This article explores the application of large language models (LLMs), specifically GPT-3, in predictive chemistry and materials science. The study demonstrates that GPT-3, when fine-tuned to answer chemical questions in natural language, can perform comparably or even outperform conventional machine learning models, particularly in low-data scenarios. The model's ability to adapt to various tasks, including classification, regression, and inverse design, is highlighted. The research shows that GPT-3 can predict molecular properties, reaction outcomes, and even design new materials by answering questions in natural language. The model's performance is validated across multiple datasets and applications, including predicting the phase of high-entropy alloys and designing molecular photoswitches. The study also emphasizes the model's ability to generate novel molecules with desired properties, even when the training data is limited. The results suggest that LLMs can serve as a powerful tool for chemical and materials science, offering a simple and effective approach for tasks that require minimal data. The study further discusses the potential of using LLMs for inverse design, where the model can generate molecules with specific properties by answering questions in natural language. The research underscores the importance of leveraging the vast knowledge encoded in LLMs to advance predictive chemistry and materials science.This article explores the application of large language models (LLMs), specifically GPT-3, in predictive chemistry and materials science. The study demonstrates that GPT-3, when fine-tuned to answer chemical questions in natural language, can perform comparably or even outperform conventional machine learning models, particularly in low-data scenarios. The model's ability to adapt to various tasks, including classification, regression, and inverse design, is highlighted. The research shows that GPT-3 can predict molecular properties, reaction outcomes, and even design new materials by answering questions in natural language. The model's performance is validated across multiple datasets and applications, including predicting the phase of high-entropy alloys and designing molecular photoswitches. The study also emphasizes the model's ability to generate novel molecules with desired properties, even when the training data is limited. The results suggest that LLMs can serve as a powerful tool for chemical and materials science, offering a simple and effective approach for tasks that require minimal data. The study further discusses the potential of using LLMs for inverse design, where the model can generate molecules with specific properties by answering questions in natural language. The research underscores the importance of leveraging the vast knowledge encoded in LLMs to advance predictive chemistry and materials science.