In 2024 a machine settled a value of the busy beaver function no human could ever compute by hand — vindicating Kurzweil’s bet on the limits of computation, and the wall that stops both sides.
Kurzweil bet AI would advance by machine breeding. Every fact he cited was right; the forecast lost to gradient descent by a thousandfold — then quietly came back married to LLMs.
Batch 54: ten claims from 2005 about AI being everywhere, plus one forward bet on VR. The 2005 examples were all real. None of them was the mechanism that mattered.
He called the 2029 Turing test right and the substrate wrong. Ten predictions on strong AI, uploading, and self-replicating intelligence, scored against April 2026 reality.
Genetic algorithms were supposed to deliver human-level AI by 2025. Transformers did — on time, and nothing like the path Kurzweil described.