Improving accuracy of GPT-3/4 results on biomedical data using a retrieval-augmented language model