Kurzweil’s 2005 defense of speech recognition got every claim right and the architecture wrong: HMMs lost to transformers, but the destination arrived bigger than promised.