home

Leo Pharma GenAI Training
June 30, 2025
Welcome to your GenAI training portal. This is today's training starting point. We have designed a dynamic learning environment that will adapt and expand based on your feedback and needs.
You will be able to access all data and exercises after training if you want to revisit any material or exercises.
Objective:
To inspire and empower GCD leaders (and IT colleagues) to deepen their understanding of AI's potential and applications, while fostering individual initiative and accountability for developing AI skills and applying them in Clinical Drug Development.
The aim is to:
Gain hands-on experience with our LEO internal AI tools and explore their practical applications.
Understand how AI can enhance efficiency in areas such as communication, analysis, and decision-making.
Identify specific use cases where AI can be applied in your role.
Commit to incorporating AI into at least one area of your work and share learnings with the group.
About your instructor - Torben Madsen (Click to expand)
Education
Graduate Diploma in IT Engineering Management, DTU - Technical University of Denmark (2019-2023)
Experienced consultant and instructor specializing in AI implementation, with focus on integrating ChatGPT and other generative AI tools into business processes. Proven track record of creating practical AI solutions that drive efficiency, improve customer service, and generate new revenue streams.
Key AI Expertise
ChatGPT implementation in business contexts
AI chatbot development
AI-enhanced training and educational methods
Process automation through generative AI tools
Skills
Artificial Intelligence implementation
ChatGPT and generative AI tools
Business process optimization
Change management for AI adoption
AI-enhanced training development
Agile methodologies
Certifications
Professional Scrum Master II (PSM II)
Six Sigma Black Belt
Professional Scrum Product Owner II (PSPO II)
www.linkedin.com
Workshop rules
Please ask questions as we go.
If you are stuck or if it's confusing, let us know.
Wait for the long discussions until after the workshop.
The workshop is hands-on - just copy and paste from this website (don't write by hand).
Help me when I ask for check-ins.
Thumbs up
All is good
Thumbs down
I need time or help
Safety
It is crucial to protect confidential data when working with AI services. Even when using approved internal AI tools, never share sensitive information such as:
Customer data and personal information
Business-critical information
Employee data
Important Security Guidelines:
Use only IT-approved AI tools and platforms
Be aware of where your data is processed and stored
Always follow the company's AI policy
Contact the IT department if you are unsure about which information you can share
Remember: Even with internal tools, you must always comply with data protection regulations and exercise caution with sensitive information.
This also means that in all processes where you use AI, you are the final approval step. 
We call this "Human in the Loop."
AI as the great explainer
AI can adjust the explanation level to your needs and is endlessly patient.
Exercise - 15 min.
Try the following prompts - one at a time:
Copy the following to your AI model: "Explain what an LLM like ChatGPT is as if I were 10 years old."
Copy the following to your AI model: "Explain what a prompt is using a Godfather analogy."
Please reflect for 5 minutes - if you and your team are able to learn faster and more, what possibilities will that open up?
Let's establish your baseline.
20 min. exercise.
Copy the following prompts: (One at a time)
1) "On a scale from 1 to 5, where 5 is expert, I consider myself to be a (write nr.).
What are your recommendations for me to get better at GenAI? I have access to the following OpenAI models: 4.1, 4o, 04 mini, 03"
2) "I work as a (list job title), and some of my many tasks are (write some of the tasks you spend the most amount of time on or you think are the most dull). How does this information help you?"
3) "Please be precise, give me a specific training plan for the next 10 days. I wish to spend x amount of minutes every day. Please ask clarifying questions before you answer, including asking me what capabilities I would like to learn."
We control how we use AI - not the AI itself
It is not "classical" (deterministic) software as we know it, but is built with our brains as inspiration - a neural network. The AI has been trained, not programmed - just like a human. In fact, writing to AI as an assistant works best; some of the most effective users are managers and parents (they are used to giving tasks to other people).
Copy this prompt: "Explain why GenAI gives different answers, even though my colleague uses the same prompt as I do?"
Although we cannot control the development of the AI technology itself, we have full control over how we implement and use it in our work. It is about making conscious choices about when and how AI tools can most effectively support our processes. By focusing on the application rather than the technology itself, we can ensure that AI becomes a valuable partner that supports - not replaces - our expertise and judgment.
It's normal to forget that you have access to super capabilities.
You are just a browser away from accessing billions of operations of processing power per second.
You with AI will almost always beat you without AI.
If unsure, AI can help. Ask your model: "I am about to do the task (task description). Explain how you could assist me in increasing complexity."
