**Codex跟我迭代十几轮，Claude一天就全作废？ — Transcript & Summary | SozAI**
Source: https://sozai.app/transcript/codex-iteration-claude-invalidated/

Exploring Claude 3 Opus's superior website-building capabilities and the importance of narrative sovereignty in the AI era.

## Key Takeaways

- Claude 3 Opus significantly outperforms GPT-4 in website building, showing a clear generational improvement.
- A document-first workflow is crucial for scalable and transferable AI-driven development.
- Narrative sovereignty is essential for individuals to control their information and presence in the AI era.
- Community value lies in the connections between members, not just in centralized distribution.
- Effective AI use depends on clear, audience-aware prompting and iterative refinement.

## What the video covers

- The creator compares websites built by Claude 3 Opus and GPT-4, highlighting a generational leap in quality and aesthetics.
- Claude 3 Opus was used to iteratively improve a personal homepage, emphasizing the value of a document-first approach.
- The video introduces the 'Super-Intelligence Academy,' a community platform with over 20,000 interconnected profiles visualized as clickable nodes.
- The creator discusses automating data updates using GitHub Actions and APIs to maintain the website content dynamically.
- Aesthetic design was guided by AI instructions to achieve a high-end look with interactive effects like a spotlight following the mouse cursor.
- A free tool powered by the Grok 4.5 model was developed to help users leverage public contacts effectively, demonstrating practical AI deployment.
- The video stresses mastering narrative sovereignty in an age of information overload and algorithmic dominance.
- The creator explains the importance of providing AI with clear, context-rich prompts tailored to the audience.
- The concept of community is redefined as a network of meaningful connections rather than just groups or newsletters.
- The video concludes with reflections on what it means to be an AI expert, focusing on effective communication with AI rather than coding skills.

## Chapters

1. 00:00 Introduction to Claude 3 Opus and Website Building
2. 00:58 The Importance of Narrative Sovereignty
3. 02:03 Iterative AI Prompting and Audience Context
4. 02:58 Super-Intelligence Academy and Community Visualization
5. 04:10 Automated Data Updates and Aesthetic Design
6. 05:15 Developing AI Tools with Grok 4.5 Model
7. 06:26 Reflections on AI Expertise and Effective Use
8. 09:02 Conclusion and Community Invitation

Answers

## Questions about this video

What makes Claude 3 Opus better than GPT-4 for website building?

Claude 3 Opus delivers a significant generational leap in aesthetics and functionality, producing more polished and high-end websites with interactive features that GPT-4 lacks.

How does the document-first approach benefit AI-driven development?

Creating a detailed document first allows the AI to understand the project comprehensively, making it easier to hand off tasks to different tools or collaborators and ensuring consistent execution.

What is narrative sovereignty and why is it important?

Narrative sovereignty is the ability to control your own story and information in an age dominated by algorithms and information overload, helping individuals maintain autonomy over their digital presence.

## Full Transcript — Download SRT & Markdown

00:00

Speaker A

True experts can identify the essential connections between things, simplify the complex, and find the most effective path forward. Here is the situation: yesterday morning, I received a message from Youchao, who told me that Claude 3 Opus is exceptionally good at building websites.

00:14

Speaker A

My channel's audience knows Youchao; he is a startup founder and is incredibly busy, so if he went out of his way to tell me this, I had to try it for myself. So, after I got up, I

00:23

Speaker A

followed his advice, and the results truly exceeded my expectations. Here is the website I built with Opus, and here is the one previously built with GPT-4; you can clearly see that they feel like they are in completely different

00:33

Speaker A

leagues. I have polished my site over dozens of rounds and gone through several major iterations to get it to this point, but compared to this Opus-built site, there is a generational gap. After finishing it, I immediately realized the immense value

00:46

Speaker A

of a personal website in the AI era, so I quickly posted about it in the community and am recording this video to share what this site does, how it was made, and why we should do this. It is about identifying what is truly

00:58

Speaker A

valuable amidst a mountain of seemingly noisy or varied information. If I had to summarize in one sentence why this is important: it is about mastering your own narrative sovereignty, whether you are facing other people or AI; in an age of information overload and

01:11

Speaker A

algorithmic dominance, this is especially crucial. But first, let's take a look at what exactly Opus 3.5 created for us; this is the post I shared. In this post, you can see the version I previously built with ChatGPT; I actually thought it was quite good

