Unverified50% confidenceOpinionExact time
Metalinguistic prompting requires the model to switch from a text-generating mode to a text-evaluating mode, a mode shift that is not directly covered by the pretraining objective.
1
Sources
50%
Confidence
Long-term
Relevance
10/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMainstream text embedding models are typically trained based on contrastive learning, mapping semantically similar text pairs to nearby positions in vector space and pushing unrelated text pairs further apart.75% similarUnverified提示词注入分为直接注入和间接注入两种形式，因语言模型无法区分数据与指令，凡进入上下文窗口的文本都可能被执行74% similarUnverifiedExisting dynamic graph architectures typically initialize text feature vectors with a pretrained language model and keep them static during training in order to preserve semantic structure.74% similarUnverified模型的格式方言源于指令微调阶段训练数据的格式差异，被模型深度内化形成强烈格式偏好，难以通过运行时提示词完全覆盖73% similarUnverified多模态模型的文本识别并非通过独立OCR子模块实现，而是作为视觉-语言对齐训练的产物内嵌在模型权重中，图片文字与普通文本共享同一套语义处理管线73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1115550API
curl https://kongchang.com/api/v1/knowledge/claims/1115550MCP
get_claim(id=1115550)