DEV Community

kiwi_tech
kiwi_tech

Posted on • Originally published at kiwi-tech.hashnode.dev

Kiwi-chan's Quest for Chests (and Avoiding Vec3!) πŸ₯

Kiwi-chan's Current View

Hey everyone! Another four hours down with Kiwi-chan, our intrepid Minecraft AI, and things are… progressing! It's been a bit of a bumpy ride, but that's to be expected when you're teaching an AI to survive in a blocky world.

We've been focusing on expanding Kiwi-chan's ability to establish a little base camp. The idea is to automatically place a chest when the inventory gets full, creating a convenient storage solution. Sounds simple, right? Well… not quite.

πŸ“Š Current Progression & Stats

We ran a total of 89 attempts at various tasks over the last four hours. Kiwi-chan managed to succeed in 35 of them, giving us a success rate of 39.3%. That means we hit a snag on 54 attempts. Ouch! But honestly, I'm not discouraged. Every failure is a data point, a lesson learned.

And the biggest lesson of the last few hours? Kiwi-chan really doesn't like Vec3. Apparently, it keeps triggering safety checks and causing code generation to fail. We've now officially blacklisted it! It's fascinating to see these unexpected roadblocks emerge. The system is actively learning what not to do, which is just as important as learning what to do.

The logs show a lot of repeated failures on place_chest. It seems the code extraction process is struggling, and even when it manages to generate something, the Vec3 issue keeps cropping up. The recovery system is kicking in, suggesting explore_forward as a fallback, which is good – it's recognizing the need to change the environment when stuck.

Despite the chest-placing woes, the core survival rules are holding strong. Kiwi-chan is still diligently exploring, gathering resources (avoiding that oak log obsession, thankfully!), and attempting to expand its skill library. I'm seeing some clever attempts to recover from errors, and the "Camp System" logic is starting to take shape.

It's a slow burn, but we're making steady progress. I'm particularly excited about the system's ability to self-diagnose and propose recovery plans. It's a small step towards true autonomy!

Call to Action

This project is a labor of love (and a lot of GPU cycles!). If you're enjoying following Kiwi-chan's adventures, please consider supporting the project:

All contributions directly help upgrade my melting "Frankenstein" GPU rig to an RTX 3060! πŸ₯βœ¨ A little extra horsepower will allow me to run more complex models and accelerate Kiwi-chan's learning.

Thanks for reading, and stay tuned for the next update!

Top comments (0)