How to Systematically Understand a Vertical Industry as an Ordinary Person in 4 Hours

Author: danny

My friends often ask me, “Why does it seem like you know about everything or every field?” Besides some past experiences or current projects, most of the time, I learn on the spot. Today, I want to share how I use AI tools and NotebookLM to pursue self-education as an ordinary person.

First, I want to clarify that this approach is aimed at: systematic and structured learning and understanding of a specific niche/field/concept, and building your own knowledge system and map. If you only need to get a basic idea of some concepts—like what is this xx—then asking mainstream AI tools on the market is probably enough.

Currently, there are several bottlenecks and limitations when using AI to learn and understand a new subject:

1. Hallucinations: AI (most likely) will generate fabricated data and information, especially in specialized fields, due to insufficient training data and sources.

2. Lack of detail: Due to copyright and other issues, AI won’t read entire articles or books itself. Training materials are usually reviews or comments from others, and information in niche areas is particularly scarce.

3. Inability to precisely describe problems: If you haven’t previously engaged with the topic, you probably can’t clearly articulate what you want to learn, nor understand the causes and effects, let alone systematically collect data and form a structured learning framework.

Theoretical part

My method is quite simple: use the academic “citation (quote/reference/impact factor) network” to refine information, then employ AI for evidence and divergent thinking to create a “left-brain/right-brain” duel, structuring your understanding of a new subject.

Streamlined workflow:

• Find valuable papers
• Upload them to NotebookLM
• Use AI tools to generate prompts
• Q&A learning within NotebookLM
• Add valuable papers back into NotebookLM
• Continue learning in NotebookLM
• Repeat

Complex workflow:

Step 1: Follow the trail (Time: 0.25 hours)

Don’t search “What is XX” or “How does this work.” Instead, directly find the “cornerstone” of the field.

Call AI (Gemini / Perplexity): Ask directly, “In [specific niche], who are the three recognized leading figures? What are the 1-3 highly cited classic papers that laid the foundation for this field?” (For example, in LLM, focus on papers like “Attention Is All You Need”). Represents “this life.”

Download primary literature: Extract references from these 1-3 core papers, and download all the key references they cite. Represents “past life.”

Refine high-frequency secondary literature: Cross-reference references in the primary papers, select the top 10 most cited or frequently appearing papers, and identify the top 5. Represents “future.”

Core logic: Following the masters’ perspectives is the lowest-cost shortcut. Don’t underestimate this step—you’re downloading the most essential evolution of ideas in this field over decades.

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

Upload all the classic papers identified in Step 1 into Google NotebookLM.

Generally, for classic articles, these two are enough:

Why NotebookLM? Because it never hallucinates. It only answers based on the data you provide.

Through strict literature filtering, you cut off internet junk info, establishing a pure, highly focused knowledge base for the field.

Step 3: Cross-examination among different AI models (Time: 1-3.5 hours)

This is the core of the workflow. You let different AI models with varied strengths interrogate your knowledge base, forming structured knowledge paths and logical deductions, ultimately leading to your own insights.

Ask proactively, not passively. Active questioning (interest-driven) stimulates thinking.

Find anchors: Ask Claude, Deepseek, Gemini, or Perplexity: “What are the core controversies and underlying theoretical frameworks in [XX field]?”

Close-loop questioning: Take these core controversies back to NotebookLM: “Based on the literature I uploaded, how do the experts address these controversies? Please provide specific sources and reasoning.”

Dimensional reduction review: Copy the rigorous answers generated by NotebookLM, then feed them into a logically strong AI like Gemini or Claude. Instruct: “Critically evaluate these viewpoints, point out logical flaws, limitations of the era, or blind spots. Based on this, what are the 3 deeper questions I should pursue?”

Cognitive spiral ascent: Use AI-identified gaps and new questions to revisit NotebookLM for further answers.

Practical application

Let me give an example: “What exactly are LLMs?” 😂

Step 1: Follow the trail (Time: 0.25 hours)

Ask Gemini and Claude simultaneously—here are their responses:

gemini

claude

Then, recall what your middle school teacher said: scientific theories are always built upon previous knowledge, with a “past,” “present,” and “future.” So, ask AI to research which papers these core articles reference (usually in “literature reviews”) and which later papers cite these core works. Have AI help you filter.

Step 2: Build a structured knowledge base

Due to some limitations of the original LLM features and AI permissions, you need to manually download (or have your assistant do it).

Generally,

Download and upload to NotebookLM (currently supports about 300 papers).

Step 3: Cross-examination among different AI models

Start by asking simple, intuitive questions in NotebookLM, then discuss and explore your understanding with other AI models. Later, send your conclusions back to NotebookLM for rebuttal, argumentation, supplementation, and correction.

NotebookLM’s responses and annotations:

Repeat this process several times until you can outline your own mental map.

If you want to be more rigorous, ask NotebookLM to generate test questions for you.

By then, you’ll have a decent understanding of the field (at least knowing the past, present, and future, so you can talk for 5 more minutes when asked).

Postscript

Save your “knowledge base” (and keep it updated in real-time, maybe with your assistant). Create a dedicated folder—for example, I keep all articles related to “contract trading” in one folder. When analyzing a situation, just pull up this folder, describe the data and cases, and perform near hallucination-free analysis.

It’s not that current AI models can’t do deep thinking and analysis; it’s that you’re not using the right tools. (An important parameter in LLMs is the constraints and input conditions.)

Using AI is a skill, but making AI empower humans to become stronger is another skill.