The singularity is already here

Last week when I was finishing up a blogpost on the future or development teams in the 2030s, I found out something about myself that was truly disturbing. So, disturbing as a matter of fact that I could not continue writing before trying to figure out what was happening. And who I really was.

It all began earlier that day when I stumbled upon a random article on the vast interwebs about a new set of tools to detect AI generated content. I was intrigued and found the subject well worth a closer look. I don’t have any issues with generated content per say, as long as it’s well written and fits the intended purpose of the author. I would even encourage the use to be quite honest.

It is better to have synthetic content that fits the bill rather than the genuine article written and produced by fellow human, but that falls short. So, after having read the article, I started testing. Who was ‘cheating’ on their homework?

It turs out that there was a lot more generated content than I would ever have imaged. Large newspapers and news sites, corporate websites and private and personal blogposts all had the marks of the AI beast upon them. And there is nothing wrong with that at all. The paragraphs that where syntactic, according to the tools, where all well written and coherent and I spotted no AI shenanigans like obvious factual errors or the AI admitting being just that, an AI.

Lastly but not least I turned the tools on myself. Just for fun!

And to my dismay, the results were not good. They were horrendous! Apparently, I’m part machine now, twisted and evil, as 45% of the blogpost I just had written was derived from a GTP according to the tool.

Surely, it couldn’t be true. I just wrote those word myself. What was happening?

I tried another article I had written, and the results were better at only 25%. So, I tried another. This time coming in at almost 70%. I was baffled, confused and, to put it mildly, slightly annoyed.

This could not be happening. What was going on? I decided to rewrite it all, to make it more human and rid myself of this AI curse. But the mark of the beast did not wash away so easily. Every rewrite gave more or less the same results. I focused in in a paragraph and rewrote it again and again and the results where more or less the same. Pointing to only one conclusion. I’m an AI!

After a bit of an existential crisis, I decided to press on and to find out more about this strange affliction and why I had it. And after reading up on the subject I think I have it all figured out.

These tools look at the overall text for certain clues that it might be written by an LLM. As AI are trained on large datasets of millions and millions of wordy documents, reports, blogposts and anything that tends to contain a lot of text it has a certain style, a fingerprint of its own so to speak.

One of these giveaways is that an AI is often extremely ‘wordy’ and prefers long sentences with a multitude of decorations and lofty phrases. It has a tendency to use words that are not unknown but uncommon and lastly to emphasize its message to sound smarter than it is.

So, returning to that paragraph that I could not humanize with this knowledge I found the culprit. A single word, “significant”, was the factor that made the sentence synthetic. Removing it at once turned the text, and me, human and to 100% to boot. Now, why did this happen? How could my writing style be so synthetic and what is the root cause of all this? And does it matter at all? As I stated earlier, I have nothing against synthetic content. So, why did it annoy me so much to be classified as just that, synthetic?

After pondering the question for some time, I have finally found the answer to all of the questions above and even learned something about myself. Or at least this is the closes approximation of an answer that I could come to without visiting a psychologist, hypnotist or voodoo priest.

The conclusion is simple, I’m a Large Language Model, but a very human one at that. Since English is not my fist language, I tend to gravitate to phrases and feels a bit posher, eloquent and more native to bridge the gap of me being just a swede in a vast ocean of genuine English speakers, and writers.

When I was growing up, swedes and especially politicians spoke English in a certain, unpolished, manner that was often referred to as Swenglish. To the uninformed about this extremely local trend, it can be summarized as the use of English words in a Swedish sentence but without any actual change in how the vowels are pronounced or the sentence is structured. The end result is similar to English but not quite there, just like an LLM.

That is not the only reason from my past that factors in to this. As a child I suffered from dyslexia. And this I think might be the real or main reason for why my poor choice of words and incoherent sentences might be picked up as nothing but the mad ramblings of a GPT that has gone off the deep end. I’m still struggling with that self-image of being the stupid kid that could not read, or spell for that matter, so I have a tendency to overcompensate and produce rambling sentences with no real purpose or an end in sight.

To an extent I’m an LLM. As a grownup I read a lot, my affliction started to subside in my twenties, and my preferred language is English so as a direct side effect the content I consume is in it. My brain processes it, finds meaning and information and phrases and words that just goes together.

These phrases, once learned, are part of my arsenal of linguistic patterns and syntaxial decorations and when the need arise might be rehydrated, reconstituted or regurgitated into new sentences of my own. And endless barrage of words fills the page in my ongoing endeavor to rid myself of my ideas, thoughts and the incoherent nonces that fills my every woken moment.

The end product, a page or two, that hopefully conveys the message I want to get across but served up as a delicate and delicious word salad. That is just my style, I guess.

So, rounding things off. I’m still human and I will not change my style of writing. Mostly because I don’t think I can to be honest. But I will persevere and continue this onslaught of short blogposts for as long I can. Not because it is easy, but because it is hard. (I might have stolen that last part, just like an LLM).