Artificial intelligence-powered large language models (LLM) need to be trained on massive datasets to make accurate predictions—but what if researchers don't have enough of the right type of data?