What if AI fails the moment the problem changes?
Most AI systems today are great at recognizing patterns — but what happens when the problem itself is different from anything they’ve seen before?
That’s exactly the challenge behind the ARC Prize 2026 on Kaggle.
Getting Started
I recently joined the ARC-AGI-2 competition on Kaggle, which challenges participants to build AI systems capable of solving problems they’ve never seen before.
Unlike typical machine learning tasks, ARC is not about training on massive datasets.
Instead, it focuses on something closer to human intelligence:
reasoning and generalization.
My First Approach (Baseline)
To get started, I built a very simple baseline model.
The goal was not to solve the problem yet — but to:
- Understand the dataset structure
- Learn how submissions work
- Successfully make my first submission
Here’s the basic idea:
- Read input grid
- Generate output grid with the same shape
- Fill it with zeros
Yes — it's intentionally simple.
Code & Repository
You can check my first implementation here: https://github.com/Bangkah/arc-agi-learning
This repository documents my learning journey as I explore reasoning-based AI.
First Submission (Kaggle)
Why Start Simple?
Because ARC is not about jumping into complex models.
It’s about:
- Understanding transformations
- Identifying patterns
- Building logic step-by-step
Even advanced AI systems still struggle with ARC tasks.
Why ARC is Hard
ARC tasks require:
- Understanding abstract patterns
- Applying transformations
- Generalizing from very few examples
Unlike traditional machine learning, there is no training phase with millions of samples.
Each problem is a completely new challenge.
First Submission Result
My first submission didn’t perform well — and that’s expected.
The real goal was:
✔ Understanding the pipeline
✔ Getting familiar with the problem
✔ Starting the journey
Key Insight
ARC is fundamentally different from most AI challenges.
It’s not about:
❌ memorizing patterns
It’s about:
✔ discovering rules
✔ applying logic
✔ adapting to new problems
What’s Next?
My next steps:
- Solve ARC tasks manually
- Build rule-based solutions
- Improve generalization step-by-step
💡 Final Thoughts
This is not just a competition — it’s a challenge to rethink how we approach AI.
And I’m just getting started.
Have you tried solving ARC tasks?
I’d love to hear how others approach reasoning problems like this.
Top comments (0)