“…Notably, these solvents result in significant aqueous stability improvement, displaying retention indexes (ratios between aqueous photocurrent at 200 s and the photocurrent at 1 s in the photoelectrochemical experiment where the sample is submerged in water) of 98.83, 97.37, 86.34, 82.58, 88.58, and 80.28%, compared to the control sample (stability is only 36.14% for the sample fabricated via the classic “DMSO + DMF” solvents). In particular, the most effective quinary solvent system identified by the “textual descriptors + DFT descriptors + GA” model is “DMSO + DMF + hexane + octane + toluene”, displaying a significantly enhanced stability from 36.14 to 98.8% for the halide perovskite film, which is noteworthy considering the rapid dissolution in water for the CH 3 NH 3 PbI 3 film. − The enhancement of the photocurrents is owing to the versatility of the textual descriptors and DFT descriptors integrated into the genetic process; this may comprehensively evaluate materials compositions, structures, processing, properties, and performance from language model-based ontology and atomic/electronic levels to deliver more powerful predictions. In this way, scientific insights are condensed from millions of literature articles into predictive mathematical formulas through genetic iteration, and more information provided by DFT and NLP is leveraged to advance the ability to forecast material properties.…”