01:25

Speaker A

—it was tidy and had all the necessary information, but once you open the Opus version, you can feel the distinct generational leap in aesthetics. I can also show you the specific prompt I used, which is what I started with yesterday. What I did was

01:38

Speaker A

first open Claude, and then I let it work directly on my codebase. You see, yesterday I basically just said this: I described my personal homepage and noted that the entire history is hosted on GitHub, trusting that if I pointed

01:50

Speaker A

my AI toward this project, it could grasp the whole thing. Then I told it, you should perform a systematic review and provide bold suggestions for updates. The suggestions it gave initially weren't actually very good, honestly. At first, it provided some

02:03

Speaker A

text-only versions, so I told it that my audience isn't just hardcore techies, but the general public. I gave it some more context, and it tried again.

02:11

Speaker A

Later on, I realized that since I need to add many personalized components—which I'll talk to everyone about shortly—I decided to have it draft a document so I could hand it over to Cursor to build. This is where I told

02:20

Speaker A

Cursor to read the document. This is actually a key point we teach in our course: "document-first." Whatever we are building, we should ultimately end up with a document, and then you can take this document and use it everywhere. I’m currently switching

02:32

Speaker A

from Claude for work to Claude for code. I could just as easily hand this document to Cursor or others to execute, which is why this document is so incredibly important. After I gave it this document, it started working, and

02:43

Speaker A

after a day of progress, I got this current version. You can all see it has quite a lot of content now. In this version, I created something truly, incredibly interesting. This really interesting thing isn't just the stuff above—what I’ve done, my

02:58

Speaker A

conversations, or how I’ve been—but this "Super-Intelligence Academy." We are building a journey. You see, there are over 20,000 dots here, and each of these 20,000-plus dots can be clicked on. For instance, if you click on me, it takes you to my profile page, and

03:15

Speaker A

this is my content. Everyone in our community is here. As you know, I run an AI community, and many people call their newsletters or WeChat groups "communities." A WeChat group is a community, but a newsletter is definitely not. What a community really

03:29

Speaker A

means is that every single one of us can connect with everyone else. With 20,000 people here, those links can generate 200 million connections—that’s the (n multiplied by n-1) divided by 2 equation. So, these connections are the true value and the

03:41

Speaker A

real meaning of a community. When I created this Hyper-College, I told everyone that I didn't want a centralized distribution system, but rather something that fosters connections between people. I think this map actually visualizes that concept very well; I find it quite

03:55

Speaker A

interesting. I also explained in my article exactly how I went about doing this. Basically, every month, I use a GitHub Action to pull data from Circle via my API and store it in a Vercel Blob, which updates automatically each

04:10

Speaker A

month without me having to do a thing. As for the aesthetics, I told the AI, "See how the spotlight is shifting? Make it look high-end, don't make it look cheesy," and it really delivered, even with effects like the light following

04:23

Speaker A

my mouse cursor—see that? The light moves along with my mouse, which I think is really cool. The aesthetic is truly excellent; I never would have come up with that on my own. I can tell it what I want, but the way it actually

04:33

Speaker A

implements the details is far more impressive than what I imagined. Looking further down, there’s another very interesting point: this "Wenwen" certificate. If you follow me, you know that I made all my contacts public a long time ago, so anyone can use my

04:47

Speaker A

contacts, and many friends have commented saying they used them with great results. But later, I thought that even though calling these contacts sounds simple—just a line of code or a prompt—many people still lack the skills, so I built this free tool for

05:06

Speaker A

everyone to use. Is there a future for it in the AI era? Will it be replaced?

05:15

Speaker A

How do we pivot? Let’s just pick one and give it a try. It’s powered by the Grok 4.5 model, allowing it to search the original text to perform the task. I can also explain to everyone how this was made; I actually started

05:28

Speaker A

yesterday afternoon, just using my phone to tell it to use my contacts, but the tricky part is how to deploy this model for others to use, right?

05:38

Speaker A

Luckily, we have our own "builderspace," which I’ve mentioned many times. In the community toolbox, we’ve provided a module that lets you log in, call many different APIs, and even directly deploy the kernels you’ve invoked, so it’s not just a frontend. Inside, it

06:00

Speaker A

first intelligently retrieves all my relevant articles; although it’s a so-called RAG, by pulling all my articles and synthesizing them, it finally forms an answer. You see, is there any future in this? If you’re interested, you can pause and check the

06:16

Speaker A

quality of this answer yourself. I think the layer being replaced is the one involving judgment on materials, which isn't very nuanced. Yes, I believe it will disappear—it’s a waste of time. So, it really is what I said; it’s not just empty talk,

06:26

Speaker A

it’s a statement with a real attitude, and I think all of this is correct.

