· 2 min read · Analysis
AI Plays with Entropy to Recreate Reality
The fundamental process behind generative AI is not just advanced technology, but a philosophical recreation of how we perceive and model reality.

Contrary to popular belief, neural networks don't actually work like the human brain. But modeling reality by subtracting and adding entropy does mimic aspects of how we think.
The Dance of Construction and Deconstruction
Generative AI works in a two-phase process:
First, it deconstructs: It focuses on removing noise from signals, encoding only essential information. It reduces reality to its minimal expression necessary to recreate it—creating models (the famous "latent space"). This is called "training."
Then, it reconstructs: It deliberately adds noise (entropy) to create "new" signals or predictions from those models.
The result is as if large portions of reality were simply a particular point of view—some "coordinates"—in a multidimensional space that contains everything at once.
Why LLMs are not "Stochastic Parrots"
Here's the touch of genius that many overlook: randomness is precisely what separates LLMs from being mere repeaters.
Without entropy (randomness), then they would indeed be repetitive—they would have zero entropy and would systematically repeat the information they were trained on.
The paradox is that it's controlled chaos that allows creativity and variation in responses.
What This Means for Your Business
If you understand this process, you can leverage AI in a much more sophisticated way:
-
For marketing: Generating different versions of the same message, and running A/B tests to know what works best.
-
For analysis: Use AI to find the essential patterns (low entropy) in your complex data.
-
For innovation: Generate new ideas even if at first glance they may seem absurd, that serve to see things from new perspectives.
The Philosophy Behind the Technology
What's interesting is that this process—reducing entropy and increasing it to generate variations—is how humans solve problems and understand the world. First we understand, and based on that process we can predict what will happen next.
Generative AI is not simulating intelligence. It's recreating the fundamental process of how information organizes and reorganizes to create meaning.
More updates
· Engineering
The Case for AI-Native Companies
AI-native companies produce roughly 100x the output of legacy firms that adopt AI on top of old hierarchies. Here is why organizational design, not software, is the deciding factor.
Read update · 6 min read
·
GTM Engineering and Growth Engineering Are Not the Same Role
GTM engineering automates revenue systems outside the product. Growth engineering runs experiments inside it. Here is where each discipline starts and stops.
Read update · 3 min read
· Product
Design of a webapps factory as a separate project
Design work on a factory for small, niche web apps. Two conclusions: it has to be a separate product from our client platform, and the slow part is not the code.
Read update · 1 min read
