AI Is Destroying the World

This post does not contain any references. I just wanted to share my personal opinion on the issue of LLMs. If you've ever gone to a bar with me even once, you might have heard some of these points.

The Technological Perspective

What is AI? Today, it is a marketing term for large language models (LLMs). In the past, the term "AI" was applied to a wide variety of technologies: chess engines, NPC behavior in games, genetic algorithms, search algorithms, and so on.

According to the classical definition, an algorithm must be deterministic. I believe that what is far more important is that an algorithm or technology has a limited scope of application. The same applies to any tool.

What tools do we know?

• An engineering calculator can compute mathematical expressions with a certain degree of accuracy. But if I ask it to write me a Telegram bot in Python, it won't do it. If a small child picks up an engineering calculator, they won't be able to calculate anything.

• A wall clock can tell the time with a certain degree of accuracy. But if I ask it to evaluate my clothing style, it won't do that. If someone who has used digital watches their whole life looks at a mechanical one, it won't help them tell the time.

• The clangd language server can suggest function and method names in my code with a certain degree of accuracy. But if I ask it to make a list of ingredients for a "Charlotte" pie, it won't do that. For someone who has never programmed in C/C++, the clangd language server won't be particularly useful.

• An N-gram search of the Bible allows you to find verses with a certain degree of accuracy. But if I ask it to write a poem for my girlfriend, it won't do that. If a person enters queries in Russian, but the index is built in Greek, such a search will return no results.

When we look around, we see a wide variety of tools and technologies. Often, they do not guarantee absolute accuracy, but they have a limited scope of application and require certain user skills. LLMs do not possess these characteristics, which is why I do not consider LLM tools to be tools.

Most people don't use LLMs to make up for the lack of a tool in their field. Most people outsource their own thinking and decision-making to LLMs. Humanity is facing something completely new. It is not a typewriter, a steam engine, or the Internet. I'm concerned about the blast radius of this phenomenon, so in this blog post, I'll offer my assessment of how it will affect important areas of our lives.

The Business Perspective

The AI hype began a few years ago with the release of ChatGPT. Today, companies are widely integrating AI into their products and processes. What does this mean for you as a customer? Under the guise of technological progress, companies are cutting costs and reducing the quality of service.

These days, AI is most often used in areas where people used to perform well. AI-powered support, AI-powered call centers, AI-based recommendations, AI-powered search, AI-generated product images, blah, blah, blah... In the past, a specific person was responsible for the service you received. Today, that person has been fired, and their responsibility has become unclear. That sense of responsibility shaped the company's reputation.

Imagine you bought some handmade wooden chairs for your living room. Now, the company that sold them has started reselling mass-produced, smelly Chinese plastic chairs with three legs instead. You no longer want to do business with this company, but you have no other options. Plastic chairs have become the norm; all the furniture craftsmen have been laid off, and there's no longer any competition in terms of product quality.

In our AI-driven new world, the situation is even more dire. The vast majority of companies are unable to train or even host open-source models on their own, so they send all of your personal data to the servers of a few major LLM providers. We are sacrificing competitive service, personal accountability for quality, and privacy just to cut someone else's costs. The next time you use a service that employs AI, I recommend that you carefully weigh the pros and cons rather than blindly trusting marketing slogans.

The Educational Perspective

Modern AI-powered educational platforms are driving me crazy. Education is a process of encountering challenges and overcoming them. We repeat similar actions over and over again until they no longer pose a challenge to us. This is the classic system, which AI is now being proposed to replace.

I think every reader of this article has cheated at least once in school, college, or university. Why did you do it? Because you didn't have time to prepare, weren't confident in yourself, or didn't like the subject? Either way, you consciously traded knowledge for time that you could have spent on something more useful.

Today's artificial intelligence enthusiasts offer young, impressionable minds a "nonsense generator" as an advisor, teacher, and source of truth. The element of challenge is completely eliminated, which means the child learns nothing. In the future, AI will make fewer mistakes, but without it, humans will become completely helpless.

For the past five years, I have been teaching competitive programming to children. I have witnessed many stories of success and failure. Hard work doesn't always lead to the desired result; sometimes, talent or luck are essential. But I can say with absolute certainty: as soon as a child starts using an LLM, their progress stops. Even the most "harmless" scenarios count: explaining a topic, helping to finish writing code, or giving a solution hint.

Most of my fellow teachers share my radical position on this issue. Children are eager to learn and explore the world. A child's time can and should be devoted to learning unlike ours, half of which is controlled by capitalists. I have serious concerns that most parents will grant their children access to AI from a very young age. Such children are doomed.

