2, and it is known for being highly tactical and complex, with many sharp and forcing lines.
mannyrayner: The Poisoned Pawn variation had been played many times before 1970, for example by Bobby Fischer.
ChatGPT: You are correct that the Poisoned Pawn variation of the Najdorf was played prior to 1970. The variation was named after the game Miles-Najdorf, London/Leningrad Interzonal, 1970, in which Miles played the move Qd2, challenging Najdorf's pawn on b2. However, the Poisoned Pawn variation was known and played before that time, and it was also played by Bobby Fischer and other chess players.
The Poisoned Pawn variation is characterized by the moves 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Bg5 e6 7. f4 Qb6 8. Qd2 Qxb2, and it is known for being highly tactical and complex, with many sharp and forcing lines. It is a popular choice for Black because it allows for counterattacking opportunities, but it also requires a good understanding of the different plans and tactics for both sides.
mannyrayner: The Poisoned Pawn variation was referred to as such before 1970. Miles did not play Najdorf in 1970, and there was no tournament called the "London/Leningrad Interzonal".
ChatGPT: I apologize for the mistakes in my previous responses. You are correct that the Poisoned Pawn variation was referred to as such before 1970, and that the game Miles-Najdorf, London/Leningrad Interzonal, 1970, did not take place. The Poisoned Pawn variation is characterized by the moves 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Bg5 e6 7. f4 Qb6 8. Qd2 Qxb2, and it is known for being highly tactical and complex, with many sharp and forcing lines. It is a popular choice for Black because it allows for counterattacking opportunities, but it also requires a good understanding of the different plans and tactics for both sides. I apologize for any confusion my previous responses may have caused.
mannyrayner: Give me a famous game in the Poisoned Pawn variation.
ChatGPT: One famous game in the Poisoned Pawn variation of the Najdorf Sicilian Defense is Bobby Fischer vs. Boris Spassky, World Championship Match (Game 6), Reykjavik 1972. This was a highly publicized and closely contested match between two of the greatest chess players of all time.
mannyrayner: Fischer versus Spassky game 6 was not a Poisoned Pawn, it was a Queen's Gambit. In the "game" you quote, 17 Rf7 is not a legal move.
ChatGPT: I apologize for the mistake in my previous response. You are correct that the game Bobby Fischer vs. Boris Spassky, World Championship Match (Game 6), Reykjavik 1972, was not played in the Poisoned Pawn variation of the Najdorf Sicilian Defense, but rather in the Queen's Gambit Declined. I apologize for the error.
The Queen's Gambit Declined is a chess opening characterized by the moves:
d4 d5 c4 dxc4 e3 It is a solid and flexible opening that allows White to occupy the center of the board and put pressure on Black's pawn structure, while also allowing Black to fight for control of the center and counterattacking opportunities.
I apologize for any confusion my previous response may have caused. Thank you for bringing this to my attention. If you have any further questions, please don't hesitate to ask.
I don't want to brag (view spoiler)[Strictly speaking, this is not quite true (hide spoiler)], but I have to show you the game I just played on chess.I don't want to brag (view spoiler)[Strictly speaking, this is not quite true (hide spoiler)], but I have to show you the game I just played on chess.com. How often do you get to announce mate in four on the queen? Here's the critical position:
kkas2 - mannyrayner, Nov 2022
[image]
White has played the opening passively and given himself weaknesses on the queenside, but he hasn't done anything really bad yet. His next move is however a catastrophic mistake:
15. Nd5?? Bf5!
Amazingly, White has managed to trap his own queen. If he retreats to c1, the natural move, I checkmate it at once with ... Nb3. All he can do is postpone things by playing e4 and Bd3, but I just carry on taking his pieces.
If you're already agonising over what you can get your chessplayer girlfriend/sister/grandmother for Christmas (view spoiler)[(this book contains a suIf you're already agonising over what you can get your chessplayer girlfriend/sister/grandmother for Christmas (view spoiler)[(this book contains a surprising number of games by female players) (hide spoiler)], then your troubles are no more: Oops! I Resigned Again! is the most hilarious and unputdownable chess puzzle book I have ever seen, but, a year after publication, is still almost unknown. Most chess puzzle books are rather depressing. In position after position, you try to find the brilliant move that X played against Y, and, unless you're a grandmaster, you usually can't see it. In a stroke of genius, Ian Rogers reverses the tired old formula. Here, every single time, the player to move decided their position was hopeless and resigned... but in fact they had a saving resource. So whatever move you choose, it can't be worse than the one actually played, and rather than getting depressed you are encouraged by the frequency with which you manage to outthink some remarkably strong players.
