Could Your Cat Make a Website? 💻🐈 - Version 1.1

(or: How to Solve the AI-Anxiety Problem Among Developers and Professionals from Any Other Field)

Date: August 13th, 2024

By Jay (Contact me at jaywebsites[@]protonmail . com)

Share, but give me (Jay) credit: jaywebsites.com/couldyourcatmakeawebsite.html

Let's Analyze This Problem

First, let's make a model and simplify what a cat is and what a human is.

Now the human, if they trained their neocortex correctly, could make a simple website, or even something more complex like a sophisticated web app. The cat can train its lizard brain all it wants; it won't be able to produce a website. Obviously, the neocortex is key. In order to work around this and still make the cat produce the website, we would have to give it a very powerful tool. Let's say a big red button; if the cat presses the button, it will send a signal to a computer which will autogenerate a website.

Website-producing Cat: [Animal] + [Lizard Brain] + [Big Red Button]

Now, this might worry other cats a bit. But if they as well would get the button, it would worry the first cat since it would become less special.

Human Comparison

Now let's take the old situation among humans.

Here, Human2 cannot do what Human1 can do. Human1 could produce a more complex website, while Human2 cannot (at least not very easily).

This is great. If you are Human1, you can make more money and have a good career.

Now the current situation since the popularization of LLMs:

Let's be clear. Human1 still has an advantage. But the distance to Human2 has shrunk a lot. Human2 could produce something as well now. This is the core of the AI-anxiety problem.

Dissecting the Problem

The edge Human1 had comes from the fact that they trained their neocortex. But now you can rent a sort of artificial neocortex called an LLM, that's already trained, for 20 USD a month (and there are good free options as well).

We are becoming superhumans. If you take the theories of Kurzweil and the practical implementation of Musk with Neuralink, the future becomes clear. As stated in the subtitle of Kurzweil's new book, we will merge with AI. With the LLM. We kinda already have a very close relationship with our smartphone.

Now this doesn't solve Human1's problem. Human1 wants to keep an edge. By working hard, by learning hard, by putting in more effort, by making smart decisions. This is the nature of the "professional."

To solve this problem, we have to dissect it.

First, take the professional's output: A website. A simple app. A simple piece of music. An essay. A podcast. A legal contract. These were the relatively small outputs of the pre-LLM era. We cannot think in these terms anymore. You cannot compete with output like that anymore.

Envisioning New Levels of Production

We have to change the characteristics of the output, in terms of quantity, quality, and complexity. The problem is that we have to start envisioning a level of production that doesn't exist yet. Something that Yaron Brook has explained very well is that it's hard to envision what will be produced by the many minds of the market, and this leads to all sorts of scary rationalizations about AI-generated (no pun intended) unemployment.

I disagree with the current binary debate:

  1. There will be a massive amount of technological unemployment, so we need a universal basic income.
  2. The market will smoothly solve the issue by producing new jobs that we cannot envision yet.

    3. I think this will happen:

  3. The market will solve the issue, but it's going to be anything but smooth. It will require deep thinking by the professional upper layer of the working class (yes, I know that sounds elitist, but at the moment, they are under attack, manual laborers are safe from the current wave of LLM automation at the moment). They have to think very closely about their craft and they have to become visionaries who can see into the future.

Let's say, start with quantity. Let's say you produce 10 websites in the time it would normally take you to produce 1. Now this is a bit of an antithesis to the nature of the professional. His/her edge is generally not mass production, but production that requires (more) thinking. But the thinking is starting to get taken over by the LLMs. So purely this route will trap us in a loop.

If the professional can make a site in 4 days, and the non-professional in 6 days, increasing the production purely in terms of quantity won't help much. Robert Kiyosaki has written about this way of thinking; he gave an example of someone trying to compete by carrying more buckets of water, while the other person took some time and built a water pipeline. Increasing purely the quantity of output can't be the full solution; the edge will be minimal and, in the end, you might get wiped out by the competition.

Now let's go to quality. Instead of a simple website, you could produce one that is content-rich, and the content could be co-produced by AI tools. You could make a much more immersive site in the same amount of time. Bringing programming, narrative, visuals, and sound together.

Now, this is something that is nowadays frowned upon. Post-WWII in the West, and especially since the dominance of Apple launched the iPod, minimalism is the standard in web design. The experimentation of the 90s (think gif-filled sites with background music) has been murdered by Apple just as they later killed Flash.

However, if done well, it could make a comeback. This is an interesting example because it consists of multiple elements:

  1. The engineering: code
  2. The art: narrative, visuals, sound
  3. The art direction: work done by producers, directors
  4. High-level design choices: software architecture

Of course, there are other elements as well, on the tech side: potential platform technology (web app), XR possibilities, and on the commercial/marketing side to make something that actually works in the marketplace, makes money. But let's focus on these 4 for now.

