Armchair backfitting, but: If there were language samples that had been weighted as both persuasive and rare, this model appears to be resorting to them when i

Armchair backfitting, but: If there were language samples that had been weighted as both persuasive and rare, this model appears to be resorting to them when it was challenged to provide a more consistent answer to the sunflower seed oil question. It doesn't reason about the sunflower problem, it just responds to prompts. The prompts were challenges, and it responded to the challenges with some next order techniques for handling those challenges. Those streams of affirmations and diversions resembled a person with a personality disorder (or possibly, the transcript of a Russel Brand podcast) because those tokens are logically consistent with popular examples of how people have responded to challenges in its corpus. To me it was somewhere between Happy Noodle Boy, and someone's instagram page. I've had less coherent interactions with people.