Unverified60% confidenceFactTime unknown
LLMs are biased toward instructions located at the edges of the prompt (the very beginning and the very end).
2
Sources
60%
Confidence
Long-term
Relevance
6/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe underlying working mechanism of LLMs is to predict the next token.73% similarUnverifiedSemantics-driven LLM policies produced the strongest 'core–tail' retrieval traces, compatible with a truncated power law.72% similarUnverified大多数LLM会优先响应距离最近、最明确的上下文指令69% similarUnverified使用 LLM 编程时，提供正确的上下文比打磨提示词措辞更能提升输出质量68% similarUnverifiedWhen an LLM consistently misidentifies a command as a different tool, pre-dispatch hooks should be used to inject explicit constraints rather than relying solely on prompt engineering.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/37941API
curl https://kongchang.com/api/v1/knowledge/claims/37941MCP
get_claim(id=37941)