Realizing the cooking recipe of materials synthesis through large language models

Abstract

LLMs offer a promising and viable direction to convert materials synthesis descriptions into recipe-like outputs effectively preserving the order of synthesis steps. LLMs show true potential to guide experimental design using materials literature.

Descriptor Design for Perovskite Material with Compatible Molecules via Language Model and First-Principles

“…8,9 With the advent of a big language model 10−12 represented by ChatGPT and various BERT/GPT derivatives, natural language processing (NLP) has received increasing attention in the field of materials science. The emergence of such large language models has facilitated the exploration of vast scientific literature resources that are previously underutilized, 13,14 helping integrate scientific literature information into daily material research studies and thereby promote prediction of new materials. 15,16 Nevertheless, the practical implementation of big language models encounters formidable challenges, arising from substantial demands on computational resources, sheer volume of model parameters, significant time costs associated with development and deployment, and considerable expenses incurred in handling the computational complexities.…”

“…Over the past few years, artificial intelligence (AI) and data-driven approaches have been deployed throughout science and industry. − For materials science, AI has emerged as a powerful tool offering innovative solutions to long-standing challenges; for example, machine learning algorithms can analyze large material scientific data sets associated with chemical compositions, crystal structures, material processing details, and properties to accelerate the prediction and optimization of novel materials, leading to advancement of applications including energy storage, catalysis, and nanoelectronics. − In contrast, the traditional scientific paradigms exhibit trial-and-error limitations that may result in diminished predictive accuracy in experimental validation. , With the advent of a big language model − represented by ChatGPT and various BERT/GPT derivatives, natural language processing (NLP) has received increasing attention in the field of materials science. The emergence of such large language models has facilitated the exploration of vast scientific literature resources that are previously underutilized, , helping integrate scientific literature information into daily material research studies and thereby promote prediction of new materials. , Nevertheless, the practical implementation of big language models encounters formidable challenges, arising from substantial demands on computational resources, sheer volume of model parameters, significant time costs associated with development and deployment, and considerable expenses incurred in handling the computational complexities . Consequently, how to achieve swift and accurate material predictions via machine learning remains an urgent concern.…”