Artificial intelligence is now transforming almost every aspect of our daily lives. It writes texts, generates images, answers questions, filters information, supports programming, and increasingly influences education, media, communication, and work.

Everyone talks about AI. But only few people understand what actually happens in these systems. And even many experts are increasingly reaching their limits. Because modern language models no longer consist of a few clearly comprehensible rules. They consist of billions of trained parameters, high-dimensional vector spaces, and dynamic interactions. Systems that can be described mathematically, but can hardly be fully grasped intuitively.

We understand the formal processes. But we often no longer fully understand the overall dynamics that emerge from them. That's exactly what makes modern AI both fascinating and hard to grasp.

When Simple Rules Suddenly Create Something New

This phenomenon is not new. In nature, it's called emergence: from the interplay of many individual components, properties arise that can hardly be derived from the individual parts alone.

  • A single neuron does not think.
  • A single wing does not fly.
  • And no single component lifts an airplane into the air by itself.

Only the interplay creates something new.

Throughout history, humans have repeatedly harnessed such emergent processes, often most successfully when nature served as a model. Airplanes are modeled on the aerodynamics of birds. Neural networks, in turn, are loosely modeled on biological nervous systems.

Modern AI also does not arise from a single 'intelligent' mechanism. It emerges from the interplay of many mathematical concepts:

  • Weights
  • Training
  • Parallelization
  • Vector spaces
  • Attention
  • Probabilities
  • ... as well as vast amounts of data

The individual building blocks can usually still be followed. But from their interaction emerges an overall system whose behavior becomes increasingly difficult for even experts to grasp intuitively. And it's precisely at this point that modern AI begins to seem almost 'alien,' even though it's based entirely on mathematical and physical processes.

The Problem of Invisibility

When I delved more deeply into the technical fundamentals myself, it was surprisingly difficult to find truly comprehensible explanations. Many resources simplify the systems so much that essential mechanisms are lost. Others are aimed almost exclusively at mathematicians or computer scientists. Many videos, in turn, often move too quickly over central processes. And written texts quickly hit conceptual limits: a single sentence can already raise five new questions that remain unanswered.

  • What exactly are 'weights'?
  • Why does a network suddenly recognize patterns?
  • How does attention really work?
  • Why do language models need GPUs?
  • How does language suddenly emerge from probabilities?

Many of these processes are difficult to understand because they're dynamic. You need to see them.

Why I Developed Animations for This

That's why I started visualizing the fundamental mechanisms of modern AI systems as interactive, time-controllable animations. Not as marketing graphics. Not as abstracted 'AI magic.' But as an attempt to make inner processes visible that are normally invisible.

Like under a microscope, the animations make visible what normally remains hidden. The animations show step by step:

  • how artificial neurons process signals,
  • how training works,
  • why GPUs are so important for AI,
  • how attention calculates meaning between words,
  • and how language models generate new text token by token.

Many of these processes normally happen in fractions of a second and remain invisible. Through time control, you can pause the processes, rewind them, and observe them calmly. Exactly what I wished for back then.

Understanding Does Not Mean Being Able to Fully Explain Everything

We can describe the mathematical operations of modern AI systems exactly and formally. But from billions of interactions emerge dynamics that can no longer be grasped intuitively like classical machines.

AI is not magic. Ultimately, it's 'just' human behavioral patterns cast into mathematical concepts. And yet from this emerges something whose overall dynamics increasingly exceed our imagination.

My goal is not to fully 'decode' AI. I even believe that an important part of modern AI consists of understanding and acknowledging where our intuitive imagination reaches its limits.

Rather, my goal is to lead to exactly this boundary of thought. Because once we've climbed the mountain, we gain a better overview. We see farther. And we recognize where the boundaries of our understanding lie. We may not go higher. But in my view, there is no better starting point for human decisions than this.

The animations are meant to make this climb as easy as possible. Because a society increasingly shaped by AI should at least have an idea of what it's talking about.