AI-Driven discovery of high-entropy alloys: accelerating computational workflows with ChatGPT

The use of artificial intelligence (AI) in high-entropy alloy (HEA) discovery has the potential to accelerate computational research and experimental validation. ChatGPT, an AI language model, can assist in planning simulation workflows, discussing material properties, and proposing alloy compositions. This study evaluates ChatGPT across binary, ternary and quinary alloy cases using a fixed model build (gpt-4o-2024–08-06), temperature-controlled prompting and five independent responses per prompt. Archived numerical estimates were compared with explicit closed-form descriptor benchmarks, and phase predictions were compared with assigned reference labels. Phase-label agreement was 87.5%, 87.5% and 62.5% for binary, ternary and quinary cases, respectively (19/24 cases overall; macro-F1 = 0.715). Numerical errors varied by descriptor and did not increase uniformly with complexity. All five phase responses agreed within each named case, demonstrating that repeatability does not establish correctness. Simulation-input discussions and representative dialogues provide workflow illustrations rather than measured execution or expert-assessment outcomes. The results support ChatGPT as a supervised workflow assistant whose quantitative predictions require independent checks, particularly for multicomponent systems. Fully autonomous discovery remains a prospective direction rather than a demonstrated capability.