Unverified50% confidenceFactExact time
The same prompt string does not equal the prompt the model actually sees, because chat templates inject system text and some SDKs add defaults.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedLLM models do not directly read characters or words; the tokenizer splits text into chunks according to a fixed vocabulary, and what the model sees is a string of chunk IDs.69% similarUnverifiedModels labeled with Instruct or Chat are instruction-tuned conversational versions, while those without labels are Base foundation models68% similarVerifiedThe 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.68% similarUnverifiedEmbedding models (such as text-embedding-3-small/large and text-embedding-004) and the LLM responsible for conversation are two independent models.68% similarUnverifiedSystem prompts are not hard code constraints, but more like a piece of natural language text with a relatively high priority.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1065391API
curl https://kongchang.com/api/v1/knowledge/claims/1065391MCP
get_claim(id=1065391)