AI is often misunderstood as a resource hungry entity and the public debate swings between two extremes. This course offers students a foundation to understand how generative AI works under the hood. Through live demonstrations, practical firsthand examples and hands-on tasks, the participants gain the factual knowledge needed to evaluate tools and claims independently.

What this workshop aims to cover:
- Prevalent misconceptions regarding data, training, privacy and the ecological footprint;
- Popular AI model architectures;
- Different kinds of ‘training’ and their goals;
- Hardware reality and limitations;
- Examples of use cases beyond simple chatting.

Objectives

Remembering: Recalling the technical meaning of the ChatGPT acronym

• Participants can break down what chat, generative, pre-trained and transformer stand for. This provides an immediate, non-intimidating entry point to demystify the technology and properly locate Generative AI within the broader hierarchy of Machine Learning and Deep Learning;

Identifying core computer hardware concepts and terms

• Participants recognize the vocabulary related to running models (VRAM, RAM, memory bandwidth, quantization, context window, ...). This makes abstract software concepts more approachable and clarifies how models require specific hardware rather than just knowing a vague definition of "cloud";

Understanding: Distinguishing between different types of AI models

• Participants can explain the fundamental difference between a base model, a post-trained instruct/chat model, and a LoRA and are familiar with the different possible modalities beyond text;

Applying: Navigating popular online software repositories and platforms. 

• Participants can directly interact with platforms like Hugging Face and OpenRouter. Instead of being passive consumers of commercial chatbots, they can apply their foundational knowledge to independently search for models. 

Analyzing:  Differentiating checkpoints from the software harness

• Participants can deconstruct an AI tool into its distinct layers. They can separate the weights from the software environment that orchestrates model inputs, executes tools, and makes the system appear "smart";

Evaluating: Re-evaluating personal preconceptions about AI

• By comparing their quiz inputs with the results and analyzing the firsthand evidence gathered from live demos and practical examples, the participants become more capable of forming a fact-based judgement of AI capabilities. 

Target audience

Students

Students (BA, MA, Doctorant at Unifr or HES-SO) from all disciplines who wish to discover how AI can be used to enhance their academic research and skill.

Responsibles and speakers

Speakers

Maryam Asadi

Dates and locations

Period	Location
19.11.2026 from 13:15 to 17:00	B205