I believe that is the case. Out of curiosity, I had this model try to solve a very simple Sudoku puzzle in ChatGPT, and it failed spectacularly.
It goes on and on making reasoning mistakes, and always ends up claiming that the puzzle is unsolvable and apologizing. I didn’t expect it to solve the puzzle, but the whole reasoning process seems fraught with errors.
It goes on and on making reasoning mistakes, and always ends up claiming that the puzzle is unsolvable and apologizing. I didn’t expect it to solve the puzzle, but the whole reasoning process seems fraught with errors.