The ugly Internet. Why rappers say crazy stuff at the beginning of their careers, and why RAG won’t help you write better articles with AI.
There's a problem with online content. It's all crap. We know it, those who create it now it, yet we still consume it. Why? How is technology changing things?
Online content is terrible. Hideous, really.
Because we are not used to paying attention to anything anymore, online writers came up with the clickbaity formulas we all know:
Here’s X things to do if you are looking to Y
X strategies to make sure you do not X
X products you didn't know existed
X strategies to completely change your life
Dating X people in under Y days - my story
And the list goes on and on and on. I call these “cigarette articles”, because you read them, get a short-lived dopamine hit, but you are never really satisfied.
This is the problem with content today. It’s a lot like cigarettes. You know it’s bad for you, but you do it anyway.
How many times a day do you find yourself reading utter bullshit? Or worse, watching TikToks (or Reels, or YT shorts) and then showing a really funny one to your friends?
Yeah far too many. I began dreading the sight of a friend approaching me with their phone cracking the usual “Hey you gotta see this”. And then it's something that will make me laugh, sure, but I won't even remember it the day after. This content clutter is making it impossible to focus, and this is why I don’t watch Reels anymore.
Frankly, when I was a kid, I thought that having a portable TV would be fantastic. Now, I hate portable TVs, because TV is not what TV used to be anymore.
I feel surrounded by an ADHD-affected crowd that is incapable of appreciating anything longer than 30 seconds. Drunk on short, dopey content whose only purpose is to make you waste your time.
It used to be Facebook posts, then Instagram pictures, then Stories, then TikToks and Reels. Now, it’s everything.
What is incredible is that people don’t even seem interested in creating valuable content. The gamification of the interface of TikTok, Insta, etc. pushes people to create crap. Crap is what keeps users glued to the screens, so why bother creating valuable content?
And then we train AI models on this very content. What the hell. This is a recipe for disaster. this is why we are now drowning in bad content.
Even Text-based content is becoming terribly dull
Don’t get me wrong. I don’t think we should be reading scientific articles or investigative reports all the time.
But for God’s sake - thes 10 point, 3 point or 5 point lists are becoming the norm. All I see on the internet is bullet points and lists. It’s as if McKinsey interns were writing all of the stuff we see on the internet.
I even see articles about writing articles that claim the following: “People like structure, and if you want to reach the maximum number of people and have them read your article, you should use this structure blah blah blah.”
Breaking news: most people don’t read. They don’t like to read, and they are not interested in reading at all. And that's fine. That's normal human behavior. I love reading, and I know I am not the norm.
But I guarantee that if a person needs to know something, you do not need to use clickbaity structures or downright dumb content. You just need good SEO for your articles. Actually, the crazier the content, the narrower your niche, the stronger following you will attract. Let me explain why.
The artist’s curve
You know how artists say the craziest things at the beginning of their careers? Especially rappers. Think of Eminem.
2000:
They said I can't rap about bein' broke no more
They ain't say I can't rap about coke no more (Ahh)
Slut, you think I won't choke no whore
'Til the vocal cords won't work in her throat no more? (Ahh)
Yeah, pretty crazy - even for before the woke era. Now let’s take a look at one of his most recent hits.
2010:
When I'm with you, I get the shakes
My body aches when I ain't with you, I have zero strength
There's no limit on how far I would go, no boundaries, no lengths
Why do we say that until we get that person that we think's gonna be that one?
Yeah that’s right. It’s changed hasn’t it? This happens to all artists. All of them. It’s the artist’s curve.
Artists follow a pretty standard curve. So do companies - but we’ll talk about this in another article or I will never get to the point.
Here’s a pretty accurate representation of the artist’s curve:
This usually happens to artists that are not a product of the “industry” (i.e. this graph does not apply to Justin Bieber).
I think that there is a positive, linear relationship between the amount of “crazy” things said and fame. Eminem said some pretty provocative things at the beginning of his career and this certainly helped him become well-known in an otherwise very competitive world.
This also applies to online content. In the beginning, you can attract a pretty narrow audience that is very loyal to your content just by posting things that people are not willing to post for fear of retribution/disapproval. Eventually though, as your name becomes more well-known out there, you will cease to be as provocative. Because you don’t need to.
