Ernest Hemingway had the “Iceberg Theory,” also known as the “Theory of Omission”: you only need to write the tip of the iceberg and leave the rest underwater. If you know what you’re writing about, the readers can feel the depth the subject’s iceberg.

AI writes only the tip, but omits what’s underneath the surface. When you read it, you know it in your gut. Hemingway’s old theory might help us survive a new one; the dead internet theory is no longer a theory.

“Proof of personhood” is a crypto term. I’m stealing it. It’s the line between AI work and human work, and the line gets blurrier every moment AI advances.

How do you prove a human was behind a piece of work you stumbled upon online? You almost can’t - especially if someone hasn’t written before, meaning they haven’t developed a voice. That voice becomes the content you compare against to make sure it’s still human-written. No voice? No verification.

This is me practicing developing my voice before it’s too late, though it might already be. There are currently certain tells, traits that AI authored, but it’s getting harder to distinguish.

AI images came first, and the entire ChatGPT subreddit became a peanut gallery for laughing at boomers falling for dumb AI images of Jesus taking a selfie or Bigfoot vlogging about its life. It finally hit technology-forward professionals on LinkedIn. Slop post after slop post, what used to be a breeding ground for turning mundane and personal moments into a cringe “business lesson” is now just bots talking to one another.

I know some of you are thinking, “Gil, you actually read that garbage?” Sometimes! It’s like a filtered Facebook of old friends, former coworkers who won’t share their unhinged racist takes because there’s still some semblance of professionalism.

But it’s not just LinkedIn; those same subreddit hecklers didn’t realize that their cheap seats had already been infiltrated. The bots are laughing with them, replying in the comments, and they’re pretty good mimics.

The new AI writing “tells” are there, and by the time you see this, it might be outdated. We have the em-dashes, rest in peace. I learned to use em-dashes because when I learned to code, someone told me tech folks use them a lot, and I just added it to my writing, didn’t overuse it — but I found it useful to be precise with brevity, and I liked it. It added style, another way to write a sentence. Now I ask AI to remove them… “Hey, did you just admit to having AI write for you?”

I admit it’s useful for dumb things that are a nuisance but necessary. I subscribed this weekend to a “publisher” so I can read a specific article, and I wanted to cancel, because that’s all I came for, guess what? No unsubscribe button, you have to email support to unsubscribe… guess what, Gemini wrote this dumb email that should have been a button.

The other tells are

• “It’s not this. It’s that”
  • “It’s not procrastination. It’s fear.”
• “Honestly? That’s the revelation.”
• “Here’s the thing.”

Public people and their publicists are already catching on and are prescribing adding typos to their Instagram posts, CEO memos, and Slack messages. It’s a cat-and-mouse game that will end up with both the cat and mouse being automated.

Clearly, when the LLMs were trained, they labeled “this is good writing” versus your crusty cro magnon corner of the internet forum where everyone is just slinging insults at one another. They have been given instructions to write more like the training data they provided.

Why is writing different? The medium is the message; You write a handwritten note or card for a birthday or an event because taking time to write something down by hand is a lot of work. It’s meaningful. It says, “I spent my afternoon thinking about you, and this is a proxy to let you know, I care about you.” I write these essays, record and edit them myself. The reason you don’t see 2-3 videos a week is that I don’t let an LLM write this - AI can’t write this. This is my proof of work.

Writing is typically more serious and sincere. You collect the right words for the message and tone. Maybe we should be grateful these tone deaf CEOs aren’t doing layoff videos instead of emails.

These layoff announcements haven’t stopped since post-COVID. I remember waking up, ready to work in 2023, only to have my laptop bricked while reading an email sent at 4 AM. I don’t remember the email’s content, except that the message was clear and I was able to process it and move on. AI wasn’t as prominent then. Being in tech, I’ve read dozens of these CEO emails in the last few months, and I’ve spotted tells in each. Having AI write these messages that should require pause and reflection, only drains them of meaning. You don’t care.

Writing is also thinking… The best guest speakers you like listening to on podcasts or television speak so effortlessly about their subject matter because most of them have written books about their realm of influence. Just think about frequent good guests, even comedians, whose specials are just a chain of well-written jokes told together, which have been edited down over and over. The editing and refinement is the work.

Writing is a powerful tool for communication and building awareness on the web. People know this… personal brands open up opportunities for people. Part of this channel is me practicing this development. I made a couple of LinkedIn posts, and I can confidently say they helped me get my new job. It made it easier for recruiters to find me, and this lets me rest easy knowing I won’t have trouble looking for another job if layoffs or the like were to happen.

Because of this impact and reach, LinkedIn has turned into an AI slop factory. According to “Originality AI,” an AI detection startup, over 54% of LinkedIn English posts are likely AI-generated. I suspect that number will only keep growing.

Part of what bothers me the most about reading this slop is that we usually mimic what we read or consume. As I read more AI text, I know it’ll impact my speech, and I’ve noticed it in others as well - It’s not just annoying, it’s a contagious trait that gets me pessimistic about the technology.

I don’t want to be suspicious of AI writing; I already overthink, and it’ll just go on forever. I can see my chain of thought going… “ha, they have a typo - AI wouldn’t generate that, no wait, they probably added that themselves post hoc to seem authentic… nah… Billy wouldn’t do that, or would he?” I would have already spent more time thinking about the post than the person who asked an LLM to generate it.

How do we solve this? Besides being in person and checking for flesh and bone, it’s getting harder. Sam Altman tried tying your iris to a token with World Coin, no thanks. Different orgs are trying to create a human-verified label like you see “certified organic” on meat or veggies. I personally like “certified organic slop”.

Soon, IDs will be tied to our online accounts, so you’ll know that a human is behind the keyboard warrior account, but you won’t know that a human wrote that comment. I predict we’ll see a behavior shift in users, we’ll stop reading comments, and we’ll all get suspicious unless the platform makes it clear it’s not welcome. Or maybe some people might not care and will appreciate that content on their doomscrolling lunch break. I know you’re not one of those people.

In this new world of AI mimicry cat and mouse game, the proof is the work itself. Not a verified organic slop label. Not an 8k iris scan. It’s the hours you can’t fake, the voice you develop over years of creating. So don’t be overly suspicious, but trust your gut. If AI were to train on all of Hemingway’s work and write a new book, in his “voice”, I’m not reading it. It’s not Hemingway; the proof isn’t there.