Unverified50% confidenceFactExact time
Switching embedding models means all historical data must have its vectors recomputed, resulting in extremely high migration costs.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe full rebuild approach is simple to implement and leaves no dirty data, but is costly at the million-vector scale and cannot meet near-real-time update requirements, making it suitable for scenarios where data changes infrequently.66% similarUnverified用新数据更新模型时可能破坏原有能力，即灾难性遗忘，工业界通过失败案例库、数据增强与混合回放技术来整合新数据62% similarUnverifiedANCE通过定期刷新ANN索引从全库动态检索难负样本，但需重新编码全量文档并暂停梯度更新，代价高昂58% similarUnverifiedRAG converts historical interaction data into high-dimensional vectors via embedding models and stores them in a vector database, injecting context at inference time rather than modifying model parameters, which avoids the risk of catastrophic forgetting.57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/900839API
curl https://kongchang.com/api/v1/knowledge/claims/900839MCP
get_claim(id=900839)