AI
19. juni 2025
What is an AI model?
In its simplest form, an AI model is a set of rules and patterns that a computer has learned, typically using large amounts of data. It is a bit like a recipe: it learns to recognise what typically follows something else, and that is how you can “train” the AI.
Indhold
Introduktion Eksempler på modeller Hvad kan de bruges til Fordele og ulemper Sådan kommer du i gang Beskrivelse af modeller FAQAn AI model is a recipe that learns from data: it helps with text, images, analysis, and much more. You get the most value by starting simple—use ready-made tools, ensure human oversight, and work iteratively with your colleagues.
When you train an AI (most often using large amounts of data), it learns to recognise what typically follows something else.
For example, a model can learn that “horse” is often followed by “gallop” in a text, or how an image of a cat differs from a dog. Once the model has practised enough, it can predict, provide answers, or create something creative based on what it has learned.
Examples of AI models
Here are some well-known models:
| Model/tool | Developer | What it is good for |
|---|---|---|
ChatGPT | OpenAI | Conversations, text, ideation, brainstorming, explanations |
| Claude | Anthropic | Long texts, empathetic language, collaboration, safety and ethics |
| Gemini | Google DeepMind | Multimodal AI (text, image, audio), search, coding, analysis |
| DALL·E | OpenAI | Creating images from text—creative and illustrative |
| Midjourney | Independent team | Aesthetic, atmospheric images, art and branding |
| Stable Diffusion | Stability AI | Open-source image model, extensive visual control, local deployment |
| TensorFlow | Train and run AI models, image recognition, production and scalability | |
| PyTorch | Meta | Flexible AI development and research, text and image models |
| GitHub Copilot | Github + OpenAI | Helps with coding directly in the editor, auto-suggestions and code understanding |
Read more about each model further down, or by clicking the model name.
You can read more about our view on the use of AI, which tools we use and what for, in this post.
What can you use an AI model for?
Chat and customer service
An AI model can live in a chat on your website and help with customer questions—24/7, without breaks.
Text generation and writing support
Do you ever think: “How do I start this email?”—an AI model can suggest an opening, a headline, or summarise texts for you.
Image and pattern recognition
For businesses with images: models can sort images, find defects, or recognise objects.
Data analysis and predictions
Do you need to forecast sales, inventory needs, or risk? AI models can help you identify patterns in the numbers.
Automation
Repetitive tasks—e.g., reading forms, retrieving information, filling in fields—can be delegated to AI models.
Creative use
Music, illustrations, scriptwriting—models can help with ideas or even carry out part of the creative work.
Pros and cons of using AI models
| Advantages | Cons |
|---|---|
| ✅ Save time on routine tasks | ❌ Can fail or misunderstand—requires oversight |
| ✅ Scales—responds 24/7 | ❌ Can be expensive to train or use at scale |
| ✅ Supports creativity—ideas and inspiration | ❌ Risk of bias—the model learns from our data and can repeat mistakes |
| ✅ Automates error-prone tasks | ❌ Can feel like a “black box”—you do not always know how it arrives at its answers |
| ✅ Increases your capacity and therefore your competitiveness | ❌ AI requires significant server power and therefore has a large CO2 footprint |
Best practices: how to get started smartly
- Start simple
Use ready-made tools (ChatGPT, Copilot, Zapier) rather than training your own. You will get value quickly—without the hassle. - Data and responsibility
Consider what data you use and who has access. Remember GDPR and data protection. - Critical and human oversight
Never let the AI model stand alone—an employee must check the output and quality so mistakes do not slip through. - Iterate and adapt
Start with small pilot projects and evaluate: What works? What creates value? Then you can scale up quickly. - Build user trust
Tell the user when it is AI: “This answer was generated with support from an AI model.” Transparency creates confidence—both internally and externally. - Follow-up and learning
Gather feedback: What works best? What are the challenges? Use the experience to improve the process.
Description of different AI models
ChatGPT (from OpenAI)
ChatGPT is a language model that understands and writes text. You can ask questions, get explanations, brainstorm ideas, get help with emails, coding, and much more—all in a human conversational style.
What is it good for?
- Brainstorm and collaborate
- Write and improve text (emails, articles, scripts)
- Answer questions and explain complex topics
Claude (from Anthropic)
Claude is an AI model with a particular focus on safety, ethics and collaboration. It is designed to be “helpful, honest and harmless”—part of Anthropic’s approach to responsible AI.
What is it good for?
- Long-context understanding: Claude can remember and work with very long documents (often 100,000+ words at a time!).
- Collaboration: Often used as a kind of “AI colleague” for document work, brainstorming and ideation.
- Empathy and tone: Very good at writing in a warm and friendly tone—e.g., in emails, onboarding, etc.
Typically used for:
Document analysis, creative writing, internal support, responsible AI operations, and HR communication.
Gemini (from Google DeepMind)
Gemini is Google’s latest AI model, building on the experience from Bard (Google’s previous chatbot). It is a so-called “multimodal” model, which means it can understand both text, images, audio and video—and combine them.
What is it good for?
- Search and facts: Gemini is (of course) well integrated with Google’s vast knowledge and search results.
- Analysis and text understanding: Very good at comparing, summarising and analysing content.
- Multimodal input: You can provide both text and images and get meaningful answers.
- Code understanding: Gemini Pro and Ultra are strong at programming and debugging.
Typically used for:
Customer service, analysis, SEO, presentations, visual explanations and coding.
DALL·E (from OpenAI)
DALL·E is an AI model that creates images from text descriptions. For example, you write “an astronaut surfing a wave in space”, and DALL·E paints it for you.
What is it good for?
- Quick drafts for design or concepts
- Creative ideation and visual storytelling
- Illustrations for content, marketing or presentations
Midjourney (independent development team)
Midjourney is also an AI model for image generation—but with its own distinctive style. It is especially used to create artistic and atmospheric images.
What is it good for?
- Aesthetic images with wow factor
- Branding and creative concepts
- Mood boards and visual ideas
Stable Diffusion (from Stability AI + open-source community)
Stable Diffusion is an open-source image generator, which means you (or your developer) can download it for free and control what it should do yourself.
What is it good for?
- Option to run locally (i.e., without the cloud)
- Complete control over image style and output
- Used both creatively and professionally
TensorFlow (from Google)
TensorFlow is an open-source framework for building and training AI models. It is primarily used by developers and researchers who want to create their own AI solutions from scratch—for example, image recognition or text analysis.
What is it good for?
- Can be used for text, images, numbers and audio
- Used to train custom models
- Widely used for machine learning in production (large systems)
PyTorch (from Meta/Facebook)
Like TensorFlow, PyTorch is a tool for developing AI models. Many researchers prefer it because of its flexibility and “human” approach—it is easier to experiment and get results quickly.
What is it good for?
- Many modern AI researchers choose PyTorch
- Rapid prototyping and experiments
- Often used for research in language models and image analysis
GitHub Copilot (from GitHub + OpenAI)
An AI assistant for developers that lives in your code editor. It suggests the next line, functions, and helps you write faster and better code.
What is it good for?
- Save time on routine tasks
- Suggest code as you type
- Understand and explain code blocks
FAQ: Frequently asked questions about AI models
Are AI models safe to use?
Does it cost a lot of money?
Do we need to be able to code to use AI?
Will AI replace employees?
Can AI models be used in all types of industries?
Can we train our own model with our data?