Reimagining Our Model

Let's look at our model again, and put 1 and 2 in them:

Human1 (Pro): [Animal] + [Lizard Brain] + [Trained Neocortex] + [LLM*]
Output: code & narrative, visuals, sound

Human2 (Joe Schmo): [Animal] + [Lizard Brain] + [Neocortex] + [LLM*]
Output: code & narrative, visuals, sound

*and other AI-generating tools

Not much of an edge. Of course, if you are a professional developer or musician, you could produce something of higher quality, but the gap will slowly narrow there.

Now when we get to 3) art direction and 4) software architecture, it gets interesting. Yes, an LLM can help with art direction and with software architecture. But the human is the dominant factor here. And this requires a lot of experience, knowledge, smart thinking, effort, etc. Just try to make a content-rich platform purely with AI. You will see very quickly that AI is weak in putting it coherently together.

There is an interesting example I need to give here. Under this essay, I will write a bit more about myself, and I will mention Sierra there. This was a game developer/publisher that dominated PC game sales during the 80s until the mid-90s. After a 25-year hiatus, the founders Ken and Roberta Williams came back to make another game. The game is called Colossal Cave 3D. It was released for many platforms including VR platforms. Roberta Williams is the game designer and her husband Ken handles the coding and business side of things. Basically, she is the art director (grown from her core as a writer and game designer) and he is the software architect.

I played the Quest 3 version of the game, and I was very impressed. They built it with a very small remote team (much smaller than what they were used to working with). Recently, there was a panel at the Adventure Game Fan Fair, where they and other legends from Sierra (and other game developers etc.) came together. They both answered questions about how they produced the game.

Let's look at this again:

  1. The engineering: code
  2. The art: narrative, visuals, sound
  3. The art direction: work done by producers, directors
  4. High-level design choices: software architecture

Ken said any coding he did was done with an LLM window open next to it. Roberta explained that the visual assets were mostly bought from the Unity store (and polished and tweaked a bit), the sound was rights-free and the music they found online. The narrative mostly was already there because it was based on the original 1977 game. Now not all these things were AI-generated, they used existing assets as well, but it's not hard to imagine that it all could be AI-generated.

However, it would be extremely hard for any normie team with a bunch of AI tools to build this game, probably impossible.

The result the normie team would get would be bad because art direction and software architecture at the grandmaster level of Roberta and Ken Williams cannot be produced by AI. There are games that you can sideload in your VR headset that were made with existing assets or AI-generated assets if you try them you will see immediately what I mean. It just looks like crap because they are missing experienced people that can pick the right Lego pieces and put them together in the right way. It was done so naturally in the game mentioned that I did not even see they used existing assets, everything fitted so well, because Roberta's mind works like Rachmaninoff putting all the notes in the right order. Ken made the right tech and business decisions as well by releasing the game on many platforms, guided by his experience as captain of industry game publisher.

Human1 (Pro): [Animal] + [Lizard Brain] + [Trained Neocortex] + [LLM/AI Tools]
Output: code & narrative, visuals, sound
Output: art direction, software architecture

Human2 (Joe Schmo): [Animal] + [Lizard Brain] + [Neocortex] + [LLM/AI Tools]
Output: code & narrative, visuals, sound
Failed output: art direction, software architecture

Now we are getting a bit of our edge back. But it's not enough.

Final Element: Complexity

Nintendo used to produce playing cards from 1889 until the 1960s. In 1949, Hiroshi Yamauchi took over as CEO. The playing card market got saturated in the 1960s. Yamauchi led Nintendo like an ancient shogun and saw business almost as war. He appointed people that could come up with something to revive the business. In the 60s and 70s, they experimented with toys (higher complexity than playing cards), and in the 70s and 80s, they transformed into a video game company (even higher complexity).

We are experiencing what Nintendo went through in decades now in a couple of years. We are building playing cards and are wondering how we should compete. The market has moved extremely rapidly because of AI. We were thrown into a new world, in the span of 2 years. These changes require constant monitoring.

It's not the answer, but the question that is the hardest (Musk, Douglas Adams), and we are asking the wrong questions.

The products that we as professionals used to produce come from the pre-AI era. Just like the playing cards came from the pre-personal computer era.

We should use our tools to the fullest and train our neocortex in those fields that cannot be taken over easily by AI.

We should reason by first principle (Musk, ancient Greek philosophers), and see what we can build in this new era, with these new powerful tools. Now that might mean that we come to conclusions that "feel" or sound very futuristic and unrealistic. They are unrealistic in the sense that they don't exist yet in our current world, but as long as they do not defy the laws of physics, we should not shy away from them.

