https://towardsdatascience.com/a-new-method-to-detect-confabulations-hallucinated-by-large-language-models-19444475fc7e