~27%. More than 1-in-4 answers wrong in real deployments. Not in a lab. In production. AI accuracy in a demo means nothing. The gap between benchmark and production is where AI projects die.
nicornle