Unverified50% confidenceFactExact time
从表征到有限词汇的映射过程本质上是一种有损编码（Lossy Encoding）
1
Sources
50%
Confidence
Long-term
Relevance
7/7/2026
First Seen
Sources
Why Large Language Models Are Starting to 'Invent Words': The Nature of LLM Language Boundaries
hackernewshackernews7/6/2026
Related Claims
Unverified文本编码器将提示词编码为固定维度嵌入向量，abliteration针对的是token生成过程中的拒绝行为，与嵌入向量输出情况不同77% similarUnverified实验数据显示，语法约束解码使小型语言模型生成Bash代码的语法正确率大幅提高，引号不匹配和结构不完整等错误几乎被完全消除76% similarUnverifiedConstrained Decoding was not invented by Outlines; its academic roots can be traced to formal language theory and compiler design, where early constrained generation was used in beam search for speech recognition and machine translation.73% similarUnverifiedThe Seq2Seq architecture compresses a source-language sentence into a vector representation via an encoder, and then a decoder progressively generates the target-language word sequence.73% similarUnverified程序化生成的纹理能被极端压缩是因为它们具有低柯尔莫哥洛夫复杂度，即存在简短程序可完整描述它们72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/168697API
curl https://kongchang.com/api/v1/knowledge/claims/168697MCP
get_claim(id=168697)