Artificial Intelligence (AI), especially Machine Learning (ML), is a transformative technology that can be integrated into a variety of software applications in the near future. However, this brings not only the promised benefits but also significant risks.

This workshop covers the fundamental terminology and concepts of modern AI. It demonstrates how Artificial Intelligence can be applied and critically evaluated in various work environments. The seminar addresses important aspects of AI projects, highlight fundamental weaknesses of AI models, and point out dangers to avoid.

Additionally, a current selection of existing AI services is presented, and their typical use cases explained. The focus is on the basics and workflow of machine learning. The knowledge conveyed is therefore largely AI provider and technology stack independent and generally applicable.

The course covers both current hype AIs like ChatGPT and GPT4, as well as models from other application areas, such as image recognition and processing. Practical exercises include not only a reflection of the learned theory but also an introduction to the use of OpenAI models. The workshop is continuously supplemented with current developments and refers to them.

Agenda

What is Artificial Intelligence?

• Examples of Artificial Intelligence that illustrate the breadth of applications
• Supervised, Unsupervised, and Reinforcement Learning
• Generative vs. Predictive AI
• Embeddings and Vector Databases, e.g., for AI-based search applications
  
• Case Study: How DeepBlue defeated the chess world champion
• Practical Task: Recognize application areas in your own industry

Introduction to Large Language Models (LLM)

• How can ChatGPT etc. generate text?
• System and User Prompts
• What are the fundamental weaknesses of this approach?
• Randomness and Creativity in LLMs
• Use of non-public and current data with Retrieval Augmented Generation (RAG)
  Multi-Modal LLMs
• Practical Task: Prompting OpenAI’s LLMs

From Linear Regression to Artificial Neural Networks

• How machines learn from data
• What is an Artificial Neural Network?
• Application examples of Artificial Neural Networks
• Strengths of Artificial Neural Networks and Artificial Intelligence

The Machine Learning Workflow

• Training, Validation, and Testing of Models
• Finetuning existing models
• Transfer Learning from existing models
• Practical Task: Hallucinations of large language models

Weaknesses of Artificial Neural Networks

• Uncertainties in predictions
• Explainability of predictions
• Problems with unknown data
• Bias
• Practical Task: Live-Jailbreaking (“Hacking”) Google’s image generator*
  * Execution and success cannot be guaranteed.

Economic Aspects of Artificial Intelligence

• Costs in Machine Learning
• Self-improving systems and AI economies of scale
• Legal and Ethical Questions without obvious solutions
• Case Study: Tesla’s self-driving cars

 Brief overview of known services

• Text and Multi-modal Models (GPT variants, LLaMA…)
• Image generation models (DALL·E, Stable Diffusion, Imagen…)
• Transcription models (Whisper)
• Simple and inexpensive alternatives (FastText, Pre-trained Image Recognition…) 

Target Audience and Requirements

Apart from a certain affinity for information technology, the course has no prerequisites. It is primarily aimed at decision-makers who are considering integrating AI components and want to better assess the potential and risks.