Unverified50% confidenceFactExact time
IBM points out that more context does not mean better performance, and that performance may degrade as context expands, a phenomenon referred to as Context Rot
1
Sources
50%
Confidence
Long-term
Relevance
8/29/2026
First Seen
Sources
Related Entities
Related Claims
Unverified大上下文窗口会导致模型性能下降，让模型变笨，因此现代Harness的价值在于让模型精准构建所需上下文而非全部塞入68% similarUnverifiedA larger context window does not automatically lead to better model output results.66% similarUnverified更长的上下文窗口并未消除注意力稀释效应，反而带来更高的单次调用成本64% similarUnverified上下文窗口的大小不等于有效利用的能力，窗口越大指令衰减的潜在风险反而可能越突出64% similarUnverified模型性能的瓶颈不仅在于参数量，更在于表示的质量，嵌入空间坍缩的模型即使参数再多也无法充分发挥潜力63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/821007API
curl https://kongchang.com/api/v1/knowledge/claims/821007MCP
get_claim(id=821007)