SlideChain: Verifiable Semantic Provenance for Educational Content

This paper addresses the critical issue of trust and reproducibility in AI-generated educational content, particularly in STEM fields. It introduces SlideChain, a blockchain-based framework to ensure the integrity and auditability of semantic extractions from lecture slides. The work's significance lies in its practical approach to verifying the outputs of vision-language models (VLMs) and providing a mechanism for long-term auditability and reproducibility, which is crucial for high-stakes educational applications. The use of a curated dataset and the analysis of cross-model discrepancies highlight the challenges and the need for such a framework.