I'd be impressed by an AI that can go against tons of training data by reversing the polarity of a common turn-of-phrase.

I was proud of that, and am slightly bummed to hear someone thinks it sounds AI-like. Also, I had a typo, its supposed to be "faint" (light) not "feint" (fake-out), though the latter worked out in context, and works by adding a homonym layer to the turn of phrase already turned on its head. 2 layers of smug, conceited cleverness - or 1 layer and a typo if you're not feeling charitable - that gets in the way of clear communication seems very human to me, YMMV.