Unverified50% confidenceFactExact time
在大语言模型底层，系统提示词、用户输入和模型输出都是同一个token序列的不同片段，依靠特殊分隔符和训练时植入的行为模式维持边界，而非硬件或操作系统级别的隔离
1
Sources
50%
Confidence
Long-term
Relevance
7/12/2026
First Seen
Sources
Related Claims
Unverified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75% similarUnverified同一大语言模型在不同硬件上可能产生不同推理结果，甚至同一硬件上重复运行也可能出现微小差异75% similarUnverifiedThe Coze platform has integrated multiple large language models at its underlying layer, including the Doubao large language model.74% similarUnverifiedThe root cause of prompt injection lies in the fact that LLM architecture encodes system prompts, user inputs, and externally retrieved content uniformly into the same continuous token sequence, making it impossible for the model to distinguish trusted instructions from untrusted data at the semantic level.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/491525API
curl https://kongchang.com/api/v1/knowledge/claims/491525MCP
get_claim(id=491525)