Unverified50% confidenceFactExact time
Google's Minerva model improved its mathematical problem-solving ability by further pre-training on a large volume of scientific and technical papers and LaTeX source code.
1
Sources
50%
Confidence
Long-term
Relevance
8/14/2026
First Seen
Sources
Related Claims
Unverified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72% similarUnverified斯坦福大学与谷歌联合研究表明，多步推理、代码调试、跨语言翻译等任务在特定模型规模阈值处呈现明显能力跃升70% similarVerifiedThe Chain-of-Thought prompting paper published in 2022 by Jason Wei et al. of Google Brain was the first to systematically demonstrate that adding step-by-step reasoning examples significantly improves the accuracy of large models such as PaLM on mathematical reasoning tasks69% similarUnverifiedModern large models significantly improve performance on complex mathematical problems by learning to progressively decompose problems and verify intermediate steps during training (Chain-of-Thought reasoning mechanism).69% 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.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/748113API
curl https://kongchang.com/api/v1/knowledge/claims/748113MCP
get_claim(id=748113)