A recent review preprint posted on Preprints.org by researchers Taghipour et al. explores using artificial intelligence, including foundation models and machine learning, to integrate heterogeneous geoscientific data for mineral exploration and prospectivity mapping. The paper has not undergone peer review. The review outlines a mineral exploration pipeline with three stages: Stage 1 (Area Selection) uses regional-scale (∼100s km) geology, satellite imagery, and coarse geophysics; Stage 2 (Target Generation) operates at the district scale (∼10s km) and relies on data integration, processing raw data such as geological maps, geophysical surveys, geochemical assays, remote sensing, and unstructured text into stacked ‘evidence layers’ for machine learning models; Stage 3 (Target Testing) involves prospect-scale (∼1 km) physical validation, such as drill core sampling. The escalating global demand for minerals, especially critical minerals, intensifies the need to discover new deposits. Traditional expert interpretation is time-consuming and difficult to scale over large terrains. The heterogeneous nature of geoscience datasets and complex geological spatial dependencies present challenges for AI, including scarcity of labeled positive samples, multimodal data formats, and spatial non-stationarity. Mineral exploration is naturally a positive-unlabeled learning problem, and spatial autocorrelation can yield overly optimistic results with random splits. No single foundation model currently spans all exploration modalities.