06:30

Speaker A

And it’s about what determines the transformation, what needs changing rather than just swapping labels, as attacking performance gets you closer to the result. Yes, or it’s these—these are actually things I’ve said, and I think they’re pretty good, so

06:39

Speaker A

you can go take a look. The reason I can create things so quickly is that I rely on this underlying system we have.

06:45

Speaker A

So, I really just mentioned it last night, he went to do it, and after he finished, I published it—actually, it was already published. Then I’d suggest making it faster, changing the name, or swapping the model. Providing the thought process to the user with

06:56

Speaker A

minimal effort is what allows so much to be created, which is a very important point. So, what is an "AI expert"? What does it take to be considered an AI expert? In my view, we haven’t taught any coding; the

07:06

Speaker A

methods we talk about are just speaking directly to the AI, but the thing behind it is actually very power.

07:19

Speaker A

. Whenever something broke, they’d go to a shop to fix it. One day, a screw fell out, and he needed to find a replacement. The author suggested that the tab from a soda can would work perfectly, so just use that. But his

07:31

Speaker A

friend replied, "How can I use such a cheap tab on my expensive motorcycle?" I wanted to go to the store to find a really good capacitor to replace it, and the author said that he actually had no idea that the capacitor on the

07:42

Speaker A

soda can was far better than the one he found in the store. Because that soda can capacitor is actually made by extremely expensive machinery, so its quality is very high. Even though it's common and cheap, there is profound

07:53

Speaker A

expertise behind it. The author not only knew that the capacitor was functional, but he also recognized that its quality was actually very high. But a rookie, or his friend who doesn't understand machinery or industry, would just think, "I need to go to the store

08:07

Speaker A

to find something very fancy, that’s the right way to do it." This is actually what we see all the time, many of these instructors will tell us we need to use some fancy prompting techniques or a bunch of tools to get

08:18

Speaker A

things done. In reality, true masters can find the essential connections between things, simplify the complex, and find the most effective solution.

08:26

Speaker A

This includes other examples, like when I saw Song Yadong fight someone, he hit a video creator. He used just a bit of technique, and that creator was in pain for a long time and couldn't get up. If you ask a fighting champion like Song

08:39

Speaker A

Yadong, "What makes your fighting so effective?" It’s in that one quick, powerful punch hitting the vital point, right? It’s about being accurate, fast, and forceful, rather than doing flashy, showy moves like "White Crane Spreads Wings" or "Eighteen Dragon

08:50

Speaker A

Subduing Palms." No, things like the " Eighteen Dragon Subduing Palms" only look impressive in martial arts novels; what real combat experts use is that one punch. If you look at Messi's movements, you’ll see that truly great players don’t do showy bicycle

09:02

Speaker A

kicks; every one of their movements is the simplest and most effective. Anyway , digging aside, this is my personal page. Aside from listing my work, it also features two very impressive modules. The significance of these two modules is that they present the

09:17

Speaker A

results of my long-term accumulation in an intuitive way; when I say my community has over 20,000 people, it's just a number that people can't grasp, but here you can see what each person is like. The substance behind it

09:30

Speaker A

becomes much more apparent. I hope that explains it clearly. How is it done? How to do it is to simply tell the AI, "Take my content, distill it, summarize it, and then build this website." You can copy my prompts, or even tell it

09:44

Speaker A

that a certain website is good and ask it to replicate it, or copy any personal website you admire. I believe it can handle that well, but the true substance is what we need to build ourselves. This leads to our third

09:56

Speaker A

point: why you should even create a personal website. Basically, there are three reasons, just as I mentioned in the text. The first is to reclaim your own narrative sovereignty. Think about it—what is your impression of me? I don't know. You might be thinking of

10:11

Speaker A

one right now, perhaps based on a few videos you’ve seen or things I've said. But the impression in your head is likely not the one I intend to project. I can't control what's in your mind, nor can I control how I'm

10:23

Speaker A

perceived across different platforms, but I can control providing a definitive entry point. I put my story there, so if people want to learn about me, they can come to my personal site to see how I tell my story, what my

