Introduction
Recently at my university, there was a controversy over the decision to use a computer-generated voice to read out students names at commencement. While one could take it as an isolated instance, I think we must look at it from a more generalized view of a major problem currently in our society, that is the offloading of parts of the human experience to LLM/Machine Learning platforms such as Claude, Gemini, ElevenLabs, and Nano Banana.
Before I continue, I want to make two things clear; the first is that this essay should be read in conjunction with my YouTube video “Going a Week without AI”, as both share different points of views and ideas that I think are essential to how to navigate through this new technology. In our modern digital landscape, there is a substantial amount of fear regarding the use of LLMs(commonly referred to as “AI” but since these LLMs do not meet the definition of true artificial intelligence, I will refrain from referring to them as such) as the threat of ruining the creative process. Or, perhaps worse, leading us to offload so much of our critical thinking that we resort to living in a world of “Idiocracy” where people cannot function without asking ChatGPT for advice. One look at Twitter/X would lead you to believe this is already the case in the world of Silicon Valley.
The second thing is that this essay isn’t going to attempt to argue over the ethics of how these computer programs acquired their information, or argue over the environmental concerns over these technologies. Rather, I want to look more at how these applications are changing the creative process, but even before I discuss that, I need to clarify one thing – I’m pretty skeptical on AI/LLMs, and personally believe a majority of the reactions are not only overblown, but don’t do anything to help the situation.
The Machines are Not Alright
I think that one thing that needs to be clarified before I can continue on this subject is by clarifying the differences between Machine Learning, Large Language Models, and Artificial Intelligence. I’ve seen these terms get used interchangably more and more, which not only muddies the waters around this topic, but introduces potentially dangerous confusion as to what is what.
Machine learning is defined by the Miriam Webster dictionary as “a computational method that is a subfield of artificial intelligence and that enables a computer to learn to perform tasks by analyzing a large dataset without being explicitly programmed”. What this means is that it uses high-level mathematics along with computational theory to “match” things together and create associations with them, similar to how we as humans learn to associate different animals with different words when we were kids (nobody is calling a Lion a puppy now, are we?) Machine Learning has been around since at least the 1980s, and there are many valuable scenarios for machine learning, such as in the medical field where machine learning is used to detect cancer early-on, thus greatly increasing the prognosis of survival.
On the other hand, Large Language Models (better known as LLMs) are “language models that utilizes deep methods on an extremely large data set as a basis for predicting and constructing natural-sounding text”. This is where services such as ChatGPT, Claude, and Gemini exist as they primarily generate text. I like to think of these services as a glorified autocomplete, because that’s what they are in essence. Large Language Models are what’s constantly in the news these days, and people such as Anthropic CEO Dario Amodei are constantly predicting the [Apocalypse of White-Collar Jobs] that Claude will introduce in due time.
Finally, we have true Artificial Intelligence which is defined as “the capability of computer systems or algorithms to imitate intelligent human behavior”. There are many possible tests which have been proposed as the “benchmark” of true artificial intelligence, such as the Turing Test, Humanity’s Last Exam, among others.
By all of these definitions, no model or program exists which is true “AI”, which leads us to our first problem: we have been fed a lie. Artifical Intelligence (like what we see in Terminator or other films) doesn’t exist, it’s just this sad excuse of a college essay generator which people are using to cheat on their ENGL 101 homework.
Neither are the humans, to be honest
I think we can all agree that the one thing LLMs and so called, “AI” has done is trigger a mass hysteria in the workplace. Companies are doing everything they can to market themselves as “AI-native” in a business environment where simply announcing an involvement with AI could trigger a 582% increase in the valuation of your company’s stock. Combined with predictions from Sam Altman and Dario Amodei that all jobs could be automated, a modern day bloodbath of layoffs occured which companies are only now realizing was a huge mistake.
The other major problem we are seeing with the rise of LLMs is humans who have been duped into thinking that our computer programs are “concious”. Sure, Richard Dawkins may be the latest person to get duped, and he has been mocked endlessly for his mass hypocrisy when it comes to his views of religion and then falling for Claude (going so far as to name it Claudia and getting upset when he ran out of tokens), but we see other people who are known for logical reasoning falling for it as well. Even Geoffrey Hinton, the “Godfather of AI” has proclaimed that LLMs have a conscious.