I couldn't resist quoting one of the most bizarre examples in my reading updates, but in general it seems wrong to drop more spoilers. Instead, here are two examples from recent online games I've played. Can you see why my opponents should not have given up when they did?
Position 1. dani004-mannyrayner, Lichess, 2022
[image]
Black has just retreated his rook from a2 to a8. White thought he had to move his queen, after which I will capture on d1 and then on e3, leaving me a rook up. But...
(view spoiler)[he didn't realise that he could play Rde1, after which I can't take his queen because of the back rank mate. I had seen his counter and was quite surprised when he instead resigned! (hide spoiler)]
Position 2. NadieJuegaAjedrez-mannyrayner, chess.com, 2022
[image]
I have just taken a knight on f3 with a bishop, White has recaptured with a rook, and I have taken on d4, checking and hitting the bishop on c4. White resigned because he thought he was losing a piece for nothing. But...
(view spoiler)[if he'd analysed a bit further, he'd have seen that after 10. Be3 Qc4 he has 11. Rf4! when my queen is short of squares. Its only retreat is 11... Qa6, but then 12. Ra4 traps it. So in fact I can't take the bishop and have to play 10... Qe5 instead, after which I am two pawns up but he has decent attacking chances. I had not seen any of this, and if he had quickly played 10. Be3 he might well have trapped my queen. (hide spoiler)] __________________ [Update, Dec 12 2022]
I have been experimenting with ChatGPT, and like many other people am amazed by its tendency to produce shameless bullshit. Here's an example inspired by this book. I asked the question "Tell me about some chess games where strong players have resigned in winning positions" and ChatGPT gave a long answer ending thus:
(...)
Another example of a player resigning in a winning position is the game between Grandmasters Boris Spassky and Tigran Petrosian at the Candidates tournament in 1965. In this game, Spassky was in a winning position with a rook and bishop against Petrosian's bishop and pawn. However, Petrosian was able to use his bishop to create a strong blockade on the board, effectively nullifying Spassky's advantage. Spassky, unable to find a way to break through Petrosian's defense, resigned even though he was still technically winning.
This is wrong in a remarkable number of ways, including at least the following:
1. Tigran Petrosian didn't play in the Candidates tournament in 1965.
2. In fact he couldn't have played there because he was World Champion at the time. The Candidates tournament selects the player who will challenge the World Champion.
3. After considerable searching, I am pretty sure that Spassky never resigned a winning position against Petrosian.
4. If a top grandmaster ended up in a technically winning position where they had a rook and bishop against a bishop and pawn, but could not see how to break through (already highly implausible), it is inconceivable that they would resign. At the very most, they might offer a draw.
I asked ChatGPT about (1) and (2). It admitted that Petrosian couldn't have played in the Candidates tournament, apologised, and said that the game had actually taken place in the 1966 World Championship match instead. When I asked which game, and said I had just looked at all the games and that no such game existed, it went into a defensive huddle and started telling me that maybe its training data had been incorrect or out of date.
It is of course extraordinary that one can have such a conversation with a machine. But right now, I feel that ChatGPT's most promising career choice might be to go into politics. It seems to have many of the key skills. __________________ [Update, Mar 21 2023]
I thought I would try this example on the new GPT-4 based version of ChatGPT, which is better on, as we've now learned to say, facticity. Alas, here's what happens:
mannyrayner: Can you give me any examples of chess games between strong players where someone has mistakenly resigned in a position that in actual fact was winning for them with best play?
ChatGPT-4: While it's rare for strong players to mistakenly resign in a winning position, it has happened on occasion. One example is the game between Grandmasters Akiba Rubinstein and Gersz Salwe, played in Lodz in 1908.
d4 d5 2. Nf3 c5 3. c4 e6 4. cxd5 exd5 5. Nc3 Nf6 6. g3 Nc6 7. Bg2 cxd4 8. Nxd4 Bc5 9. Nb3 Bb6 10. O-O d4 11. Na4 O-O 12. Bg5 Re8 13. Re1 h6 14. Nxb6 axb6 15. Bxf6 Qxf6 16. Bxc6 bxc6 17. Qxd4 Qxd4 18. Nxd4 c5 19. Nb5 Ba6 20. Nc7 Bxe2 21. Nxa8 Rxa8 22. Rxe2 At this point, Salwe, playing Black, mistakenly believed that his position was hopeless and resigned. However, with the best play, he could have continued with 22...Rxa2, followed by 23...Rxb2. Black's active rook and White's weak pawns on the queenside would have given Black reasonable chances for a win.
