ML based voice interfaces have been kicking around in hype cycle for 6 or 7 years. Its is a painstaking task to build a decent one, and, lets face it users have stopped expecting much so it is hard to get excited about the return on effort needed to build them. This talk implements one live onstage in 20mins with a GPT rather than hand crafted dialog training and tries to figure out how much of a game-changer this approach will be.