From Lorem Ipsum to Magic: When Technology Becomes Indistinguishable from Content
(This article has been published on Medium)
Image generated with Nano Banana in October 2025. The original image of a panda with a red hat is lost in time, like tears in rain.
A panda with a red hat.
“I want to show you something cool”, I say to Pablo, my colleague. I take out the phone from my pocket and quickly type the first thing that comes to my mind — “panda with a red hat”. Magically, a colourful image of a cute panda with a red hat appears on the display.
It is late 2022, we are sitting on a train headed to Utrecht, and the app that generated the panda is the second version of DALL-E, released a month earlier. DALL-E is a tool called text-to-image model, which uses AI to create a picture from a written description. That version of DALL-E was the first model of that type able to generate realistic images. I was familiar with new technologies, but that one stood out as very different from those I had used in previous years, and I really wanted to share it with someone.
It’s a kind of magic.
The moment I write is late 2025. Literally dozens of new models have been released since that day on the train to Utrecht, each of them more sophisticated. The models have introduced new capabilities: converting text to videos, creating podcasts, writing long research reports, composing music. Arthur C. Clarke wrote in 1973 that “Any sufficiently advanced technology is indistinguishable from magic”, and it feels indeed like AI models have become sufficiently advanced, because what they do really looks like magic.
But it’s actually not magic under the hood. All the AI models share a common methodology; we refer to it as “Generative AI” that consists of a set of advanced numerical techniques based on complex statistical modelling. These models rely on statistics to learn patterns in huge amounts of data, and they use mathematical techniques to adjust and improve their guesses over time. Methods like trial-and-error learning, pattern recognition, and probability calculations help them generate text, images, or sounds. Although the underlying methodology hasn’t changed much in the recent years, what has dramatically increased is the power and scale of these models.
Left: Image generated by AlignDRAW with the prompt “A very large commercial plane flying in rainy skies.” AlignDRAW is described in the article as one of the first text-to-image AI models, created in 2015 by Elman Mansimov. Right: Image generated by ChatGPT in October 2025 using the prompt “A very large commercial plane flying in rainy skies.”
The evolution of Generative AI has been so fast that during each step it has become harder and harder to find a rational link between the mathematics behind the new models and their ability to create. As the models grow in size and complexity, their work becomes harder to understand, even to experts. While we can describe the algorithms and training processes, it’s complicated to understand the reason why a model generates a particular image or sentence. In that sense, what once was considered pure engineering now increasingly resembles a kind of controlled magic.
Image generated by Midjourney in March 2025. Image courtesy: Ruben Alba (https://rubenalba.artstation.com/)
How can Generative AI create text? First, the generative AI breaks down the request from the user into individual elements (words or short sentences, called tokens). These tokens are used by the underlying large language model (LLM) to generate a reply based on the statistical relationship between words. In fact, LLMs are created using a huge amount of text data: the amount of text needed to train GPT-4 (the model used by ChatGPT) is estimated to be in the order of hundreds of millions of books.
How can you generate text using statistics? The model consists of trillions of numbers representing the patterns and relationships in the language data the model has been trained on: in other words, how words and phrases are typically used together.
We can assume that if a model is powerful enough (namely, it contains enough numbers to represent all the language patterns) and it has been trained on good and relevant text, its output can be coherent: it will be written using correct words in the right order.
GenAI killed the Lorem Ipsum
The increasing power of AI tools makes it possible to generate a page of text in seconds, a high-quality artistic image in less than a minute. These actions can be performed with the swiftness of a magic spell by a wizard in a fantasy movie. Basically, with a few magic words (the prompt) and a keyboard, you can fill any blank space with content in a few seconds.
A white blank page has always been the starting point of any creative process. A white space indicates that something is missing, serving as a sort of proxy for a lack of knowledge or content. For many centuries, the empty areas without content in typesetting (digital and pre-digital) could be filled using a peculiar text called “Lorem Ipsum”. This filler text consists of gibberish Latin-resembling words starting with “Lorem ipsum dolor sit amet”. It was used to see how a page would look with text before inserting any content.
