Before drafting with AI, agree on what the words mean
A shared word can hide two different ideas. This framework asks people to surface that gap when working with AI. It offers examples, not a trial proving better results.
Beyond Prompting
What the source reports
This paper proposes a way to repair gaps in meaning between a person and an AI system. The problem can start with a word both sides use but understand differently. Its examples suggest checking two meanings of a key adjective before drafting. That move can make a hidden disagreement visible. The source offers a framework and examples. It does not report a trial showing that this step improves writing outcomes.
Original source
- Title
- Alignment Games: A Framework for Conceptual Repair in Human-AI Collaboration
- Author
- Subramonyam, Hari, Agrawala, Maneesh, Follmer, Sean
- Publication
- arXiv
- Date
- Monday, September 28, 2026