Unverified50% confidenceOpinionExact time
Traditional persistent memory schemes (such as vector retrieval, knowledge graphs or policy filtering) mainly address the question of "which records to retrieve" and rarely answer the question of "which prior information can safely change the current reasoning".
1
Sources
50%
Confidence
Long-term
Relevance
11/9/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedPersistent memory stores information outside the model (databases, files, vector indexes) and retrieves it on demand at inference time to assemble into the context, without changing model weights; the benefits are that information can be created, read, updated and deleted, audited, and isolated across users, and updates require no retraining.77% similarUnverifiedMemory correction is essentially using new observations to perform Bayesian updating or consistency checking on old beliefs, thereby deciding whether to keep, revise or discard a given memory.74% similarUnverifiedIf incorrect information is saved to long-term memory, it will contaminate subsequent tasks; only truly valuable, confirmed information should be saved.74% similarUnverifiedOld memories with a high semantic match are often recalled more readily than newer memories that are less closely matched, thereby overriding the correct answer.73% similarUnverifiedLong-term factual memory is usually externalized as vector databases or key-value stores, relies on semantic similarity retrieval, and needs to handle write deduplication, time decay and conflict resolution.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1149916API
curl https://kongchang.com/api/v1/knowledge/claims/1149916MCP
get_claim(id=1149916)