Prompt Engineering

A vague instruction leaves the model to guess what matters. I find it more useful to identify the audience, the information available, and the constraints before asking for an answer.

Examples help when a format or distinction is difficult to describe. They are a way to communicate expectations, rather than a guarantee that the result will be correct.

For a larger task, I want intermediate results that I can inspect. A proposed structure, a small change, or an explanation of assumptions is easier to judge than a long answer that hides several decisions.

That keeps the work connected to the original requirement as it develops.

I do not treat a confident response as evidence. With generated code or factual claims, I still need to check the relevant sources and behavior. This is especially relevant to the AI-assisted implementation documented in my Nutras website work.