15 Times to use AI, and 5 Not to
The Art of Using AI: A Practical Guide
Please spend 5 minutes reviewing the headlines in the article (click the toggle).
Ask if you have any questions.
When to Use AI:
To Generate a Wealth of Ideas: AI can produce a vast number of ideas, which is invaluable for brainstorming. The sheer quantity increases the chances of finding a truly innovative concept.
When You're an Expert: If you have deep knowledge in a subject, you can quickly assess the quality of AI-generated content and use it to your advantage.
For Low-Stakes Summarization: AI is excellent at summarizing large volumes of text where minor errors are not critical.
To Repurpose Content: AI can efficiently adapt a single piece of content for various audiences and formats, such as turning a policy document into multiple training materials.
To Overcome Creative Blocks: When you're stuck, AI can provide numerous suggestions to get you moving forward, even if you don't use the ideas verbatim.
When AI Outperforms Available Humans: If an AI model is demonstrably better than any human expert you can access, it's a logical choice, provided its potential errors aren't catastrophic.
As a Learning Companion: Use AI to ask endless questions and gain context while reading or learning about a topic.
To Explore Diverse Solutions: AI can generate a wide variety of creative and stylistic options for a task, allowing you to curate the best one.
For Proven Use Cases: In fields like coding, where research has shown AI to be a significant help, it's wise to use it.
To Test Reactions: AI can simulate how different audiences (hostile, friendly, naive) might react to your work.
For Entrepreneurial Endeavors: AI can act as a versatile co-founder, assisting with tasks across various disciplines where you lack expertise.
To Gain Specific Perspectives: You can use AI to generate feedback from simulated personas with particular viewpoints.
To Automate Trivial Tasks: AI can handle mundane, ritualistic work that serves little real purpose, freeing up your time for more valuable activities.
For a Second Opinion: You can present data to an AI to see if it reaches the same conclusions as you.
For Tasks AI Simply Does Better: This category is rapidly expanding as AI capabilities grow.
When NOT to Use AI:
For Deep Learning and Synthesis: To truly learn and internalize new information, you need to do the reading and critical thinking yourself.
When High Accuracy is Crucial: AI can "hallucinate" and produce plausible-sounding falsehoods that are difficult to detect.
If You Don't Understand Its Flaws: Be aware of AI's failure modes, such as its tendency to be overly persuasive or sycophantic, before relying on it.
When the Effort Itself is the Goal: The struggle to understand a concept or create something new is often where the real learning and breakthrough moments happen. Don't shortcut this process.
When the AI is Simply Not Good Enough: AI has a "Jagged Frontier" of abilities—it excels at some unexpected tasks while failing at seemingly simple ones. It's crucial to discern where it is and isn't effective.
In essence, knowing when and how to use AI is a form of wisdom. It requires understanding that AI is a powerful tool best used to enhance our own expertise, not replace the essential work of critical thinking and deep learning.
Source
Tips for Using Language Models
Context Window
Maximum text an AI can process in a conversation. Larger windows allow for detailed prompts and referencing earlier information. Be sure to use models with large context windows when you have a lot of data to process—otherwise, the model may answer less accurately.
Hallucinations
AI-generated false information that appears credible. Always verify important facts from authoritative sources.
Temperature Setting
Controls randomness in outputs. Low (0-0.3): factual, technical responses. High (0.7-1.0): creative, varied outputs for brainstorming.
Prompts
Instructions guiding AI responses. Clear objectives and specific context yield better results. Consider role-playing or step-by-step formats.
Understanding these concepts helps you leverage language models effectively while maintaining safety standards at Leo Pharma.
Tips for using language models
1
Be curious
Test what the AI can do. It's okay to ask all kinds of questions.
2
Experiment
You learn best by practicing. Try different ways of asking questions.
3
Improve
Start simple and refine your questions based on the responses you get.
A meta prompt is often a good way to start: "I want help from you to resolve x issue. Ask the most important questions before you provide the answer."
GenAI models make educated guesses (and sometimes get it wrong)
They are prediction machines.
Tip: If you receive an inaccurate or unsatisfactory response, ask the AI to try again with more specific instructions. For example, give it more context. Each attempt can yield improved results as the model processes your feedback. Remember that persistence and clear communication lead to better outcomes.

Try asking the AI: "I need to communicate that the project is late. Please provide 3 examples of short statements. Write them in a table."
When the AI answers, respond with: "Try harder, give me some more examples."
How an LLM works (click to expand)
Loading...