百万上下文不等于一份好简报
作者实测把整份长资料丢给 Astra 只得到平静的总结而非决策，并给出四条写简报的替代做法。
GPT 6 Astra can hold about a million tokens. That is still not a brief I dropped a long folder into GPT 6 Astra in the new ChatGPT and asked for “the point” It gave a calm summary of everything. It did not give a decision. The window was full. The job was empty That is the habit around this model. People paste the whole repo, the whole PDF, the whole week of notes, and call it prompting. Astra can hold it. Holding is not choosing What I do in GPT 6 Astra: Name the owner of the answer in line one Name what to ignore Paste only the pages that can change the answer If the file is huge, I write the question first and let Astra ask for the page it needs GPT-6 Astra is built for long work. A brief is still a stop condition. If you cannot say what would make the reply finished, you did not prompt Astra. You rented its window
这条是引用着另一条写的：「实测 Astra 自主性：会自己开工也会越界」