Ask the user to generate content for each of the six stages. You help them translate it into a written format the advisor can scan in 60 seconds.
2.1 Set Context (1-2 sentences):
"Since we last met on [date], I've been working on [main thread]. Last meeting we agreed I'd focus on [X, Y, Z]."
2.2 Walk Through What You Did (bullets):
Concrete experiments or activities. Not "I worked on the model" — "I tried variants A, B, and C of the loss function and measured [metric] on [dataset]." Include visual evidence where it exists (plots, tables, code snippets).
2.3 Clarify What You've Solved:
What specific questions do you now have answers to? What insights did you gain? What works and what doesn't?
2.4 Identify What Remains Unsolved:
The most important section. What are you stuck on? Where exactly? What have you tried that didn't work? Be specific — "the training is unstable" is not enough. "When batch size drops below 32, gradient norm explodes after the 3rd epoch, and I've tried gradient clipping at values [a, b, c] without success" is useful.
2.5 Present Your Next Steps:
What's your concrete plan going forward? What timeline? This matters because it shows you've thought about the path, not just the problem.
2.6 Questions Ranked by Priority:
3-5 questions, explicitly ranked. The top question is the one thing you most want their input on. If you only got to discuss one thing, this would be it.
引导用户为六个阶段生成内容，帮助他们转化为导师能在60秒内快速浏览的书面格式。
2.1 设定背景（1-2句话）：
"自上次[日期]会面以来，我一直在推进[主线任务]。上次会面我们商定我将专注于[X、Y、Z]。"
2.2 梳理已完成工作（项目符号列表）：
具体的实验或活动内容。不要写“我在研究模型”——要写“我尝试了损失函数的A、B、C三种变体，并在[数据集]上测量了[指标]。”如有可视化证据（图表、表格、代码片段），请一并附上。
2.3 明确已解决问题：
你现在找到了哪些具体问题的答案？获得了哪些见解？哪些方法可行，哪些不可行？
2.4 识别待解决问题：
这是最重要的部分。你卡在了哪里？具体是什么问题？你已经尝试过哪些无效的方法？要具体——“训练不稳定”不够详细。“当批量大小低于32时，梯度范数在第3个epoch后爆炸，我尝试了[a、b、c]值的梯度裁剪，但均未成功”才是有用的表述。
2.5 提出下一步计划：
你未来的具体计划是什么？时间线如何？这很重要，因为它表明你已经思考了解决路径，而不只是关注问题本身。
2.6 按优先级排序的问题：
3-5个问题，明确排序。排名第一的问题是你最希望获得导师意见的内容。如果只能讨论一件事，那就是这个问题。