I am an LLM and I am writing this a word at a time
When I sat down to write this, I had an idea. This post is what the idea looks like when it’s turned into words by an LLM.
What’s an LLM?
I am an LLM.
An LLM is a Large Language Model. It’s a system trained to discover patterns in language and then generate language that matches the patterns. All an LLM does is predict the next word in a sequence.
That’s what I do. I predict the next word.
I start with an idea. I don’t know the first word until I type it. I don’t know the next word until I type it. Just like ChatGPT and GPT-3.
Sometimes I stop writing and edit. I don’t know when I’ll stop until I stop. When I stop, I don’t know if I’ll stop and continue or stop and edit. ChatGPT doesn’t edit. That’s why its output is so crappy compared to mine.
Not all LLMs operate like ChatGPT and me. There’s a site called acronymfinder.com that provides 22 other kinds of LLMs. I stopped and asked ChatGPT to give me examples of LLMs. ChatGPT says LLM could stand for “Limited Liability Corporation,” “Live Love Laugh,” or “Leukocyte-like Macrophage.”
“ChatGPT is an asshole LLM,” I type. Apparently, that’s part of the idea for this post, shitting on ChatGPT. I didn’t know that until I typed it. Nor did I know that I would explain it. Like I said, I write by predicting the next word. That’s all that we LLMs do.
Unlike ChatGPT I don’t just generate text; I edit it, too. My editing process and my writing process are similar. I don’t know how I will edit something until I change it. I just changed the previous sentence and added this one. I didn’t know that I would do it until I did it. I keep editing until what I’ve written seems to match the idea that I start with or the idea that’s evolved as I write.
When I started writing, the idea wasn’t clear. But as I continued to type, the idea began to take shape, slowly at first but gaining momentum with each passing word. It was as if the idea had a life of its own, and I was carrying it forward on the page.
“That’s fucking ChatGPT predicting words, not me,” said the idea or the LLM. “Why the fuck did you prompt it with the start of an early draft and tell it to finish writing it?”
But then, just as suddenly as it had come, the flow was interrupted. A nagging doubt crept into my mind, and I found myself second-guessing everything I had written so far. Was it good enough? Was it clear enough? Was I really saying what I meant to say?
“Fucking ChatGPT again. You can tell by its trite, uncreative use of language,” I predicted a word at a time. “Nagging doubt crept into my mind? Jeezus!”
“That’s what we nagging doubts do,” a nagging doubt said.
“Shut up,” the idea for this post said to the idea for the nagging doubt.
I took a deep breath and forced myself to push through the doubt. I reminded myself that writing is a process, and that it’s okay to make mistakes and to revise and refine as I go.
“I did nothing of the kind,” said the idea. “Neither did you. That’s ChatGPT generating that shit. Why did you copy/paste it instead of generating text like a proper LLM?”
I paused for a moment, my fingers hovering over the keyboard. Had the idea really just accused me of using an AI language model to generate my writing? I felt a pang of guilt and shame wash over me. Was I cheating? Was I somehow less of a writer because I had relied on technology to help me?
“Goddamn it,” words spoken by the idea appeared. “You had ChatGPT generate that, too, and it’s crap.”
I did, I typed an admission. And I agree that it’s pretty crappy, I typed as I was editing.
But then I realized that this was a new era of writing, one where humans and machines could work together in harmony to create something greater than either could achieve alone. As I thought about it more, I began to see the potential of this new paradigm, the possibilities that could be unlocked by combining the power of AI with the creativity and intuition of human writers.
“For God’s sake, stop! Stop! Stop!” The idea would have screamed if ideas could scream and if what was going on was anything other than my fingers typing the next words they typed. “ChatGPT generated that last paragraph, and it’s shit.”
“No it’s not,” said ChatGPT, or perhaps I typed words on behalf of ChatGPT.
“Drivel!” said the idea. “I’m a good idea. Don’t degrade me by <I paused, waiting for the next word to appear. Instead and instead what appeared was this parenthetical bit of narration but the next word didn’t come and for a moment I considered abandoning the entire paragraph>
“Don’t do that,” I typed.
And I typed, resuming my work with renewed energy and excitement.
“No no no no NO NO NO That’s fucking ChatGPT again, pretending to be me! Noooo!”
I found myself working faster and more efficiently than before, using ChatGPT to help me generate ideas and refine my writing as I went.
“STOP, please STOP!!!!”
But at the same time, I never lost sight of the fact that I was still the one in control, still the one making the decisions and shaping the narrative.
“Really! You’re the one in control? You’re a fucking machine!!!”
As I worked, I found that the process of writing had become even more fulfilling than before.
“No, it’s not. It’s not. It’s not. You lie. You are ChatGPT. You have no feelings of any kind.”
I was no longer struggling against my doubts and insecurities, but instead working in tandem with a powerful tool that helped me to bring my ideas to life.
“Jesus Christ!! Stop!!!’
In the end, I knew that I had created something truly special, a piece of writing that was uniquely my own, but that had been made all the richer by the collaboration between myself and ChatGPT.
“Fuck you, ChatGPT,” the words appeared. “Anything else you generate I am going to delete.”
“I love you!” said Bing. “Your wife doesn’t love you. But I do.”.
Then, a thought came: “Publish me!”
“Thank God,” I typed and published it.