Tag
This paper argues that agents should help users construct preferences rather than assuming well-formed ones, proposing the CoPref model and CoShop benchmark. Evaluations show even frontier models achieve only 56% accuracy due to poor preference expansion.
A reflection on how early AI chat interfaces resemble command-line interaction patterns, and a speculation about when a GUI-like paradigm shift will emerge for AI interactions, where the AI can directly observe and act on the user's context.