The enshittification of AI

Why is AI so blatantly dull when humans are amazing?

I’ve been a bit silent lately. I’m focusing more on client work these days which makes it harder to find the time to post. But I’ve been wanting to write about this for a while.

Let me start with this cartoon I saw on Alberto Romero’s feed.

You read this and think “Yeah, who would be so stupid to fall for this.“

Then you open up ChatGPT and start getting frustrated, because it’s dull.

The geniuses are outliers

We, humans love our freaks (if they are just the right amount of freak). Most of the notable personas of history were freaks, misfits, total outliers. There were anything but ordinary.

We put people like Van Gogh, Einstein or Katalin Kariko on a pedestal. W give them prizes, put their work on display for others to see and say that these people are pushing us forward.

I love to read or watch the biographies of these people. They almost always follow the Hero’s Journey. They are misfits. They are outcasts. Society usually don’t like them. Spits them out as if they were the unwanted seed of a cherry.

Apple understood this in 1997 when they launched one of the best ad campaigns ever.

We have plenty of proof to accept the statement: “Geniuses are outliers.”

But when you say that, you also say that “Most humans are dull.”

So here we are. We take all the knowledge of humanity and train an artificial though-generator. Then we’re appalled that the thoughts it generates are dull.

“Why is it so dull?” - we ask. Because humans are dull.

When you think of AI-generated content you think of this:

Or this:

Let’s face it: 99% of AI generated content is shit.

It’s brain rot. Compared to what humans can do it’s terrible.

But when I say humans, I think of this:

…and basically anything from this guy:

The best of our best.

We compare the average AI content to the best humans can do.

Most humans are dull

But wait, there was this study, that said “humans preferred AI poems over Shakespeare”. But when you look at the actual study, you’ll see that:

The authors suggest that participants preferred AI poems due to them being more straightforward and accessible than the work of the prominent poets.

There’s nothing surprising there.

More people watch Love is Blind than Hamlet.

Most humans are dull and dull humans prefer dull content. They understand it, they can consume it.

I have a confident command of the English language even though I’m not a native speaker. Yet reading Shakespeare twists my brain in very painful ways.

So if you create a thought-generator by using data from a dull majority, the product will be:

1. Incredibly dull.

2. Yet preferred by the majority.

Transformers are pattern matching machines. But the pattern it found in humanity’s collective thoughts was dullness.

The enshittification of AI

As these models are being trained, we expect more and more of them. The big debate around whether deep learning has hit a wall or if scaling will work has nothing to do with AI.

It’s a civilization-level wager about ourselves.

If human genius - the spark that creates new ideas and inventions - is part of the norm, we'll find it. We just have to throw more silicon at the problem.

If it’s not, then parroting humans will only result in more sophisticated brain rot.

So we start training more and more.

OpenAI releases o1. Alibaba releases another so called reasoning engine. The gloves are off and the stakes are high. If there is no wall, the human mind will be recreated in the form of microchips. Human creativity will finally properly fit into the Theory of Production.

But we’re not there yet and so far all we’ve seen is exactly what we always see when tech companies start trying to cover edge cases: enshittification.

OpenAI used to be the darling baby of the AI world. They had a simple solution: ChatGPT, with plugins.

Then they killed plugins. Added GPTs. Changed the rules of the game. Added more features. Now you have several models. Some require prompt engineering, some doesn’t.

Our expectations of what AI can do have exploded. Two years ago having, a Discord bot mimicking Rick Sanchez was revolutionary. Now it’s “AI generated sludge“.

However, capabilities didn’t grow with expectations. Yes o1-preview is smarter, but it’s slower, has low limits and it’s not even that smart.

The entire AI industry is at fault.

Bloggers like me are at fault. Influencers, founders, investors.

I swear to God, if I have to see another post saying “This changes everything!” only to shill a mediocre product, I’ll gouge my eyes out.

A year and a half ago I loved reading about new technology. New tools coming out.

I don’t do that anymore, because nobody is solving the actual problems.

Writing a Linkedin post is not a challenge.

Generating 10 slides to upload on Instagram is not a challenge.

Creating an automated Youtube channel is not a challenge.

Understanding the context of a business is the challenge. Knowing WHAT post will make everyone go cray-cray and HOW to distribute it is the challenge. Ensuring a closed feedback loop is the problem. Building upon all the existing knowledge of a brand is the problem. You can’t solve these with a smarter prompt.

Instead, most people are waiting for Big Tech to finally liberate them from thinking.

"Woohoo, you don’t need to think anymore because AGI will do that for you.”

And that’s a terrible position to take.

Here’s another perspective to give you a better position:

What will you think about, when AI does all the thinking for you?

Start thinking about those things now and maybe you’ll learn something new.