The point of this paper is not to claim that LLMs are dumb. It's applying rigor to the notion that they "reason" in a way that's fundamentally non human. They'r

The point of this paper is not to claim that LLMs are dumb. It's applying rigor to the notion that they "reason" in a way that's fundamentally non human. They're more like search engines than abstract reasoning machines. There will always be problem structures that are too different from anything in the training set for them to solve