Unverified50% confidenceSolutionExact time
即使输入完全无意义的假词，文生图模型也不会拒绝处理，而是通过BPE拆解为已知子词并生成看起来连贯的角色图像
1
Sources
50%
Confidence
Long-term
Relevance
8/5/2026
First Seen
Sources
Related Claims
UnverifiedEarly diffusion models often generated garbled pseudo-text when rendering words within images.73% similarVerifiedThe byte pair encoding (BPE) or WordPiece algorithms widely used in modern large models are essentially data-driven morphological segmentation methods.69% similarUnverifiedMainstream large models generally use Byte Pair Encoding (BPE) or its variants (such as SentencePiece and tiktoken) for tokenization.68% similarVerifiedDiffusion models treat text as visual texture rather than semantic symbols, causing generated characters to often look similar but lack correct meaning—a structural defect in their text rendering.66% similarUnverified大语言模型会编造看似真实、实则不存在的文献引用，这种现象被称为「引用幻觉」（citation hallucination）66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/691017API
curl https://kongchang.com/api/v1/knowledge/claims/691017MCP
get_claim(id=691017)