An AI application may combine a user’s request with selected conversation history, documents, or other inputs. The response depends on which information is included and how the application presents it to the model.
Do not assume that every earlier file or message is available in full. Applications can limit, summarize, or select context. Check the documented behavior and restate a critical constraint when the task requires it.
Give the tool the relevant material and ask it to identify missing information rather than invent it. A clear input boundary makes a response easier to interpret and helps explain why two similar-looking requests can produce different results.
- Identify which material is actually included.
- Restate critical constraints when needed.
- Keep missing information explicit.
A small working example.
A draft generated from an incomplete conversation may omit an earlier decision. State the necessary context explicitly instead of assuming it remains available.
A note to keep beside it.
Separate what was generated from what was checked. A useful result includes enough context for a reader to recognize the decisions that still need human judgment.
Follow a related question
Write the question beside the idea.
Capture the question firstList the assumptions behind a prediction.
A thoughtful view of what comes nextKeep learning
Related background to continue exploring this subject.
Google: an introduction to language models NIST: AI risk management framework
