Unverified50% confidenceFactExact time
在代码数据上训练的LLM不仅提升了编程能力，还显著增强了模型的逻辑推理能力，代码训练和逻辑推理能力之间存在正向迁移效应
1
Sources
50%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
VerifiedPre-training on code not only improves performance on programming tasks but also significantly enhances a model's logical reasoning capabilities, and has become standard practice for top-tier foundation models.85% similarUnverifiedThe inclusion of code data has been shown to significantly enhance a model's logical reasoning ability even on non-code tasks, a phenomenon known as the spillover effect of code data.79% similarUnverifiedLLM的代码生成能力源于对开源代码的统计模式学习，而非真正的逻辑推理78% similarUnverifiedTraining strategies for mathematical LLMs include continued pre-training on large-scale mathematical corpora on top of general pre-training, followed by supervised fine-tuning with chain-of-thought reasoning data, combined with reinforcement learning using process reward models.78% similarUnverified在代码语料比例更高的混合数据集上训练的模型，在数学推理基准（如MATH、GSM8K）上的表现显著优于纯自然语言训练的同规模模型，这一结论获斯坦福大学和MIT的多项研究证实75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/532698API
curl https://kongchang.com/api/v1/knowledge/claims/532698MCP
get_claim(id=532698)