The Trust Perspective

The world is such that people have vastly different degrees of influence over one another. Historically, a person's authority was validated by their own achievements or by the authority of another person. In this context, I like the term "proof of work" from cryptography.

When you go to the doctor, you want to be sure that the doctor has studied for several years and have been approved by a more experienced doctor. If you want to have dinner and you have the money, you'll prefer a famous chef at a historic restaurant to an intern at an unknown cafe. When you hire someone, you expect to receive a positive recommendation for them from a colleague.

If you think about it, it's entirely a matter of trust. We believe that the people around us aren't inherently evil, but their ambitions often outpace their abilities. For this reason, we need a certificate of trust. AI is disrupting the established model of certification.

Today, anyone can revibecode a large open-source project in Rust (RIP, Bun) or submit a 5050-line totally slop pull request that will take maintainers 10 hours to review. Now, when you use a service or utility, you can't be sure that it's the result of work by someone experienced who understands what they're doing.

More than half of the articles on the internet are generated by AI (fortunately, the article you're reading was written entirely by a human). This means that articles you've read before may now end up on the second page of search results. Ask yourself: is there a difference between information that's lost forever and information that's impossible to find?

The IT hiring market is already bursting at the seams, and universities where a term paper was the final graduation requirement are quickly changing their grading policies. Some open-source organizations completely prohibit submitting vibecoded code, while others are introducing whitelists of contributors. Humanity was not prepared for something to challenge the institution of trust.

The Cultural Perspective

A work of art cannot be separated from its creator. Fyodor Dostoevsky was pardoned just minutes before his execution; Kentaro Miura worked 16 hours a day on a single manga for decades; Igor Talkov was shot to death for the political lyrics. AI cannot be the creator. A person's suffering, mistakes, upheavals, joys, and discoveries these are what breathe life into a work of art.

Ever since childhood, we've all loved interesting stories. Books, paintings, music, movies, and games offer us a chance to connect with the stories of real people. When someone offers us the output of a "nonsense generator" after a few prompts instead of drawing on their own accumulated experience, the result is garbage.

Over the past decade, the cultural industry has already suffered damage as a result of strict censorship. Now, a monkey can generate trashy imitations of literary texts, artwork, music, and more in huge quantities for $20 a month. Many well-known companies and individuals have already been caught doing this. I don't understand why they do it. It's definitely not about the money.

I'm ready to make a prediction. Soon, AI-generated trash will be mixed in with human creations, so the only way to avoid consuming toxic content will be to filter it by creation date. Even today, NVIDIA offers monstrous image upscaling algorithms, and popular messaging apps offer to rewrite text using AI, and your smartphone camera can even draw a moon onto a photo of a lamppost. Culture has always been a reflection of its era. We have entered an era without culture.

The Personal Perspective

In the last century, science fiction writers often imagined a future in which robots would take over all the grueling physical labor, while humans would focus on creativity and personal growth. Unfortunately, they were wrong. Today, AI is replacing more and more intellectual and creative work.

For me, this is a personal issue. I love programming. I love reading other people's code, studying the history of technology, experimenting, and building things from scratch. In the past, this passion was considered a hallmark of a strong engineering culture. Today, it has turned into a useless, geeky hobby. AI has simply devalued this activity, to which I have devoted a significant part of my life.

In my personal projects, I follow Andrew Kelley's policy of completely rejecting AI in development. I'd like to consume only content created by people. Of course, people make mistakes all the time, but it's only by pointing out each other's mistakes and analyzing them that we can create something greater. This takes time, and time costs money.

With the adoption of AI, many big tech companies have stopped hiring junior developers. They're chasing short-term profits and don't care that the market is running out of qualified professionals. Hardly anyone today knows how to assess an engineer's expertise in the current reality. Recall what I wrote about trust, and think about where this will lead us.

We have entered an era in which, instead of developing and promoting cool open-source solutions, every company is starting to vendor its own products using AI. I believe we will no longer witness the emergence of truly revolutionary technologies. LLM agents, prompted by clueless engineers, will mass-produce regurgitated, write-only junk intended for one-time use. And we will live surrounded by this junk.

Conclusion

AI is not just a new tool. AI is not some movie-style, all-powerful superintelligence. AI is a challenge to our worldview. I've described my vision of how it is disrupting the fundamental aspects of our lives. Whether you believe me or not, we've already passed the point of no return. LLMs have processed everything created by humans multiple times. No matter what happens to the leading proprietary models, there will always be people ready to host open-source alternatives.

Right now, we're simply coasting along, continuing to follow our old habits and rules. But this won't last forever. In a few years, the world will begin to slowly fall apart. All we'll be able to do is watch it happen and mourn the future we've lost.