correcting large language model errors in thesis