Unverified50% confidenceFactExact time
不同embedding模型生成的向量在语义空间中不可直接比较，一旦更换模型，向量库中所有历史向量都必须用新模型重新生成
1
Sources
50%
Confidence
Long-term
Relevance
8/26/2026
First Seen
Sources
Related Entities
Related Claims
Unverified词表大小直接决定模型嵌入层和输出层的维度，分词器变更意味着模型无法简单从前代继承权重72% similarUnverifiedAfter replacing the embedding model, the entire knowledge base must be re-vectorized.68% similarUnverified裁判模型应在模型版本升级和输入分布漂移这两个触发条件下强制重新校准66% similarUnverified提示词并没有真正提高模型的能力，它只是给模型提供了更多相关信息64% similarUnverifiedOn this specific tool schema, newer Anthropic models actually perform worse than older models.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/802096API
curl https://kongchang.com/api/v1/knowledge/claims/802096MCP
get_claim(id=802096)