How can an ordinary person systematically understand a vertical field in 4 hours?

Author: danny

A friend asked me why I seem to know everything about everything or every field. Aside from past experiences or current projects, I often learn on the fly. Today, I'll share how I use AI tools and Notebooklm to facilitate self-learning for ordinary people.

First of all, I want to say that this article is for systematic and structured learning and understanding of a specific field/thing/concept, and building your own knowledge system and graph. If you only need to understand some of the concepts and know what this xx is, then asking the mainstream AI on the market will probably yield similar results.

Using AI to learn about something new currently faces several bottlenecks and limitations:

First, it's an illusion. AI will (most likely) give you some fabricated data and stories, especially in niche areas, because of insufficient corpus and learning materials.

Secondly, there isn't that much detail. Due to issues like copyright, AI won't read through the entire article or book on its own. The training materials are usually other people's reviews and comments, especially information in specific sub-fields.

Third, it is difficult to accurately describe the problem. If you have never been exposed to this topic before, you probably will not be able to describe the problem you want to know well, nor will you know the causes and consequences of these things, let alone systematically and structurally collect information and form a systematic learning framework.

Theoretical section

My approach is actually quite simple: use the academic "quote/reference/impact factor network" to refine information, and then use AI evidence and divergent thinking to engage in a "left-right brain battle" to structurally understand a new thing.

Data-saving workflow:

Find valuable papers - add them to Notebooklm - generate prompts using AI tools - learn through Q&A in Notebooklm - add more valuable papers to Notebooklm - learn through Notebooklm - repeat this process.

Complex workflow:

Step 1: Following the clues (Time: 0.25 hours)

Instead of searching for "what is XX, and what is the principle behind it", directly look for the "pillar of stability" in that field.

1. Calling on AI (Gemini/Perplexity): Directly ask: "In [a specific subfield], who are the three universally recognized leading figures? What are the 1-3 highly cited classic papers they published that laid the foundation for this field?" (For example, in the LLM field, focus on papers like "Attention Is All You Need "). This represents the "present-day" [experience/representation].

2. Download first-order references: Extract the references from these 1-3 core articles, and download all the core references they cited. This represents the "past life".

3. Extracting high-frequency second-order references: Cross-referencing the references of first-order references to select the top 10 most cited articles and the top 5 most frequently appearing articles. This represents the "later" [period/stage].

The core logic: Seeing the world through the eyes of the masters is the cheapest shortcut. Don't underestimate this step; you're downloading a chart of the most fundamental intellectual evolution in this field over decades.

Step 2: Building a structured knowledge base (Time taken: 0.25 hours)

All the classic documents selected in the first step were uploaded to Google NotebookLM at once.

Generally, for classic articles, these two are sufficient: https://scholar.google.com/ or https://arxiv.org/

Why NotebookLM? Because it never creates hallucinations. It answers questions based solely on the information you provide.

Through rigorous literature screening, you have artificially cut off junk information on the internet, creating a pure and highly focused knowledge base for this field.

Step 3: Inter-AI battles (Time: 1-3.5 hours)

This is the core of the entire workflow. You allow AIs with different characteristics to cross-reference within your knowledge base, forming structured knowledge paths and logical deductions, ultimately leading to their own insights.

Replace passive learning with active questioning. Actively asking questions (out of interest) stimulates brain thinking.

1. Find anchor points: Ask Claude, Deepseek, Gemini, or Perplexity: "Regarding the field of xx, what are the core controversial issues and underlying theoretical frameworks in academia/industry?"

2. Closed-loop questioning: With these core controversies in mind, return to NotebookLM and ask: "Based on the literature I uploaded, how did the masters answer these core controversies? Please provide specific literature sources and reasoning logic."

3. A more nuanced approach: Copy the rigorous answers generated by NotebookLM and send them back to Gemini or Claude, who possess strong logical analysis skills. Instruct them: "Critically examine these viewpoints, pointing out logical flaws, limitations imposed by the times, or blind spots. Based on this, what three deeper questions should I ask next?"

4. Cognitive spiral ascent: Taking the vulnerabilities and new questions pointed out by AI, return to NotebookLM to seek answers.

Practical training

Let me use "What exactly is LLM (large language models)?" as an example 😂

Step 1: Following the clues (Time: 0.25 hours)

I asked both Gemini and Claude – hey, you're the one who gave such an answer!

gemini

claude

Then you suddenly remember your middle school teacher saying that scientific theories must be connected to the past, present, and future. So you ask AI to help you research which papers these core articles referenced (usually in the "literature review"), and which subsequent articles cited these core articles, and let AI help you filter them out.

Step 2: Building a structured knowledge base

Due to some original LLM features and AI permission requirements, we need to download it manually (or you can have your lobster do it for you).

Generally speaking, https://scholar.google.com/ and https://arxiv.org/ are perfectly sufficient.

After downloading, put it into notebooklm (currently one library supports about 300 entries).

Step 3: Inter-AI combat

You can start by asking some simple, intuitive questions on Notebooklm, then discuss and explore your understanding with other AIs, and finally send your conclusions back to Notebooklm for it to refute, demonstrate, supplement, and correct.

Notebooklm's answer and comments:

Repeat this process several times until you are able to create your own mind map.

If you want to be a bit more hardcore, you can ask Notebooklm to give you a test to check your knowledge.

By now, you have a certain understanding of this field (at least you know about past lives, present lives, and future lives, so you can talk for 5 more minutes when someone asks).

postscript

Save your "knowledge base" (and update it in real time, you can let Lobster do it), and create a separate folder - for example, I compiled theoretical articles related to "contract trading" into a separate book. When you need to analyze something, you only need to call up this folder, describe the data and cases, and you can analyze it with basically "no illusions".

It's not that current AI models are incapable of deep thinking and analysis; it's just that you're not using the right tools. (A very important parameter in LLM is the constraint condition and the input condition.)

Using AI is one capability, but making AI empower humanity is another.