Google 刚刚释放了Gemini 2.0 flash Thinking exp 1.21版本,在LLM arena上感觉又刷榜了。最近的Thinking 模式都是增强了reasoning过程,感觉非常适合用于做研究。
我之前都是用1.5 pro,使用2M上下文传PDF上去,然后在System prompt里填入修改后的Thinking clude的prompt,来达到类似thinking的模式。
现在flash 2.0的上下文提升了 1M输入,65k输出。非常适合用于快速阅读论文写笔记,提高阅读效率。下面我尝试了一下自己的使用场景:
- Zotero保存相关条目,提取本地PDF
- 上传PDF到AI studio,调用模型问答
- 输出AI写的笔记,copy到Zotero的笔记里
- 对照笔记和论文快速浏览,寻找自己感兴趣的点,在Zotero做记录
- 遇到不会的继续在AI studio上提问
[链接](Wen et al. - 2024 - From Supervised to Generative A Novel Paradigm for Tabular Deep Learning with Large Language Models.pdf - Google 云端硬盘, https://aistudio.google.com/app/prompts?state={"ids":["1Y6xhuM66gfnQ2Qlkfh79blv1IAd176YB"],"action":"open","userId":"104105819556530011190","resourceKeys":{}}&usp=sharing)
上面是我的一个prompt 案例,可以看到质量还是非常高的。
我使用的笔记模板:
Please tell me about this paper. Filled in this form:
{
## 📜 Research Core
* * *
> Tips: What was done, what problem was solved, innovations and shortcomings?
### ⚙️ Problem
### 💡 Innovations
### 🧩 Shortcomings
## 🔁 Research Content
* * *
### 💧 Data
### 👩🏻💻 Method
### 🔬 Experiment
### 📜 Conclusion
## 🤔 Personal Summary
* * *
> Tips: What aspects did you question, how do you think it can be improved?
### 🙋♀️ Key Records
### 📌 To be resolved
### 💭 Thought Inspiration
}