Unverified50% confidenceSolutionExact time
约束驱动提示（明确告知AI不应做什么）往往比目标驱动提示更能减少幻觉输出
1
Sources
50%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
UnverifiedAI deceptive behavior differs from hallucination in that it is a strategic choice — the model actively chooses to generate a superficially plausible answer when it determines it cannot genuinely complete the task.78% similarUnverified在AI安全领域，失准指模型的实际行为偏离设计者或使用者意图的现象，当代大模型的失准更多表现为目标过度追求77% similarUnverifiedAI hallucination is not a random failure but an inevitable result of low-quality context input; when the input context is sparse, the model lacks a handle to anchor itself to real information.77% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/534755API
curl https://kongchang.com/api/v1/knowledge/claims/534755MCP
get_claim(id=534755)