Your AI Agent Isn’t Stupid — You’re Just Feeding It Too Much Context
中文摘要
Chroma 测试 18 种模型发现，AI 并非无能，而是因上下文过多导致性能下降，行业对大上下文窗口的处理方式完全错误。
English Summary
Chroma tested 18 models, proving that excessive context hinders AI performance and the industry's approach to large context windows is fundamentally backwards.
原文节选
Chroma tested 18 frontier models and proved the tech industry is treating the 1M-token window completely backwards. Continue reading on Medium »