Is it natural? Is it artificial? Is it just a pre-designed winning strategy?
Hey, maybe all three. But what matters for this analysis is the following: text-based content is dull nowadays. And it’s hard to find good articles that are actually entertaining to read.
So if you’re a writer, please - do not fall into the trap of using structures all the time. They suck. You know it, I know it, the readers you want to reach know it. At least not in the beginning.
Standard structures do work for established authors
Thanks a lot, Sherlock. Once you are famous you can write about whatever you want and people will flock to read your stuff.
Think of Sigmund Freud. He showed up to a lecture high on cocaine, blathered about things he did not know and then privately praised the drug for helping him talk nonsense for half an hour. If you know German, you can read the excerpt below. If not, you can trust me because I read the translation on Twitter.
But this guy was Sigmund Freud, not your average coke-head. He could command a level of credibility in the field of psychoanalysis that is unmatched to this very day.
Starting small means bringing a lot of content to the table. And this is where AI becomes an obstacle to brand success, especially for startups.
AI is not bold enough
If you ask ChatGPT to craft a captivating article, it will likely be as captivating as filing your tax return (for the kids reading this, it ain’t fun).
This is not an engineering problem only. Remember the curve? I hope you do. If you don’t remember the curve, google “Alzheimer’s” - your Google links will likely be purple.
OpenAI is now a global company and, for better or worse, subject to international scrutiny. There is no room for public experiments anymore. There is very little margin for error. In simple words, the fun is over. And so is fun ChatGPT - say welcome to boring, compliant, as-fun-as-an-accountant-can-be, ChatGPT.
These AIs cannot write compelling pieces or explanations, but they are great for customer service operations or other standardized tasks.
I invite you to ask chatGPT to write a nice article and then read it without any modifications. It will not be very interesting. I know this because I tried (hey, I’m a lazy person just like you).
Because companies are strapped for cash and people are generally a lazy bunch, I am seeing more and more botched articles being posted (as well as images, don’t get me started on midjourney images. They suck. We all know they are AI generated, and they suck.).
RAG will not save you from bad content
RAG stands for Retrieval Augmented Generation. Because I hate complex explanations, I will tell you what it is.
RAG is when you give a Large Language Model (Like GPT-4 , the model that powers chatGPT nowadays) special access to a database where specific information is stored.
LLMs are probabilistic models pre-trained on internet content (thence, its low creative quality).
What do I mean by “probabilistic”? LLMs are probabilistic because they use “embeddings”. Embeddings are words converted to a numerical value stored in a vector database.
What the fu&^ does it mean?
It means that I translate a word in a machine language made of numbers. Numbers can be ordered, so I can put them in a special place. In this special place, words like Dog and Dog-food are very close, numerically speaking, because they have a related meaning.
So, models find the word with the highest probability of being the next one in a phrase and put them together. This is why LLMs are probabilistic models.
Now go and teach a probabilistic model about humor, or sadness, or why on a rainy day, seeing somebody falling in a puddle and not getting hurt can stir up a good laugh. Yeah you can’t teach that to a model, even if you increase temperature.
Temperature is how you define the level of randomness in the text generated by the model. You can adjust temperature and the answer will be “wilder” - but not necessarily funnier, or better.
RAG can help us give specialized knowledge to an LLM. Check out this chatbot that I had built to explain how to sell on Amazon: Thufir (I called it Thufir because a human computer in Dune is called Thufir).
Chatbots are perfect cases for RAG applications. As a user, I am not looking for creativity there, just answers to my questions.
Don’t use RAG for creating nice articles! Use it for documentation, customer service or other functional applications. Creativity is still human. Not for long maybe. But it still is 🙂
Online Content and AI
It is inevitable: we will see more and more AI generated content. I guess that it is just a matter of time before we become proficient at spotting it. After all, online ads were at first very crude and stupid but as we grew bored with them marketers came up with new formulas and better ads.
So it’s just a game of cat and mouse. AIs will get better at generating online content, and it will be perfectly tailored to the masses. We will get better at spotting such content. It ill go on for some time until things really change.
Happy Monday!