A beginner’s guide to generative AI

We’ve all heard of AI, machine learning and ChatGPT. But how does it really work? Here’s a beginner’s guide to the technology behind – and what might come next.

By Sven Størmer Thaulow
A beginner’s guide to generative AI

A beginner’s guide to generative AI

We’ve all heard of AI, machine learning and ChatGPT. But how does it really work? Here’s a beginner’s guide to the technology behind – and what might come next.

By Sven Størmer Thaulow

In today’s rapidly evolving ­digital landscape, buzzwords like “AI” or “machine learning” are becoming increasingly common. Even if you’re not entirely sure what they mean, chances are you’ve encountered them in some form or another, perhaps through smartphone assistants like Siri or Alexa or in online customer service chats. However, a subset of AI, known as generative AI, is emerging as a transformative force in the digital world. Here’s a closer look at this technology and its implications for the future.

Unravelling the Mystery of Generative AI

At its core, generative AI is about ­creation. Much like an artist creates a painting or a writer crafts a story, generative AI can produce new content. But instead of paint or words, its tools are data and algorithms.

Imagine having a conversation with a friend about your favourite books. As you talk, your friend might suggest a new book for you to read based on what you’ve mentioned. Generative AI operates on a similar principle. Feed it with enough conversations about books, and it could suggest or even create a synopsis of a book that doesn’t exist but fits within the parameters of the conversations it’s analysed.

The Magic Behind the Screen

The magic of generative AI lies in its ability to produce content, be it text or images. But how exactly does it do this?

For text, generative AI models are trained on vast databases of ­written content. They analyse patterns, contexts, and structures within these texts. When given a prompt or starting point, they use this training to predict and generate what comes next. It’s like teaching a child to speak by immersing them in conversations until they start forming their own sentences.

On the image front, things get a bit more complex. Techniques like Generative adversarial networks (GANs) are often employed. Here’s a simplified explanation: imagine two AI systems – one is the artist (generator) and the other is the critic (discriminator). The artist creates a picture, and the critic evaluates it. If the critic can easily tell it’s a generated image and not a real one, the artist tries again. This back-and-forth continues until the artist produces something the ­critic can’t distinguish from a real image. Through this process, the AI becomes adept at creating realistic images.

Societal Impact and the Media Realm

The proliferation of generative AI doesn’t merely affect technological circles; its ripples will be felt across society. As AI-generated content becomes commonplace, our ability to discern between human-created and AI-created material might blur. This poses profound questions about ­authenticity, trust and the ­value of ­human creativity. For media companies like Schibsted, the implications are vast. On one hand, AI can generate news reports, write articles or even create visual content at a pace unmatched by humans, offering efficiency and cost savings. However, this also brings challenges. How do media houses ensure the ­credibility of AI-generated content? And as ­audiences become aware of AI’s role in content creation, how will this shape their trust and engagement with media outlets?

Charting the Evolution of Generative AI

Like all technologies, generative AI wasn’t born overnight. It’s been a product of years of research, improvements and refinements. As computational power increases and algorithms become more sophisti­cated, the capabilities of generative AI expand. Currently, we’re witnessing AI that can draft articles, compose music and generate artwork. Yet, this is just the beginning. The trajectory suggests
a future in which generative AI can create more complex, interactive, and nuanced content. Think of virtual realities indistinguishable from our own, or digital assistants that not only understand our preferences but can also predict our needs before we articulate them.

The Next Wave of Breakthroughs

Predicting the future is ­always a gamble, but based on the current momentum, several ­exciting developments appear on the horizon for ­generative AI.

  • Personalised content: In a world saturated with content, persona­lisation is becoming paramount. Generative AI could craft experiences tailor-made for individuals. Imagine a movie that adjusts its storyline based on your preferences or a video game that evolves based on your playing style.
  • Education revolution: Customised learning isn’t new, but with generative AI, it could reach unprecedented levels. Students might have access to study materials created on the fly, precisely addressing their weak points and reinforcing their strengths.
  • Artistic collaboration: While some fear AI might replace human artists, a more optimistic view is a ­future where artists and AI collaborate. An AI could suggest melodies for a musician or sketch ideas for a painter, enriching the creative process.

In conclusion, generative AI lies at the intersection of art and ­science, holding the promise of a world where techno­logy enhances creativity, personalisation, and efficiency. At Schibsted we feel we are on the cusp of this new era, and that it’s crucial to approach it with both excitement and caution. We must ensure that as we harness its potential, we also ­consider the ethical impli­cations of AI shaping our reality.

Using an AI co-pilot: How did I make this article?

This article was a ­classic task for generative AI as it was a fairly generic piece more ­describing a well-known domain rather than being a very personal and opinionated article – so I used ChatGPT as a co-writer. I tried out a prompt describing the article I wanted. It became very “chatGPT-ish” – lots of numbered bullets with sentences. So I tried again with a prompt saying I wanted it in “New York Times” style. I got closer. I tried some more prompt ­variations and also limited it to the number of words. When I had the 80% text I wanted I started rewriting somewhat, cleaning up mistakes and adding some elements. And voila – a pretty decent article was born!

Sven Størmer Thaulow

Sven Størmer Thaulow
EVP Chief Data and Technology Officer, Schibsted
Years in Schibsted: 4
My favourite song the last decade: Thinking of a Place – The War On Drugs