Democratizing Foundations of Problem-Solving with AI: A Breadth-First Search Curriculum for Middle School Students

Collection of arXiv papers (Apr 3 2026) with several security-relevant findings: (1) A practical two-stage, training-free LLM-based differentially-private synthetic data generation (DP-SDG) workflow is proposed for educational real-world data; it lowers engineering cost and matches DL baselines but on-demand (non-DP) real-data validation produces measurable privacy leakage and case-study validation of synthetic findings was low (36%), raising re-identification/overfitting and epistemic-risk concerns for shared datasets. (2) Voice-cloning work shows clones can increase intelligibility and that,