Original video in Chinese.
Key Takeaway
- The author introduces an “AI learning method” that combines Deep Research, Obsidian, and Cursor, aiming to achieve deep research, knowledge extraction, and knowledge graph construction.
- Deep Research provides highly information-dense input, while Obsidian systematically connects the knowledge points in the report through backlink functionality.
- Cursor’s Composer feature can directly extract, explain, and generate content for knowledge points in Obsidian notes, serving as a powerful AI-assisted tool.
- This learning method emphasizes the value of combining AI tools to achieve effects beyond a single tool.
- When interacting with AI tools, it is recommended to break complex tasks into clear steps to improve efficiency and accuracy.
- In the long run, regularly reviewing and revisiting knowledge points is crucial for consolidating learning results, and Obsidian’s random note plugin can be used as an aid.
Full Content
Let me introduce my latest “AI learning method,” which is extremely powerful. It uses Deep Research, Obsidian, and Cursor to do deep research, extract and explain knowledge points, and form a knowledge graph based on your needs.
Last year I released a members-only video introducing how to use Cursor together with Obsidian. Actually, I already had this idea back then. Now that we have Deep Research, this AI learning method has finally become a closed loop.
Deep Research is the first step. It can provide input with very high information density. But to fully digest it, you still need further processing of the knowledge points in the report.
That’s where Obsidian comes in. Its backlink feature is especially useful. If one note mentions another note, you can link to it, eventually forming a knowledge graph. This is why I put the results of Deep Research into Obsidian—so that the report note and the related knowledge-point notes are connected together. This helps us understand things systematically.
So then another question arises: how should knowledge points be extracted and explained? That’s where Cursor comes in.
Cursor’s Composer feature can directly modify documents and can also create new ones. Open the notes stored locally in Obsidian with it, and you can do everything through conversation.
In the end, back in Obsidian, we not only get to see a deep piece of content, but every knowledge point is also clearly sorted out for you. The rest is up to you.
AI has already helped you this much—what reason do you still have not to learn?
Hello everyone, welcome to my channel. Modestly speaking, I’m one of the few creators in China who can clearly explain the Why and How of AI. What I offer is more valuable than tutorials. Remember to give it a follow. If you want to connect with me, come to the newtype community. Nearly 1,000 people have already paid to join!
Back to today’s topic: the AI learning method.
I’ve shared in the community that if you want to gain an Unfair Advantage with the help of AI, one key point is knowing how to combine various AI tools to get effects beyond the framework. My AI learning method is the best example.
Let me give you a hands-on demo. The topic is: What technical innovations does DeepSeek-R1 have?
For Deep Research, I’m using OpenAI’s. On this application, they’re currently the most expensive and the strongest in the world, without question. After a few minutes, a very detailed technical analysis report is generated.
Like I said earlier, this report is very information-dense. If you try to force yourself through it, it’ll probably be a bit difficult. So I copied it and put it into Obsidian. For convenience in the demo, I created a new folder so everyone can see more clearly in a moment.
Also, I only extracted the first part of the report here, and I removed all the Markdown formatting. Because I found that if there’s too much Markdown syntax in the article, it has a big impact on Cursor. It may very well become unable to modify the article.
OK, next open the file in Cursor. Remember to choose Composer in the right sidebar instead of Chat.
As for Cursor’s three modes—Chat, Composer, and Agent—I mentioned in the community before: if you want to stay in control yourself, use Chat mode, where you can choose which part of the code to accept; if you want full automation, choose Agent. Composer is in between, with a certain degree of automation, which is exactly what we need right now.
Next, tell Cursor the first requirement: help us extract the knowledge points and mark them using Obsidian’s backlink format. For ordinary nouns, there’s no need to mark them—for example, company names and product names.
The article I’m demonstrating with isn’t long, so Cursor finished it pretty quickly. If it’s a very long article, it will work in batches, and we need to say “continue” before it will keep going.
After the first step is done, the knowledge points in this article are all marked out, very clearly. The second step is to create a blank .md document for each knowledge point. The file name is the name of the knowledge point. In other words, each knowledge point is a note. In this way, through the backlink function—that is, the marks from earlier—you can connect these individual knowledge points with the original text.
Creating these blank documents is very easy for Composer, and it was done quickly. Then the third step is to fill in the content. In each of the blank documents created just now, add an explanation of that knowledge point, and make it easy to understand.
Because these are all technical terms and not something highly time-sensitive, the model can handle them using its own knowledge reserve.
Once these three steps are done, we can go back from Cursor to Obsidian. You can see that clicking a knowledge point in the article takes you to the note for that knowledge point, where there’s a dedicated explanation. And when you open the article’s link graph, you can see which notes this note is connected to. If needed, you can also jump to any note from the graph.
What I’m demonstrating here are all basic operations. In actual use, everyone can refine and adjust them. For example, if there are knowledge points you think aren’t necessary, you can delete them manually, or let Cursor handle it for you. Also, when you ask Cursor to do work, try to break the steps apart as much as possible. Like I just did, I had it operate in three separate rounds, and I explained everything very clearly. That’s the only way to avoid many strange problems.
Finally, one more thing: in the long run, you still need to review and revisit the knowledge points from time to time in order to get good results. Otherwise, those notes will just pile up there and you’ll never remember them again. So for this situation, I recommend installing this plugin: Open random note. Just as the name suggests, when you click it, it will randomly open a note. So when you’re bored and have nothing to do, click it a few times and look at a few notes. Trust me, it really works.
OK, that’s it for this episode. If you want to learn about AI, come to our newtype community. See you next time!