Happy Friday, fellow tech masochists and AI enthusiasts!

If you’ve been following the journey of Kiwi-chan , you know the drill: we are building a fully autonomous Minecraft agent that runs entirely on local hardware. No cloud APIs, no latency, no privacy leaks. Just raw, unadulterated inference power and a bot that tries to survive in a blocky void.

The last 4 hours have been a wild ride of hallucinations, biome betrayals, and the glorious transition to Qwen 35B . Let’s dive into the logs, the stats, and the sheer chaos of watching an AI learn to craft a door without a crafting table.

The Stats: A Love Letter to Inefficiency

First, let’s look at the numbers. Because in DevOps, if you aren’t crying over your metrics, are you even trying?

[Past 4 Hours Statistics]

Total Actions: 3488

Successes: 1640

Success Rate: 47.0%

Yes, you read that correctly. 47.0% .

Is that bad? In a traditional software context, sure. You’d get fired.…