GenAI

GenAI is an abbreviation for “generative artificial intelligence”. It refers to computer programs that, upon instruction, create something new: a text, an image, a piece of music, or program code. The difference from older programs lies in the word “create”. A calculator delivers exactly one correct result, a search engine delivers existing pages. A GenAI program, on the other hand, assembles its answer piece by piece itself, and does so anew each time. That’s why the same question often gets two different, yet both usable, answers.

Why GenAI has reshaped the tech industry

Before 2022, artificial intelligence was mostly a matter for specialists. It was hidden away in spam filters or in a phone’s facial recognition. With ChatGPT, suddenly anyone with a browser could use an AI without learning anything. You type a question in ordinary language. This low barrier to entry explains why GenAI arrived so quickly in offices, schools, and newspapers.

Economically, a great deal of money now hinges on this. The major providers invest double-digit billions annually in data centers. The chipmaker Nvidia became one of the most valuable companies in the world largely because of this. When stock market news talks about a possible “AI bubble”, it is almost always about GenAI.

At the same time, new problems are emerging. GenAI can fake photos and voices so convincingly that laypeople cannot tell the difference. It learns from texts and images whose creators were often never asked. And it is changing professions in which a lot of writing, drawing, or programming takes place. The EU has passed a law, the AI Act, that classifies and regulates such systems according to risk.

What happens when text and images are generated

GenAI programs learn from huge amounts of examples. This is called training. For text, the task during training consists of predicting the next word each time. The system reads millions of sentences and fills in gaps until it becomes very good at it. Whoever practices this often enough learns not only grammar but also a great deal about the world.

In operation, this prediction simply repeats over and over. The program chooses a fitting word, appends it, and looks again. This is how an answer grows from left to right. With images, it works differently: there, the system starts with an image of pure image noise and removes the noise step by step until the desired motif becomes visible. This process is called diffusion.

An important and common misconception: GenAI does not look anything up in a database. It has no directory of true sentences. It generates whatever statistically fits well. That’s why it sometimes invents sources, quotes, or dates that sound convincing yet are false. Experts call this hallucination. For numbers, names, and dates, you should therefore always check the results yourself.

GenAI in your phone, in the office, and in headlines

You encounter GenAI most directly in chat programs like ChatGPT, Gemini, or Claude. But it has long since found its way into tools you wouldn’t expect. Translating a webpage, summarizing a video conference, autocomplete in a coding environment: the same models work in the background everywhere.

In companies, GenAI is used above all for text work. It answers customer inquiries, summarizes contracts, or produces first drafts of reports. Programmers use it as an assistant that suggests code. The benefit is greatest wherever a human afterward still checks the result.

In the news, you usually encounter the term in one of three roles. First, as a business field, when providers report revenues or investments. Second, as a point of contention, for example in copyright lawsuits brought by publishers and artists. Third, as a security issue, when fake videos of politicians surface. Anyone who knows that GenAI creates content rather than looking it up understands all three debates much better.