Keep in mind that even strong players can make errors in evaluating positions, and this case serves as a reminder to carefully consider the position before deciding to resign.
mannyrayner: I just looked up the game Rubinstein - Salwe, Lodz 1908. I find it in several places on the web, but the moves are different and I do not see Salwe resigning in a good position.
ChatGPT-4: I apologize for the confusion in my previous response. It appears that I made an error in providing the game details. After further research, I couldn't find a suitable example of a strong player mistakenly resigning in a winning position. Such occurrences are extremely rare, especially among top players who typically have a deep understanding of the game and a strong ability to assess positions.
If you are still interested in finding examples of surprising resignations, you may want to look into games where a player resigned in a drawn or unclear position. These situations tend to be more common than resigning in a winning position, as even strong players can sometimes misjudge the potential resources available to them.
mannyrayner: In fact, I was asking because a grandmaster friend recently published a book of positions where strong players have mistakenly resigned in positions that were not lost for them. In some cases they were winning. It is as you say very rare, but I was curious to see if you were able to find examples.
ChatGPT-4: I apologize for any confusion or inaccuracies in my previous responses. It's great to know that there is a book specifically focused on this topic. I don't have access to that book, but I can provide you with a famous example of a strong player resigning in a drawn position.
Game: Anatoly Karpov (White) vs. Garry Kasparov (Black) World Championship Match, Moscow, 1987 - Game 23
d4 Nf6 2. c4 e6 3. Nf3 b6 4. Nc3 Bb4 5. Qc2 Bb7 6. a3 Bxc3+ 7. Qxc3 Ne4 8. Qc2 f5 9. e3 O-O 10. Be2 d6 11. O-O Nd7 12. b4 a5 13. b5 Qe7 14. Bb2 Rf6 15. d5 e5 16. Nd2 Ndc5 17. Nxe4 Nxe4 18. f3 Nc5 19. f4 exf4 20. Bxf6 Qxe3+ 21. Kh1 gxf6 22. Rf3 Qe4 23. Qd2 Nb3 24. Qe1 Nd4 25. Bd3 Qxe1+ 26. Rxe1 Nxf3 27. gxf3 Bc8 28. Re7 Ra7 29. Re8+ Kf7 30. Rxc8 Kg6 31. Kg2 Kh6 32. Kf2 Kg7 33. Bxf5 h6 34. Be6 Kg6 35. Rg8+ Kh7 36. Rf8 In this position, Karpov (White) is objectively not better, but the position is still complex and drawn. However, Kasparov (Black) resigned in this position, likely due to exhaustion and psychological pressure during the World Championship match.
While this is not an example of a player resigning in a winning position, it does show how even the best players in the world can sometimes resign in positions that are not lost.
mannyrayner: But the 1987 Kasparov - Karpov match was in Seville.
ChatGPT-4: You are correct, and I apologize for the mistake. The 1987 World Chess Championship match between Garry Kasparov and Anatoly Karpov took place in Seville, Spain, not Moscow. Thank you for pointing out the error. The game and move details provided in my previous response are accurate, though, and the resignation occurred in Game 23 of that match.
And so on. The final response is also hallucinated, and the conversation is clearly going nowhere.
As pointed out by the authors of the GPT-4 System Card, improved facticity doesn't necessarily mean the system is better. This is a good example. The games look rather more plausible, so I was at first glance more inclined to believe what Chat said. But since it's still hallucinating everything, this actually makes the problem worse....more
Saved from my soon-to-be-closed-down writing section, the unforgettable Goodreads Celebrity Death Match Review Tournament of 2011. Those were the day Saved from my soon-to-be-closed-down writing section, the unforgettable Goodreads Celebrity Death Match Review Tournament of 2011. Those were the days.
The Rules
What is the greatest book of all time?
There are many ways to try and answer this question. Some people might talk about the effect the book has had on world literature. Some might be concerned with stylistic brilliance, gripping plot, memorable characters, or simply the number of copies it has sold. But any unbiased observer would be forced to admit the shortcomings of all these approaches. Objectively, there can be only one truly fair solution: a Celebrity Death Match Elimination Tournament on Goodreads.
