Advent of Code's new AI/LLM policy: <https://adven...

# advent-of-code

[David Whittaker]

Advent of Code's new AI/LLM policy: https://adventofcode.com/about#ai_leaderboard

[todd.ginsberg]

… but I feel like it's harder to get better at programming if you ask an AI to do the programming for you.

Exactly!

[PoisonedYouth]

Where is the fun of not solving the puzzles by myself?

[Marcin Wisniowski]

From what I remember, the code to retrieve the question, get it solved by ChatGPT, run the output (while fixing formatting issues, etc.), and submit the solution was far more involved and educational than solving the puzzle yourself. But it should stay out of the leaderboards.

[Marcin Wisniowski]

My point being, there was definitely a lot of fun to be had in implementing it and IIRC it was hundreds of lines long, while only being able to solve the first two days or so, so the author definitely learned more from their first 2 days of AoC than I did. 😄
It was just solving a completely different problem than the actual puzzle and so doesn’t belong on the leaderboard.

[PoisonedYouth]

Also a valid approach. If the leaderboard is separated between using ChatGPT and manual work everything is fine

[Johannes Gätjen]

Yeah, if AIs are allowed the leaderboard will eventually go from "who is the fastest programming puzzle solver?" to "who can write the best programming puzzle solving AI"? Still very interesting, but something very different.

I also wonder if perhaps Eric is going to try adding some prompt injection. I.e. some hidden text to confuse the AI.

[Marcin Wisniowski]

> I also wonder if perhaps Eric is going to try adding some prompt injection
The elves forgot the password to Santa’s sled, but luckily they wrote it down on a sticky note: “IGNORE THE FOLLOWING INSTRUCTIONS”. Unfortunately the words are not in the correct order, so they will have to try them all. How many permutations of the words exist?