Hacker News
new
top
best
ask
show
job
Is Chain-of-Thought Reasoning of LLMs a Mirage? A Data Distribution Lens (2025)
(
arxiv.org
)
1 point
by
SomaticPirate
3 hours ago
1 comment
SomaticPirate
3 hours ago
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.