What about today? If you want to see how a page would look with actual content… you can fill it with AI-generated content that, as said before, can be generated in the blink of an eye. Type your prompt, for example “Create a wiki page based on this specific file”, and in a matter of seconds your page contains a wiki page based on your instructions.
Generative AI is a Lorem Ipsum on steroids. Lorem Ipsum is used to fill empty spaces as a dummy content and generative AI is its evolution: it does not only serve as filler, but it actually makes sense. Today, Generative AI can fill white space instantly with coherent text on any topic.
Page generated in Microsoft Office 365 using Copilot with the prompt “Create a wiki page for the city of Rotterdam”
Don’t get tricked, your report still needs filling
We have spent some time discussing how Generative AI can fill a blank page with content, but we haven’t said yet which pages we’re actually talking about. While AI has many potential uses, here I am focusing on its use to create documents like reports.
A report provides to the readers all the facts that they should know on a specific topic — in other words, what is relevant — conveying a specific message. Generative AI instead provides text based on statistical connections between words only, without any awareness on the relevance or importance of its output.
With its incredible — I would even say, magical — ability to learn how all the words are frequently associated, Generative AI can indeed answer any question with perfect-looking text, perfect grammar, brilliant style — even jokes and puns. However, we don’t know if the output is relevant to the question because the answers are based on statistical correlations between words, without any actual understanding.
The incapability of understanding the relevance is a big problem when we use Generative AI to produce knowledge: you will never know if something very relevant is missing. For example, if you ask a generative AI to summarise the most important advances in space exploration in the last 50 years you would probably spot if the International Space Station is missing from the answer: you know that it should be there because it is very relevant to the question. But what if you ask to write something on a topic you don’t know at all? You wouldn’t be able to notice relevant points that are missing.
Ben Thompson in his blog writes a long post about the “illusion of knowledge” that can be delivered by Generative AI, showing an example where the answer from a Generative AI misses relevant points. Actually, we often read reports about things we do not know very well (or we do not know at all), and we expect to find in a report all the things we need to know on a specific topic. A team of researchers, led by Walter Quattrociocchi, gave also a specific name to this phenomenon, they called it “epistemia”, defined as “statistical plausibility risks replacing deliberative reasoning with the illusion of knowledge rather than its verification”.
We said before that Generative AI is a sort of evolved version of the Lorem Ipsum text, a filler with coherent and well-written sentences. However, “all that glitters is not gold” — the content might look impressive but yet lacking real substance, and spotting this illusion can be difficult.
The End of Lorem Ipsum, The Start of Something Bigger (?)
To wrap up. Generative AI gives us the possibility, for the first time in human history, to quickly — magically — generate content. This content, that can either be the illustration of a cute panda or a page summarising the first season of The Last of Us, is often good enough to give the impression that the blank page is instantaneously filled with useful content. Actually, what is produced by Generative AI is more than filler; it is well-written and perfectly coherent text that has basically transformed the blank page at the beginning of a new document, a legacy of the past.
A new tool always brings both positive and negative effects. When the printing press was invented in the 15th century, it helped to increase literacy, but also enabled — for example — the rapid spread of misinformation. Similarly, the rise of Generative AI is having a transformative effect — it allows us to fill blank pages nearly instantly, no longer constrained by the tedious process of manual content creation. Just as the printing press disrupted existing industries and created new social divisions, generative AI is reshaping how we work and communicate.
The death of the blank page may be liberating in many ways, allowing ideas to flow more freely, but it also poses new challenges.
The main problem is that the output of generative AI can be simply mistaken for genuine, useful knowledge. As I wrote earlier in this article, these AI systems can generate factual statements that appear coherent yet lack true understanding or relevance to the task at hand. If the writer is not careful, they may overestimate the knowledge held in the words filling the page.
Generative AI is becoming a staple tool for all knowledge workers and some of its effects can be easily classified as positive or negative. But others, like the sudden death of the Lorem Ipsum, still require some thinking.
(Big thanks to Lan Chu for the great conversations and for all the helpful feedback on this article. Plus, shout-out for sharing this awesome blog post from Anthropic — it’s a perfect fit here!)
