Unverified50% confidenceFactExact time
GLM系列来自清华大学KEG实验室与智谱AI，采用不同于GPT自回归的自回归填空预训练目标
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
Related Claims
VerifiedGLM采用了独特的自回归填空预训练范式，与GPT的纯自回归和BERT的纯掩码语言模型都有所不同81% similarUnverifiedMultiple academic studies have shown that a considerable proportion of the high scores achieved by mainstream models such as GPT-4 and LLaMA on leaderboards including MMLU, HumanEval, and GSM8K stem from overlap between training set and test set content.70% similarUnverifiedGPT 系列、Llama 等模型的核心能力绝大部分来自预训练而非后续微调阶段70% similarUnverifiedGLM基于自回归空白填充预训练，融合了BERT的双向理解能力与GPT的生成能力，通过二维位置编码区分原始序列位置与空白内部位置69% similarVerifiedThe improvement in GPT-5.5's Pass@1 mainly stems from a larger pre-training data scale, more refined RLHF alignment, and improved instruction-following capability.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/516315API
curl https://kongchang.com/api/v1/knowledge/claims/516315MCP
get_claim(id=516315)