> As others have said, part 2 of today's was really difficult.
I suspect it was purpose-built to foil ChatGPT. I solved it on my own first and then went back and tried to help GPT4 write a solution. Even after carefully walking it through a strategy, discussing edge cases, and having it outline a solution in pseudocode, it still failed completely to translate our conversation into working code. (It didn't even work for non-tricky input.) It did anticipate the edge cases though, so that's something.
Did anyone have any better luck with ChatGPT? I wonder if LLM-resistant puzzles and generally greater difficulty (at least for the second star) will be a theme this year.
I got it to work with ChatGPT, took around 5 attempts, but that was mainly me not understanding all the edge cases. Once I came up with a strategy that would work, ChatGPT gave me working code.
My strategy was not efficient, but did work.
I walked the string twice, first LTR and replaced all found strings with numbers, then walked right to left, and replaced all backwards strings to numbers.
Then took the first digit from the left walked string, and the last from the right walked string.
I suspect it was purpose-built to foil ChatGPT. I solved it on my own first and then went back and tried to help GPT4 write a solution. Even after carefully walking it through a strategy, discussing edge cases, and having it outline a solution in pseudocode, it still failed completely to translate our conversation into working code. (It didn't even work for non-tricky input.) It did anticipate the edge cases though, so that's something.
Did anyone have any better luck with ChatGPT? I wonder if LLM-resistant puzzles and generally greater difficulty (at least for the second star) will be a theme this year.