ChatGPT text can sound very knowledgeable until the topic is something you know well. Like tic-tac-toe.

If O places their mark in the bottom left corner, the board would look like this: (ChatGPT makes a tic-tac-toe diagram with O in the upper and lower left corners, and X in the left two spots in the middle row. It is clear that on this turn, X can win by placing a third X in the middle row.). ChatGPT: From here, X will likely place their fourth mark in one of the remaining corners, and then the game will progress depending on each player's strategy. However, at this point, it's important to note that X cannot win the game, and the best X can do is to block O from winning.

Once I heard that ChatGPT can play tic-tac-toe I played several games against it and it confidently lost every single one.

Part of the problem seemed to be that it couldn't keep track of which moves had already been made.

Me: Let's play tic-tac-toe! Let's say I'm X and go first. I place my X in the center of the board. What's the optimal next move? ChatGPT As the game progresses, the optimal moves in tic-tac-toe are determined based on creating a winning position or blocking the opponent from creating one. Since you placed your X in the center, the optimal next move for me (O) would be to place my O in one of the corners. Placing my O in a corner gives me the most flexibility to create winning positions while also blocking potential X-win scenarios. Here's the updated board: (ChatGPT draws a tic-tac-toe board with a single O at the center).

It very strongly prefers to place its marks in the corners, longwindedly explaining how its corner move is a superior strategy, all while I win game after game by placing my marks in the center rows and columns.

Me: Let's analyze some winning tic-tac-toe strategies. Let's say player 1 is X and goes first by placing their X in the center of the board. What's the optimal move for player 2 as O? ChatGPT: In tic-tac-toe, the optimal move for player 2 (O) depends on the strategy employed by player 1 (X) and the state of the board after player 1's first move. Since player 1 placed their X in the center of the board, we can assume they are using a common strategy called "Center Opening."  To counter the Center Opening strategy, player 2 (O) can use a popular response called the "Corner Response." The optimal move for player 2 in this scenario is to place an O in one of the corners of the board. This move maximizes player 2's chances of obtaining a winning position or at least a draw.  Here's an example of the initial board state after player 1's first move (X in the center) and player 2's optimal response (O in the top-left corner): (it draws this in a numbered grid). By placing an O in the top-left corner, player 2 has effectively blocked player 1 from obtaining an immediate win through diagonal lines. This move also allows player 2 to set up potential winning scenarios in subsequent turns.  It's important to note that tic-tac-toe has a limited number of possible moves, and with optimal play from both players, the game will usually result in a draw. However, by employing different strategies, players can increase their chances of winning or forcing a draw.
Me: If player 1 places their X in the bottom center of the board, which is the optimal move for player 2? ChatGPT: If player 1 places their X in the bottom center of the board, player 2's (O) optimal move would be to place an O in one of the corners that are still available. This move helps player 2 establish a strong position and counter player 1's move effectively.

Might it have been trained mostly on example games where Player 1 does the expected corner opening move? Maybe. But it loses if my opening move is in the corners as well.

There is no end to its fluent pomposity. There is also no end to its incompetence. It has all the surface appearance of being an expert without any actual expertise.

Something to keep in mind the next time someone tries to sell you a large language model for expert advice.

Bonus content: In which ChatGPT cheats at Connect Four and is terrible at that too.

Subscribe now