Unverified50% confidenceFactExact time
由于编码是一条轨迹，改变token顺序会改变终点，使模型天然对词序敏感，区别于简单向量平均如CBOW求和
1
Sources
50%
Confidence
Long-term
Relevance
7/13/2026
First Seen
Sources
No MLP or Attention Required: How Point-Attractor Dynamics Learn Word Embeddings
redditr/learnmachinelearning7/10/2026
Related Claims
UnverifiedLarge language models predict the next token one at a time through an autoregressive generation mechanism, and are naturally inclined to generate more complete rather than more concise code.71% similarUnverified用户输入的每个词都会成为模型的上下文，改变模型接收到的数学输入，进而影响解码阶段的概率分布和输出内容71% similarUnverified语言的语法结构越简单、惯用模式越统一，大模型在每一步预测中面临的困惑度越低，生成正确序列的概率越高69% similarUnverified模型基于前文预测后文，Prompt中任何一个字的变化都会改变后续生成概率，而标点符号的影响权重较小69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/496329API
curl https://kongchang.com/api/v1/knowledge/claims/496329MCP
get_claim(id=496329)