Out of curiosity after finishing my solution, I tried it with chatgpt 4.0
Part1 worked after me explaining a tiny bug.
Part2 however never worked. Even after explaining exactly where the bug was in the python solution got came up with, it couldn’t fix it.
It was quite fascinating watching it try over and over with different approaches, but it couldn’t even get the example working.
This just goes to show how good of a puzzle maker Eric is if it stumped gpt4 on day1 when last year gpt3.5 did the first 5 days.
Last year, I used ChatGPT on one of the first puzzles, and ended up writing a blog post about it, where I sort of do commentary on the conversation.
It's funny to read this a year later, and filter it through my experiences with ChatGPT over the last year. Some of it still rings true, some of it would probably be much improved with GPT-4. But the places where the LLM fell down in my examples are still the same kinds of issues you get using GPT as an assistant today.
This just goes to show how good of a puzzle maker Eric is if it stumped gpt4 on day1 when last year gpt3.5 did the first 5 days.