Unverified50% confidenceFactExact time
模型权重本身在推理时是冻结的，无法将新信息永久写入
1
Sources
50%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
Adaptive Recall：基于MCP协议的AI助手持久记忆方案解析
Related Claims
Unverified链式思维、少样本提示、自我一致性等提示词工程技巧无法让模型访问训练截止日期之后的信息，无法保证浮点运算的精确性，也无法处理模型从未见过的外部文件69% similarUnverifiedWhen the model itself lacks the capability to understand a specific problem, no amount of prompt engineering can compensate for this deficiency68% similarUnverified作者不再把KEEP决策当作模型改进的证明，在最近一次审计中模型在新数据上的对比结果是不确定的，模型被保留仅因没有足够强的理由回滚65% similarUnverifiedTraditional large models have a fundamental statelessness limitation: each inference is an independent forward-propagation process and cannot actively trigger any external operations.64% similarUnverifiedStatic benchmark scores cannot capture a model's actual capabilities in open-ended dialogue, reasoning, creative writing, and other complex tasks64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/503334API
curl https://kongchang.com/api/v1/knowledge/claims/503334MCP
get_claim(id=503334)