Feed the Whole Folder: Using Million-Token Context Windows Well

Gemini, DeepSeek and Kimi now support very large context windows (hundreds of thousands to a million-plus tokens), so you can drop in a whole batch of documents, code, or chat logs at once. Use it like this: 1) paste or upload everything and start with "Here are N sources — read all of them first, don't answer yet"; 2) have the model output a list of the sources with key points each, proving it actually read them; 3) then ask cross-document questions (e.g. "which ones contradict each other", "consolidate conclusions scattered across them"); 4) require every claim to cite which source number it came from so you can verify. My sources are: [paste/upload], and my question is: [question].

现在 Gemini、DeepSeek、Kimi 都支持超长上下文（几十万到上百万 token），可以一次把整批文档、代码或聊天记录喂进去。请这样用：1）把所有材料贴进来或上传，并在开头说明「以下是 N 份资料，先通读，不要急着回答」；2）让模型先输出一份「材料清单 + 各自要点」证明它真的读完了；3）再提出跨文档问题（如「哪几份互相矛盾」「把分散在多处的结论汇总」）；4）要求每个结论都标注来自第几份材料，便于核对。我的材料是：[粘贴/上传]，我要回答的问题是：[问题]。