Inference to the Best Explanation in Large Language Models