| Abstract Scope |
We introduce CVD-Diamond Agentic AI, a framework built on knowledge retrieval from the single-crystal diamond CVD growth scientific literature, and showcase an assessment of its scientific reasoning and creativity. Human- and AI-generated questions, in both multiple-choice and descriptive formats, are used to assess our framework’s reasoning with metrics like fact recall and hallucination rate. We show that, for assessing scientific creativity, semantic diversity or information entropy metrics, often used to evaluate LLMs’ creativity, are insufficient. Instead, we develop a proof-of-time creativity benchmark framework in which an AI system, exposed only to knowledge from a pre-cutoff era, is evaluated on its ability to predict post-cutoff factual recalls, forecast future trends, and anticipate breakthroughs within its domain of expertise. Using the AI framework, we map a two-decade roadmap of diamond CVD synthesis and identify potential pathways for future exploration, including synthesis strategies relevant to wafer-scale diamond electronics and NV-center fabrication. |