The paper finds that as exemplar count increases, LLMs shift from pretrained priors to in-context signals, and long chain-of-thought prompts induce longer reasoning chains that often improve accuracy.
In: The Eleventh International Conference on Learning Representations (2023)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Rethinking the Chain-of-Thought: The Roles of In-Context Learning and Pre-trained Priors
The paper finds that as exemplar count increases, LLMs shift from pretrained priors to in-context signals, and long chain-of-thought prompts induce longer reasoning chains that often improve accuracy.