I've been building AI Lab, a browser-based incremental game where you run an AI research lab. You manage compute, train models, handle alignment decay, and try to deploy something useful before your revenue collapses.
The honest reason I built it: I wanted something that makes the capability vs safety tradeoff feel real, not just something you read about. Most explanations of why alignment matters are abstract. A game can make you feel the incentive to cut corners.
Phase 1 is live now and takes about 20 minutes. It covers the basics. From overfitting vs. underfitting to compute scaling, alignment decay, and real revenue consequences. Phase 2 will add rival labs, regulatory pressure, and dataset bias mechanics. I plan to introduce 5 phases through this rode.
It's a hobby project and my first shipped game. The mechanics are grounded in how ML training actually works (loss curves, training data quality, hardware scaling) but simplified enough for someone without an ML background to follow.
Play here: https://ai-lab-azure-pi.vercel.app
I'm very curious what this community thinks. Happy to hear what's missing or wrong.
