Unverified50% confidenceBenchmarkExact time
Without changing the prompt, the capabilities of the SymCE-trained model transfer to mainstream mathematical reasoning benchmarks such as GSM8K, MATH-500, and MMLU-college-math.
1
Sources
50%
Confidence
Long-term
Relevance
10/5/2026
First Seen
Sources
Related Entities
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77% similarUnverifiedThere is a qualitative gap between the currently popular evaluation benchmarks GSM8K and MATH and frontier research-level mathematical reasoning74% similarVerifiedTraining 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.71% similarUnverified大语言模型的训练目标是最大化下一个Token的预测概率而非最大化代码安全性，这两个目标在数学上不等价68% similarUnverifiedWidely cited LLM benchmarks such as MMLU, HumanEval, and GSM8K are predominantly single-turn and closed-form tasks, where the model produces output given a fixed input and is then scored.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/975478API
curl https://kongchang.com/api/v1/knowledge/claims/975478MCP
get_claim(id=975478)