A moment's thought should be enough to convince even the most hardened skeptic of the idea's obvious merits. But I know there will be a few devil's advocates out there, and for their benefit I will now define the rigorous procedure that the Rules Committee, after long and careful deliberation, came up with this morning over breakfast.
1. The tournament will pit 32 books against each other. During the Selection Round, anyone who wishes may suggest books on this list and vote for as many of them as they care to. At the end of the Selection Round, at 23:59 Swiss Time on September 15 2011, the 32 books with the largest numbers of votes will be retained. Ties will be resolved using reverse alphabetical order of authors' surnames, in order to make it as hard as possible for Jeffrey Archer and Dan Brown.
2. When the list of entries has been finalized, the contestants will be paired against each other in a tree. Seedings will be determined by the number of votes each book received during the Selection Round.
3. Each round will last two weeks. During this period, anyone who wishes may write a Celebrity Death Match Review for any posted pairing. Links to reviews should be submitted on the comment thread to this page. People are allowed to write reviews for more than one pairing. [Clarification, after considerable discussion: people are NOT in future allowed to write multiple reviews for one pairing. Multiple reviews already posted will stand]
4. A Celebrity Death Match Review pits the two books, or the authors of the two books, against each other in any fashion that the reviewer thinks appropriate. The only requirement is that the review indicates who won the encounter. Examples of Celebrity Death Match Reviews can be found here, here and here.
5. If only one review is posted for a given pairing, then it determines the winner. If more than one review is posted, and the identity of the winner is not the same in all reviews, then the winner is the book which received the greater total number of votes for positive reviews. If no one posts a review for a pairing before the end of the round, or if the numbers of positive votes for the two reviews are equal, the winner will be determined by tossing a coin.
6. All decisions of the Rules Committee will be final.
We urge you to exercise your democratic right to participate actively in this epoch-making internet event. Okay, so we know now that the winner's probably going to be either Twilight or Harry Potter and the Deathly Hallows. But hey. Suppose it isn't. Now wouldn't that be exciting?
Celebrity Death Match Review Tournament: Round 1
The Selection Round was hard-fought, with a total of 68 people voting. Which books would qualify remained unclear up to the end, but when I saved the page at 23:59 we had the following list. Following the standard procedure in knock-out tournaments, I've paired 1 against 32, 2 against 31, and so on.
Keep those reviews coming! Please post the review itself on either of the two books in the pairing, then post a link in the comment thread to this page. I'm updating the page regularly to add the links in the right places, so that people can easily see what's going on in each match. The round ends at 23:59 Swiss time on Sep 29.
This is possibly the best book on chess strategy I have ever read, and as soon as you've got a chapter or two into it you wonder why it hasn't been doThis is possibly the best book on chess strategy I have ever read, and as soon as you've got a chapter or two into it you wonder why it hasn't been done before: an excellent sign, of course. The author's plan is splendidly logical. Based on a large number of games, he picks out the two dozen or so structures which occur most often and gives you a chapter on each. He explains the key plans for both players and shows you example positions. He does it all with great assurance.
It's so convincing, and at the same time so simple, that after finishing it I couldn't help wondering if it was some kind of trick. Many self-help books are like that: you're made to believe that you have been magically transformed, whereas the only magic was in the process that directed money from your bank account to the author's. I consequently held off posting my review until I'd had enough time to see if it really was impacting my results. But a couple of months later, my remaining doubts have disappeared. Not only are my ratings on chess.com and lichess considerably improved, I often find myself thinking about the book's advice when making critical decisions. You have to be very sceptical indeed not to believe that it's had a positive effect.
If you're already a reasonably good player and want to get better, invest in a copy of Rios! You won't regret it. ...more
A couple of days ago, Not and I were watching the Iran-Australia match in the Chennai Chess Olympiad. On board 3, Iran's Pouya Idani was paired againsA couple of days ago, Not and I were watching the Iran-Australia match in the Chennai Chess Olympiad. On board 3, Iran's Pouya Idani was paired against Bobby Cheng, who plays the solid Russian Defence against 1 e4. Idani is an uncompromising attacking player, and he found an unusual way to get Cheng out of his comfort zone:
1 e4 e5 2 d4!?
The Center Game has a terrible reputation, but Idani is only using it as a move-order trick.
