Influence traces show LLM reasoning on simple math is driven by documents that demonstrate procedures, not by documents containing the answers, suggesting a generalization strategy unlike retrieval.
23 Table 9: Example from the reasoning set that requires solving for x in a linear equation
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Procedural Knowledge in Pretraining Drives Reasoning in Large Language Models
Influence traces show LLM reasoning on simple math is driven by documents that demonstrate procedures, not by documents containing the answers, suggesting a generalization strategy unlike retrieval.