To reach complexity our entire foundation has to be shattered. The playing cards have to be thrown away.

Let's take programming languages. As computers got faster, people leaned towards programming languages that are less abstract, like Python, PHP, JavaScript. If you learned those languages in 2020, you would have a good edge as a developer (although the languages are just 25%, 75% is the metagame around development).

In 2024 it doesn't make sense to just learn Python, when there are LLMs that can produce high-quality code.

The Conceptual Director

We should not think as a coder, but as the conceptual director, who has an LLM coder employee, but also many other employees in various fields. Some of these LLM employees can combine skills like a Renaissance man, a Leonardo da Vinci.

Old position:

New position:

"Conceptual director" is a bit of a broad term. But it should include skills (some of which are still to be identified) that cannot be taken over easily by LLMs, AI, AGI, ASI, etc. Examples are art direction and software architecture (but if you are in another profession, not in tech, examples could be made for your case as well, think the higher-level director in your field).

Becoming a T-model (Gabe Newell or some earlier others) is also important. If you already have a skill you know well, that is a good thing to keep. But the | part of the T should also be strengthened by higher-level director-type skills, that you should take just as seriously as if you would take learning and becoming good at a new programming language.

The _ part of the T should include a bit of knowledge of a broad range of skills as mentioned (so art, legal, biotech, physics, finance, and more fields). The reason is that you need to be able to communicate with your team of LLM employees. You need a little (not very deep) of domain knowledge of many different fields.

New Opportunities Unfolding

Now we can produce an entirely new product. We are becoming the cat that can produce a website. Because the cat has a new tool.

Human1 (Pro): [Animal] + [Lizard Brain] + [Trained Neocortex] + [LLM/AI Tools]
Neocortex Training (edge): conceptual directing, art direction, software architecture, mini-courses in many domains, such as art, legal, biotech, physics, finance, etc., to be able to direct AI/LLMs better
Output: complex, multidisciplinary, futuristic and not seen before products

Human2 (Joe Schmo): [Animal] + [Lizard Brain] + [Neocortex] + [LLM/AI Tools]
Neocortex Training (edge): nothing (too lazy, or works in a different field)
Output: pieces of code, narrative, visuals, sound, scientific data, but not the well-integrated complete products

Remember to resist the emotional urge that makes you think it sounds too complex, futuristic, or difficult. Our new tools are highly capable as well.

Now the puzzle is solved. And everything starts to make sense. In order for us to remain professionals, we have to act like professionals in the new reality. We have to build what isn't there yet. We have to go beyond our monodisciplinary craft.

If you are wondering what potential multidisciplinary complex products might be possible, you are on the right track.


A Bit More About Me and My Interest in LLMs (Not Needed to Read for the Cat-Theory)

I started using GPT-2 and GPT-3 via a game called AI Dungeon in 2020. It's a text adventure game that uses AI to allow a user to create their own adventures. I got hooked by this game for over a year for a couple of reasons:

  1. My first coding experience was with QBasic in the mid-90s; I started making some simple text games. A bit later, around the year 2000, I used that experience with the BASIC language to make a more fleshed-out game on my TI-83 graphical calculator. It was a text adventure with some RPG elements. It was called School Quest. You walk around in your high school in the typical text adventure way (think Colossal Cave Adventure, 1977), and can encounter enemies in a battle mode (again: all text) where you could fight or flee, and there was an inventory system.
  2. The graphical adventure game, the way Sierra used to make them, is my favorite game genre. My dad used to play them in the 80s while working at the Dutch Ministry of Infrastructure and Water Management (yes, they actually played the games during working hours in the government office). Games like King's Quest, Space Quest, Larry. Although there were graphics involved, you had to give input via text, like "open door". I quickly caught the adventure bug. I developed an interest in real estate, writing, video games, computers, art, futurism.
  3. And last but not least, the power of GPT-2 and GPT-3. I was as amazed as people that were using ChatGPT for the first time a couple of years later. At the time the LLMs were not so well-polished and politically correct yet. I remember a specific moment where an unarmed character I made got very afraid because she encountered an enemy. The sentences and emotion "she" produced were extremely lifelike. I was shocked and it genuinely made me think there was a ghost in the shell.
  4. When ChatGPT (3.5, later 4.0, 4o, etc.) came out, I again obsessively used it every day, many times for hours.

Although I got very fast and handy with "prompt engineering" (not sure if I like that term too much) and became massively faster with producing code and projects, something was still bugging me. It was the AI anxiety that my skill was not going to be special anymore. Over the years, I've thought a lot about this problem, read a lot about it (I especially like how Ray Kurzweil approaches it in his latest books), and this essay is the conclusion of my thinking.

Jay

jaywebsites.com

jaybio.ai