We hypothesize that this decline is due to the fact that current LLMs are not capable of genuine logical reasoning," the researchers write. "Instead, they attempt to replicate the reasoning steps observed in their training data." https://lnkd.in/eXE64j4M