Unverified50% confidenceFactExact time
微调狭窄专业语料存在灾难性遗忘和过拟合专业域两大已知ML问题
1
Sources
50%
Confidence
Long-term
Relevance
7/22/2026
First Seen
Sources
Claude Code高阶实战：打造顶级AI开发流
Related Claims
UnverifiedThe drawback of the context condensation strategy is that it may lose details deemed unimportant, leading to misunderstandings or repeated mistakes in subsequent conversations.68% similarUnverifiedThe performance gap among mainstream frontier models on standard tasks such as general question answering, text summarization, and code generation is narrowing, a phenomenon referred to as capability homogenization.67% similarUnverified文档分块需权衡片段大小与语义完整性，过小则语义不完整，过大则引入噪音67% similarUnverified在长对话中早期的错误尝试和调试过程会持续占据上下文窗口，造成上下文污染并干扰后续推理准确性66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/587663API
curl https://kongchang.com/api/v1/knowledge/claims/587663MCP
get_claim(id=587663)