Unverified50% confidenceFactExact time
CoDR (Confidence Drift Remasking) is a training-free, sampler-agnostic method for correcting early errors in diffusion language models.
1
Sources
50%
Confidence
Long-term
Relevance
10/8/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDuring the decoding process of masked diffusion language models, once a token is committed to a masked position it is generally irreversible and cannot be corrected.65% similarUnverifiedIntroducing randomized paraphrasing or standardizing outputs can disrupt hidden encodings and strip out smuggled information.63% 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63% similarUnverifiedLarge models perform better when processing decompiled pseudocode (such as C-like pseudocode output by Ghidra), because the training corpus contains a large amount of high-level language code, while their ability to understand raw assembly instructions is relatively limited.63% similarUnverified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）63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/986619API
curl https://kongchang.com/api/v1/knowledge/claims/986619MCP
get_claim(id=986619)