But it may help us self-actualize
I constantly hear “Large Language Models just produce statistically average language”, as if that were:
A) Inherently bad
B) Different from what humans do
Then, I also hear “Language is infinite. AI can’t keep up with all of the possibilities”.I have written before (here and here) about how the creation of Large Language Models actually suggests that language is not infinite, and that we are confusing what was once too big to comprehend with what is actually unbound and never ending. Now that we can numerically comprehend language, we can see that it is not infinite.
There is tension in this discussion between linguistic theories and mathematical theories, and thus tension between old-school linguists and more modern computational linguists and mathematicians.
My previous point is that language is not infinite, it is merely massive. Our idea of “infinite” is a hyperbole, not a mathematical truth.
I am revisiting this idea because our yearning for language to be infinite is closely related to our yearning to be unique. Therefore, suggestions that language is not infinite (and is calculable) are suggestions that we ourselves are not unique. This is why creatively-minded linguists often abhor the idea of LLMs.
Before, I was focused on the mathematical finitude of language, and focused on challenging our conception of infinity. I still believe that, but I actually want to focus on the other side of my point from earlier writings: language SEEMS infinite from our finite perspective.
This is similar to the way in which we all seem unique from our own perspectives, but if we zoom out we are largely similar to others. We see ourselves as singularities in an infinite spectrum of human identity. So, if language is not infinite, neither is our method of expressing our singularity. Then, if we feel we can't express that we are unique, we begin to doubt that we are. The implication is that our expression isn’t special, and neither are we.
Here is an example:
Make a list of all the reasons you are special.
Has no one ever described themselves like that? Is that truly a unique list?
I don't know about you, but my list makes me wonder if I am special. It certainly makes me see that much of my language to describe being special is not.
This is the challenge to our ego that AI presents.
But, here is the thing: language doesn’t need to be infinite for us all to have unique ways of expressing ourselves. Language is big enough to let humanity continuously explore it, even without being infinite.
Digging...
The deeper we go into our own thoughts, feelings, and relationships, the more “special" we are. The same is true for language. Superficially, people speak similarly, and this is what a Large Language Model represents. Deeper than that, people have nuances in the way they communicate. We can imagine this being represented by a well-trained and tuned personal AI or PersonalGPT; A model that stores our idiosyncrasies. But even deeper than that, we have the things that truly make our self-expression unique. These are our organic feelings and our novel interpretations, and these will not be replaced by models (in my prediction) because there is no model to follow. This is where we deviate from the norm; where we are truly special.
This all ties back to Large Language Models like this:
If we model the “shallower” parts of ourselves and our interactions, we give ourselves more space for the deeper, irreplaceable parts.
How?
I am imagining the possibility of this as I write this blog. Honestly, I have been struggling to get the ideas on paper even though they have been circling around in my head for quite some time. I thought they were clearly defined until I started to write.
This makes me think about how far along I could be in this piece if I plugged the ideas into a well-tuned “Gabriel GPT” and had it write a first draft. Then, I could go in and make sure my ideas came through and were expressed in ways that I like. I don’t currently do this because it sits pretty low down on my list of priorities, but one day…
There is tension here, of course. I wish I didn’t need help writing. I wish I could write brilliantly, efficiently, and consistently without any help. But, real-world constraints make this difficult. I have a business to run, a family to take care of, and any number of things outside of my writing that take time and energy. And, even when the time and energy are there, sometimes I am not as focused as I would like. The fact that this bothers me is a manifestation of me wanting to be special.
The part of me that wants to be special does not want help from AI. However, the part of me that wants to do good work will take all the help I can get.
So how does this relate to infinite and finite concepts of language?
If we continue thinking about the example of what I am currently writing, how much of it is really unique? When we break it down to the exact combinations of words, their order, and their relation to the subject at hand- very little of this piece is unique. What makes it unique are the small pieces that carry the novel thoughts (at least, I hope).
The same is true for everything we write, hear, or say. Most of it fits into widely replicated patterns, while a small fraction is what distinguishes it as "ours". So, the majority of the language that is used and that will be used in the world is not even expansive, let alone infinite. Language is more expansive in the less frequently used patterns or words. Or, better said, the part of “language” that is the most diverse in use is actually the smallest part by volume and share of voice.
Utility
This is useful because it means we can be effectively helped by a model in the vast majority of our speech, writing, and translation. This will help us retain time and mental bandwidth for the tip of the iceberg that needs to be truly novel or unique.
Ego
This is threatening. I don’t like the idea of anyone or anything “writing like Gabriel” or “translating like Gabriel”. Very few people do. But, what we all have to realize is that these models work like the part of us that was never unique to begin with. As well trained as the model becomes, it will only ever “write like Gabriel” to the extent that Gabriel is a fixed, replicable identity.
We all have parts of us that are fixed and replicable. This may be painful to consider, but it comes with a bright side. We also all have a unique part of ourselves that, however small it may be, is so potent that it colors everything we do. If we continue to drop this potent color into replicable parts of ourselves, we will continue to show up as special in this world. We will continue to show up as the part of ourselves that can’t be replicated, no matter how good any model becomes.
What is really being threatened?
The sad revelation that comes with LLMs is that we aren’t as special as we thought; most of our words and ideas fit squarely into a box of statistical averages. But, as people will say, LLMs are just the aggregates of other people’s speech. So, we didn’t cease to be special with the advent of AI. In terms of the way we use language, we never were that special. The difference is that AI is now holding up the mirror for us all to see it. This is a threat to the ego. It hurts.
But, it is also a liberation to the part of us that really is special.
Just like me struggling through this piece to get to an insight that is truly mine, LLMs will allow us to navigate the replicable parts of ourselves so that we have more time to spend with our idiosyncrasies. Once we realize that 90% of what we say isn’t special at all, we are free to spend more time with the 10% that is.
But, when we insist on being 100% organic and special, it costs us a lot of time and energy. This is a wonderful opportunity when we can handle it. Struggling through this piece was a gift because it allowed me to really work through my thoughts. But sometimes, that struggle comes with too high of a price. Losing jobs, losing sleep, or not spending time with my family is too high of a price just to feel more special.
So, when I get around to training GabrielGPT to help me write these blogs, it will certainly come as a bit of a slap to the ego. But, I won’t actually be any less special. The “tip of my iceberg” will still be mine, and it will come through in my writing. We will all still be just as special and just as ordinary as we ever were, but LLMs will help us see the difference more clearly.
In terms of self-actualization, I see this as a good thing.
Do you?