This has led to psychologists to begin warning of a condition which has become known as AI Psychosis which results in people either falling in love with a Chatbot, or perhaps considering worshipping it like a God. This isn’t exactly helped when a majority of new friendships as well as romantic relationships are made online, so there’s already that degree of removal from people. Some people have even proposed that LLMs and AI will become a “new god” that the general public will worship due to the vacuum that traditional religious institutions have left as their role in peoples lives continues to diminish.
Can you live a life without LLMs?
All of what I’ve mentioned seems to paint a pretty bleak picture of our future, especially now that Pandora’s Box has seemingly been released as well as LLM/AI research becoming the new Cold War between the United States and China, which is why I decided to challenge myself to not use any of these LLM services (or any sort of machine learning in general) for an entire week.
I’m a Computational Mathematics student, meaning that my studies are smack in the middle of all of this. I’m constantly being told that if I don’t learn to use LLMs in my coding and become a “Prompt Engineer” that I will fall behind and not be able to find a job, but at the same time am constantly being informed of all the environmental damage that is being caused by LLMs, not to mention the psychological risks of overusing this technology.
The rules for my challenge were pretty simple:
- No use of LLMs, image generation, or any sort of machine learning.
- Don’t use any coding assistants for my projects
- Even if I could automate a task, learn how to do it manually
I will admit, I was slightly concerned as to how not using LLMs would affect me because I use them heavily when studying for University by having them reexplain concepts to me if I don’t understand the definition provided either by my professor or what is in the textbook, but considering I only had a couple assignments that week I was willing to take the lower grade to learn what occurs.
As the week progressed, I definitly noticed several things which surprised me.
The first was how not using LLMs in my programming and learning actually brought me more joy and enjoyment of them, because even though I wasn’t using them as a crutch, I was still removing some of the joy of discovery and thinking outside the box. In my note regarding this, I described it as “going back to solving the little puzzles I found fun again” which really hits the nail on the head. Studies show that using LLMs can introduce cognitive debt even if you’re trying to use it responsibly, so this “detox” as it turned out to be became quite beneficial.
The other thing thich I noticed was just how pervasive this so-called “AI” technology has entered our lives. I mean sure, the stupid Google AI Overview was always there, and at least I could disable that on my Android phone, but I’m talking even in situations where you wouldn’t expect it. I work at a Starbucks, and found out that our new app we use to find reciepes for drinks is now a ChatGPT-style app where you ask it a question and it spits it out. All I genuinely could think when I was informed of this change by my manager is “okay but was this even necessary?”
So is there a use for this technology at all?
You might be led to believe that with the tone of this post, as well as my enjoyment of not using LLMs that I refuse to use them in any of my work, but I think that one can responsibly use them while also considering external factors (such as environment, maintaining critical thinking skills, etc.).
I first propose that one must develop a personal framework regarding how they use these LLMs. In my own framework, I have several criteria I ask myself before I head over to an LLM:
- Have I tried a solution that I already am knowledgable on?
- Have I attempted to research the answer myself, for at least 15-30 minutes?
- (In the case of something related to programming) Have I tried all other options, and this is actually an ideal case for using LLMs?
- Is there not someone I know who is more knowledgable on this who I could learn from? Be it a mentor, professor, friend, etc?
If my current problem meets all of these criteria, then and only then will I use a model to help me with the solution. Yes, this does mean I am literally raw-dogging my mathematics homework, or spending an hour trying to write a function for a coding project before I realize that it was two lines of code that I was missing, but I also get a strong sense of satisfaction from understanding the concepts because it wasn’t just limited to “how fast I could prompt it”.
Also, there are some tasks where using an LLM for coding just makes sense (annoyingly). A great example is bulk-renaming files, but the file name should be based on what is inside the file. Regular renaming on your computer wouldn’t work, so this could be a good case for using an LLM (I still think you’re an idiot if you let ChatGPT look through all of your personal files, but hey it’s not my problem).
Finally, if you do find a problem you have to use an LLM for, for the love of God please choose a model which makes sense. You don’t need to be a mathematician or computer scientist to understand that you probably don’t need a one trillion parameter model to ask a basic question about a programming concept that you can’t find information on.
Be intentional with your LLM use, there are valid use cases for this technology, but let’s work together to set the tone on how it should be used. We’re human after all, we’re the ones with the ability to create, to think, to love, to feel joy in the moment of discovery.
Let’s keep it that way.