Tag
Task Model Induction (TMI) is a method that induces structured task models from raw computer-use traces, improving agent performance by 30% and enabling auditable, reusable skills for real-world workflows.
This paper introduces PaperPilot, a multi-turn literature search agent that constructs executable DAGs of search operators and refines workflows through user feedback, achieving significant improvements in retrieval metrics over baseline models.
Introduces Offloading Score, a measure of cognitive effort offloaded to AI, and finds that high reliance is not inherently harmful when users are engaged.