10:36

Speaker A

narrative is, and what I truly care about. So, on my personal website, you'll see how I frame things, what I actually care about, and what information I choose to prioritize, right? Along with my past judgments and the things I’ve been thinking about

10:50

Speaker A

recently. All of this is me building my own narrative, and these narratives aren't just for people to see; they’re for AI to see as well. You can try it now—if you ask your AI assistant, it will likely index my

11:01

Speaker A

homepage because I've set up the site’s UX and sitemap for it. When the AI searches, it will easily find my personal site and read through the content. So, whether for humans or for AI, I am taking control of my own

11:13

Speaker A

narrative sovereignty. The second point is that this personal homepage acts as my own territory. The reason you're seeing this video today is because the platform's algorithm decided to push it to you; otherwise, you wouldn't be able to find most of the many videos I

11:26

Speaker A

upload. I upload a lot of videos every week, and I checked the other day—on YouTube alone, I post an average of seven videos a week, and the vast majority of these you never get to see.

11:35

Speaker A

When you think of me, you have to search for me, but think about it: would you really search for a specific person on platforms like Xiaohongshu or YouTube? Most likely not. That’s why we lack a channel where others can find

11:46

Speaker A

us with certainty. Why did I buy a domain like li.ai for my personal homepage? It’s so that when people think of me, they can find me reliably, and build trust here through my own narrative sovereignty. I have also

11:57

Speaker A

placed the things I truly care about and my collaboration methods here. For example, regarding collaborations here —if you want to invite me on a show, or for institutional consulting, long-term partnerships, or custom projects, you can find it all here. My

12:09

Speaker A

past speeches and the things I’ve shared before can also be found here, which reduces my need for repeated explanations and allows me to share my past insights with everyone for free.

12:19

Speaker A

And things like my merchandise and such can also be seen here. This image was just modified by AI, and it’s a bit funny, but anyway, you can see that anything related to me can be found right here. This is incredibly

12:33

Speaker A

important in our era because, due to algorithmic distribution and the flood of information, it is very difficult to build attention and trust. If you don't have a free space, even if you have millions of followers on a platform, if

12:44

Speaker A

the platform decides to shut you down tomorrow, your followers become meaningless. Those followers have no way of truly finding you; if they don't see you in their feed, they will just forget about you. Don't think that people will actually click on their

12:53

Speaker A

following list to go look for you; they won't. When people forget, they forget. Once they stop seeing your content, you are simply forgotten. Therefore, having such a free space is extremely, extremely important. In fact, when I started this community, it was to

13:04

Speaker A

provide everyone with just such a space . When I introduce myself, I say that your self-introduction in the community can act as your internet business card; you can see that right here, can't you?

13:12

Speaker A

When we click on someone, we can see what they've posted before, what they've commented, and what their various profiles look like, allowing us to understand their background. They leave a digital footprint or "context" on our site, and this context is

13:25

Speaker A

something they can use in the future to help attract others. This is the true significance of what we call the 20,000 times 20,000 connections. I hope I’ve made it clear what this site is, how it’s built, its highlights, why

13:39

Speaker A

we’re doing this, and the value of having a personal website in the AI era . If you're interested in more, you can find my post in "Show Projects." If you want to know the specific prompts I used, leave a comment under that post,

13:52

Speaker A

and I’ll be happy to share them. I hope this video is useful to you all.

13:56

Speaker A

See you next time. If you enjoyed this video, you will definitely love the community I’ve spent the last two years carefully building, "Superlinear Academy." If my videos are where I summarize all the practical knowledge I've learned, then the community is

14:07

Speaker A

where we work together to put that knowledge into practice. Visit supern.academy or download the " Superlinear Academy" app to join for free; I look forward to learning together with you.

Topics: Claude 3 Opus GPT-4 AI website building narrative sovereignty Super-Intelligence Academy AI community document-first approach AI aesthetics Grok 4.5 model AI expert

Study this video

- [Flashcards from this video](https://sozai.app/tools/ai-flashcard-generator/?from=transcript&id=77107)
- [Quiz on this video](https://sozai.app/tools/quiz-generator/?from=transcript&id=77107)

Free, made by AI from this transcript. No signup.


---
This is the markdown twin of https://sozai.app/transcript/codex-iteration-claude-invalidated/ — the same content, without the markup.
Published by SozAI (https://sozai.app). Reuse and quotation are allowed with attribution and a link back.
Machine-readable index: https://sozai.app/llms.txt · data API: https://sozai.app/api/