Researchers at Germany's Friedrich Schiller University Jena and the Swiss Federal Institute of Technology adapted GPT-3, the large language model behind ChatGPT, to answer questions about chemical compounds or materials.
They gathered information about similar compounds or materials, formatted the data in the form of up to 30 questions and answers, and had it added to GPT-3's training set.
GPT-3 was then able to answer predictive questions about the original compound or material despite it not being included in the input data.
[
add
]
[
|
|
...
]