I would have much preferred if the author spent more time giving an objective assessment of what ChatGPT had actually accomplished at each step. The "Labyrinth

I would have much preferred if the author spent more time giving an objective assessment of what ChatGPT had actually accomplished at each step. The "Labyrinth Sudoku" description feels like a classic language model speciousness. It doesn't actually work: you can't fit the digits 1-9 if you can't use all the cells, and it hasn't modified sudoku rules in a way that makes paths relevant. Maybe you can come up with a way to salvage it, but ChatGPT didn't. The initial rules for Sum Delete should be read with this in mind: it sounds reasonable, but there's no reason to trust that it can make a puzzle at all, let alone a good one. Also, unsurprisingly, the provided puzzle isn't solvable (the 25 column requires use of the 9 in the first row). Similarly, I'd love a critical analysis of the initial code. Did it guarantee solvability? An awful lot can be swept under "improving the design".