[ijk]

I find that people generally vastly underestimate the degree to which LLMs specifically are just mirroring your input back at you. Any time you get your verbatim words back, for example, you should be skeptical. Repeating something word for word is a sign that the model might not have understood the input well enough to paraphrase it. Our expectations with humans go in the opposite direction, so it's easy to fool ourselves.