Jan 28, 2022
The paper studies prompting language models with examples that include intermediate reasoning steps.
Reasoning demonstrations improve performance on the tested arithmetic, commonsense, and symbolic tasks; generated explanations do not guarantee correctness.