Advent of Code 2024

5 min read

Every December, Advent of Code turns up in my feeds and I feel roughly equal parts keen and dread. It’s a yearly programming challenge that wraps algorithmic problems in a festive narrative.

For years, I flirted with the idea of participating. And for years, I always found a way to dodge it. Sometimes, I’d forget about it until a stray tweet reminded me a week into the event, leaving me feeling hopelessly behind. Other times, summer holidays and the thought of spending time glued to a screen felt like a crime. And, honestly? The puzzles kind of intimidated me. They felt like they were tailor-made to expose every weak spot I had. It was always easier to skip it than risk confirming my doubts.

But this year? I thought, “Fuck it! Let’s go!” Without overthinking, I jumped in headfirst. What’s the worst that could happen? A little failure never killed anyone.

What Is Advent of Code?

Advent of Code is an online event where a new coding puzzle drops every day from December 1st to 25th. Each challenge is wrapped in a whimsical holiday narrative, starting relatively simple but quickly escalating into complex algorithmic and logical brain-benders.

The puzzles are delivered in two parts. Solve part one, and you unlock part two, which often flips the original problem on its head. It’s a mix of logic, efficiency, and, occasionally, good ol’ brute force.

For leaderboard chasers, speed and algorithmic skill are key. For the rest of us, it’s a chance to learn some techniques we don’t reach for at work.

Setting Goals (And Embracing Reality)

I went in fully aware that December is probably the worst time to take on a daily challenge. Between wrapping up work projects, wrangling school holidays, and prepping for Christmas, my time was limited. I knew I’d start strong and probably taper off, and that’s exactly what happened.

Still, I set a few goals:

  1. Finish what I could: Even if I didn’t complete all 25 days, just sticking with it would be a win. How it went: Against my own expectations, I actually completed all 25 challenges! There were definitely days when I wanted to quit, especially during the grid algorithm grind, but pushing through paid off.
  2. Experiment with LLMs: I wanted to see how much tools like ChatGPT, Claude, and Copilot could actually do for me. How it went: As expected, LLMs were great for boilerplate code and algorithm suggestions, but they continue to struggle with the more complex puzzles. I ended up refining my approach in this challenge rather than discovering anything new.
  3. Learn something new: Whether it was algorithms or smarter ways to work, I wanted to walk away having learned something. How it went: I went from dreading graph algorithms to actually understanding them, and the Day 14 PNG compression hack is still my favourite thing I did all month. I also picked up some debugging tricks, like visualising grid states as images.

The LLM Dilemma: Cheating or Innovating?

Now, the elephant in the room: using LLMs for Advent of Code. The event’s creator explicitly discourages AI use for climbing the global leaderboards (see here). But I wasn’t competing with anyone. I was there to find out what these tools are actually good for.

I leaned heavily on LLMs, primarily the free tiers of ChatGPT and Claude, with Copilot in the mix. Early on, they were great for simpler problems, especially part one of puzzles. But as the challenges grew more complex, their limitations became glaringly obvious. Multi-step reasoning and edge cases? Nope. They weren’t cutting it. They were still worth having around for boilerplate and for quick refreshers on things like BFS and Dijkstra’s algorithm.

Favourite Puzzle: Day 14: Restroom Redoubt

Day 14 was my favourite, and not because it was easy. The challenge involved identifying patterns in a 2D grid, and I initially tackled it algorithmically. But after struggling with 2D pattern matching and anomaly detection, I resorted to brute force: printing 10,000 frames as PNGs to scan manually.

This was, predictably, slow as hell. So, I took my dog for a walk. And mid-walk, it hit me: PNGs use lossless compression. The Christmas tree pattern, being the most ordered, would compress the most. Back home, I sorted the files by size, and boom: the smallest PNG revealed the tree! Later, I redeemed my brute-force hack by calculating the time frame with the most consecutive robots. Was it efficient? Absolutely not. But I’m still pleased with it.

Biggest Challenge: The Grid Grind

By mid-month, I hit a wall. Many puzzles relied on pathfinding in grids - problems that practically screamed for algorithms like BFS, DFS, and Dijkstra. My lack of familiarity with these techniques made it feel like a slog. I cracked the sads a bit when I opened Day 15 to find yet another grid problem. But on the upside, I ended up learning a ton about these algorithms. The grind was real, but the payoff was worth it.

Final Thoughts

Against all odds, I finished all 25 challenges:

What started as a fuck it, let’s go! experiment turned into a month of being alternately stuck and delighted. Sure, I leaned on LLMs more than some purists would approve of, and some of my solutions were more “creative” than “elegant”. But that’s what made it valuable.

If you’re curious about my solutions or want to dig into the code, you can check out my Advent of Code solutions on GitHub.

Would I do it again? Ask me next December. But I finished the thing I’d spent years dodging, and I no longer dread opening a puzzle and finding a grid in it.