They shouldn’t be expected to improve in accuracy because of what they are and how they work. Contrary to what the average HackerNews seems to believe, LLMs don’t “think,” they just predict. And there’s nothing in them that will constrain their token prediction in a way that improves accuracy.