Awesome

Solving Advent of Code 2022 with ChatGPT

In this repository, I log my attempt to solve the Advent of Code 2022 with ChatGPT.

The purpose of this experiment is to see if modern language models can reliably solve increasingly complex code problems. While it's still not "real-life production code", Advent of Code puzzles offer an increasing level of complexity, ready to use test cases, and data on many human counterparts who are solving the same problems, so it's a good testbed to compare the machine to human performance.

All the puzzles are copied and pasted to the ChatGPT interface, and the solutions are run as-is. The only clarification I make is to use Python and assume that input is provided as a file "input.txt". After receiving a code snippet, I run it with Python and paste the output to the puzzle input field. If the answer is incorrect, I simply tell ChatGPT that it is wrong. I save the whole conversation log in a markdown file and all individual code snippets in separate python files.

Results

• Day 1: Solved at third try
• Day 2
  • Solved part one at 12th try, with hints from developer. Had to re-start chat after it became unresponsive.

(Most of this README was written by Github copilot).