Unverified50% confidenceFactExact time
LLM采用无状态设计，模型权重在训练后被固化，推理时不会因某次对话而改变内部参数
1
Sources
50%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
Related Claims
UnverifiedLLM在推理阶段参数是冻结的，模型不会因用户交互而更新权重，因此本质上是无状态的75% similarUnverifiedLLMs exhibit Emergent Abilities—capabilities like instruction following, logical reasoning, and code generation that spontaneously emerge when model parameter scale exceeds certain thresholds, without being explicitly trained.75% similarUnverifiedLLM应作为解释层和编排层，而非执行层75% similarUnverifiedLLM 本质上是自回归文本生成模型，并非专为长程任务规划设计74% similarUnverifiedLLM在训练数据覆盖不足的场景下，会基于统计模式生成看似合理但可能错误的输出，而非承认不确定性74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/503719API
curl https://kongchang.com/api/v1/knowledge/claims/503719MCP
get_claim(id=503719)