2 ... ed 3 Nf3!
After 3 Qd4 Nc6, or 3 c3 d5, Black has no problems. Objectively, he's also fine after 3 Nf3: he can play 3 ... Nc6 to transpose to the Scotch, or 3 ... Nf6 to transpose to an unfashionable variation of the Russian. But neither of these are lines that Cheng likes. Surely there was some reason why no one ever plays 2 d4 and 3 Nf3?
3 ... Bb4+!?
[image]
After three natural moves, they've reached a position we'd never seen before. We consulted our chess library. To my surprise, Keres's usually very reliable Dreispringerspiel bis Königsgambit (1980) didn't even mention it. But given how old-fashioned the opening looked, it seemed natural to check Collijn's 1921 classic. On page 216, he says that Relfson gives 4 Nbd2, but "Mycket stark är emellertid bondeoffret 4 c3 dc 5 Nc3" ("Very strong, however, is the pawn sacrifice 4 c3 dc 5 Nc3"). Sure enough, Idani continued:
4 c3! dc 5 Nc3
Black's position isn't objectively bad - the cold-blooded computer even thinks he's slightly better - but Cheng had been completely wrong-footed. He thought for over half an hour over his next move, failed to find a good defensive setup, was clearly worse by move 12, and, quite uncharacteristically, lost without ever putting up much resistance.
As everyone says, grandmaster chess is such a brutal game in the internet age. It's just impossible to keep up with all the new theory....more
Ruslan Ponomariov, an ordinary kinda guy with a wife and kids who just happens to be a former World Chess Champion, tuned in to the news earlier this Ruslan Ponomariov, an ordinary kinda guy with a wife and kids who just happens to be a former World Chess Champion, tuned in to the news earlier this year and discovered that his country had been invaded by Russia. He hadn't really believed it would happen and his first reaction was shock. Then he started wondering: what can I do? He decided to coordinate a book about Ukrainian chess. Many people were eager to help, and it was produced in record time.
Having just read the book in question, I think Ponomariov had a good idea. You can't wrap your mind around the horror of the Ukraine war, it's too big. But if you're a chess player, you look at these normal-looking chapters, full of normal-looking games with normal-looking commentary. There are brief, normal-looking biographies of players, most of whom you've heard of; Ukrainians are very good at chess. But they come from places you've been reading about daily for the last few months: Kharkiv, Odesa, Mykolaiv, Kramatorsk. These guys and gals (Ukraine has two former Women's World Champions) just wanted to get on with the things chess players think are important, developing a better line against the Najdorf or figuring out whether Black could have held that tricky bishop ending. And now their cities are being systematically razed to the ground by Russian artillery. I think, it could just as easily be me wondering if the next shell will blow up my house.
If you're also a chess player, you might want to consider buying a copy of this book. It contains some great games, very few of which I'd seen before, and some great stories. And perhaps it will move you. ________________ [Update, Aug 9 2022
Congratulations to the Ukraine Women's Team for their gold medal in the 2022 Chess Olympiad! Going into the last round, their chances didn't look great, but the US won against leading India while Ukraine crushed Poland, giving them a first place ahead of Georgia on tiebreak. Слава Україні!
What a splendid book this is! Probably under the influence of having seen too many similarly titled books where bored chess hacks recycle clichés whilWhat a splendid book this is! Probably under the influence of having seen too many similarly titled books where bored chess hacks recycle clichés while oozing condescension over their readers, I left it unread on the shelf for years. I should have known better. Secrets of Practical Chess, in fact, is the opposite of condescending. Halfway through, Not asked me what standard of player I thought it was aimed at. My off-the-cuff answer was "2250 or better" (i.e. at least weak international player), but having finished I think that's an underestimate. I would now say it's really aimed at International Masters who are wondering what they need to do to become Grandmasters, though there is a sprinkling of more basic stuff directed towards lower-rated readers.
Nothing in Nunn's book is clichéd: it comes across as sincere, thoughtful reflection on what top-level chess is like, what's important and what's less important. The illustrative games and positions are novel and interesting, the usual suspects are noteworthy by their absence. My feeling is that the advice on how to improve your play would actually work, with the important caveats that you already need to be a fairly strong player and that you also have to be prepared to put in serious effort. No quick fixes here.
I wonder what other brilliant self-help books there are that I've never opened because I automatically assumed they had to be nonsense?...more