Unverified50% confidenceTradeoffExact time
Word counting should be implemented in code as a unit-testable function, something that a prompt cannot achieve
1
Sources
50%
Confidence
Long-term
Relevance
10/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWhen the subject of a check is LLM-generated text, the reasons for a check passing or failing are not written in the code but depend on factors such as model output and prompt constraints that cannot be exhaustively determined through static reading.70% similarUnverifiedFilter/detection-based defenses (such as scanning for phrases like 'ignore previous instructions') are structurally doomed to fail, because natural language has no syntax that distinguishes instructions from data.64% similarUnverified规则层通过关键词、正则表达式、状态机处理表意固定无歧义的高频指令，可实现毫秒级路由且零算力消耗63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1000017API
curl https://kongchang.com/api/v1/knowledge/claims/1000017MCP
get_claim(id=1000017)