Reasoning & Test-Time Compute · 2025
LIMO: Less is More for Reasoning
Demonstrates that ~800 well-chosen demonstrations elicit competition-level math reasoning, arguing reasoning is latent in pretraining and unlocked by minimal cognitive templates rather than large post-training data.
Editorial record
Plain-language summary
Demonstrates that ~800 well-chosen demonstrations elicit competition-level math reasoning, arguing reasoning is latent in pretraining and unlocked by minimal cognitive templates rather than large post-training data.
Knowledge graph
Relationships
Descendants
Conceptual ancestorEvidence: Strongly supported
Chain-of-Thought Prompting Elicits Reasoning
Builds on the reasoning lineage in the archive
freshness sweep 2026
Source record
Provenance
- Record ID
- P-639
- Record created
- 2026-07-13
- Last reviewed
- 2026-07-14
- Record version
- 2
- https://arxiv.org/abs/2502.03387
- arXiv:2502.03387
Citation caveat: Citation metadata is approximate and marked unverified in the source dataset.