Artificial Intelligence (AI) refers to computer systems that can perform tasks that would normally require human intelligence. This includes things like understanding text, recognizing images, making decisions, or translating language.

Important: AI does not “think” like a human. It recognizes patterns in vast amounts of data and derives predictions from them. When ChatGPT writes a sentence, it doesn’t “know” what it’s saying—it calculates which word is most likely to come next.

None. KI stands for “künstliche Intelligenz”—the German translation of “Artificial Intelligence” (AI). Internationally, AI has become the standard term, which is why we at nuwai use it as well.

Machine Learning is a subset of artificial intelligence. Instead of telling a computer step by step what to do (classical programming), you give it data—and it independently learns patterns and rules from it.

An example: Instead of manually programming 500 rules to detect spam emails, you show the system thousands of emails marked as “spam” or “not spam.” The system recognizes on its own which characteristics are typical of spam.

Deep Learning is a specialized form of Machine Learning that works with so-called neural networks—structures loosely inspired by the human brain. “Deep” refers to the many layers in these networks.

Deep Learning is the technology behind the major breakthroughs of recent years: speech recognition, image generation, text comprehension. All modern AI systems like ChatGPT, Claude, or Gemini are based on Deep Learning.

A neural network is a mathematical model organized in layers. Data flows from the input layer through multiple hidden layers to the output layer. In each layer, the data is transformed and weighted.

You can think of it as a multi-stage filter: Each layer extracts increasingly abstract features. In image recognition, the first layer detects edges, the next shapes, then facial features—and finally a face.

Training is the process by which an AI model learns from data. The model is shown vast amounts of text, images, or other data. The model gradually adjusts its internal parameters to better recognize patterns.

Training modern language models takes weeks to months, requires thousands of specialized graphics processors (GPUs), and costs tens to hundreds of millions of francs. That’s why this step is only performed by large companies like OpenAI, Google, or Anthropic.

An AI model is the result of training—comparable to the “brain” of the system. It contains billions of parameters (numerical values) that were optimized during training.

Well-known models include GPT-4o (from OpenAI), Claude (from Anthropic), Gemini (from Google), or Llama (from Meta). Each model has its own strengths: Some are better at programming, others at creative writing or analysis.

Parameters are the numerical values within a model that are adjusted during training. You can think of them as control knobs: The more parameters, the more finely the model can capture nuances.

GPT-4 has an estimated over 1 trillion parameters. Smaller models like Llama 3 have 8 to 70 billion. More parameters don’t automatically mean “better”—but generally more capabilities and nuanced understanding.