Is Chain-of-Thought Reasoning of LLMs a Mirage? A Data Distribution Lens (2025)arxiv.org 1 pointSomaticPirate4 months ago1 commentSaveHideCopy link On HNComments−SomaticPirateOP4moA useful reminder that just because an LLM "appears" to be thinking and reasoning that it likely isn't. If it hasn't seen something similar or "in distribution" before, then it typically doesn't perform well.
Comments
A useful reminder that just because an LLM "appears" to be thinking and reasoning that it likely isn't. If it hasn't seen something similar or "in distribution" before, then it typically doesn't perform well.