I thought about using GPT-3 to summarize big tables or spreadsheets. It works okay with the davinci engine and a couple of examples (spreadsheet, description). However I guess the results could be improved by fine-tuning the model with a lot of examples.
Has anybody already fine-tuned a model on tables or knows a dataset of spreadsheets + description?