Unverified50% confidenceFactExact time
当前的LLM一旦完成训练其参数即被冻结，只能通过微调或RAG等外部机制来更新
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/28/2026
First Seen
Valid until: 11/26/2026
Sources
Related Entities
Related Claims
UnverifiedLLM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71% similarUnverifiedLLM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70% similarUnverified微调通过在私有数据上继续训练来更新模型权重，RAG则在不修改模型的前提下通过检索外挂知识库增强回答69% similarUnverified在LLM引用行为测试中固定随机性需要同时设置temperature和top-p参数并在支持时固定随机种子66% similarUnverifiedPure generative models mix all training data before producing output, making it impossible for users to trace the evidence chain behind any single conclusion—a limitation compared to the RAG retrieval mechanism.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/815980API
curl https://kongchang.com/api/v1/knowledge/claims/815980MCP
get_claim(id=815980)