Wolf Paulus – Embry-Riddle Aeronautical University, October 2025
This repository accompanies the talk “From Vectors to ChatGPT – The Building Blocks of Generative AI”, presented at Embry-Riddle Aeronautical University (ERAU) for first-year engineering students.
The session demonstrates how simple math, data structures, and a few lines of Python form the foundation of systems like ChatGPT.
It is designed to be visual, accessible, and hands-on, helping students connect engineering fundamentals to modern AI.
- How language can be represented as numbers (vectors & embeddings)
- How similarity and meaning emerge from data
- How transformer models build context on the fly
- How generative models produce language one token at a time
- How a few lines of Python code can illustrate all these ideas
“Every time you talk to ChatGPT, you’re not really talking to a robot — you’re talking to math. And this repo shows how that math learned to speak.”
| Folder / File | Description |
|---|---|
AI-2025.pdf |
Full presentation slides |
src/ |
All Python demo scripts |
requirements.txt |
Required Python packages |
img/ and mov/ |
Optional images or banner for presentation / GitHub Pages |
- training.py – train simple word embeddings from scratch
- word2vec.py – train word2vec embeddings and explore
- word2vec_3d.py – visualize word2vec embeddings training in 3D
- similarity.py – find the most similar words in embedding space
- outlier.py – find the word least similar to the group
- analogy.py – analogy reasoning (king – man + woman ≈ queen)
- bert.py – get contextual embeddings with BERT
- sbert.py – sentence embeddings with Sentence-BERT
- q_and_a.py – question answering with sentence transformers
- heroes.py – generative text example with Hugging Face
Clone the repo and install dependencies:
git clone https://github.com/wolfpaulus/VEC-2-GENAI.git
cd VEC-2-GENAI
python -m venv ./.venv
source .venv/bin/activate # On Windows use: .\.venv\Scripts\activate
pip install -r requirements.txtDownload required models for demos: ./get_models.sh
Download GPT model for local use (optional, for faster demos): python ./src/download_gpt.py
Run any demo:
python ./src/outlier.pyEach demo runs standalone and includes inline comments for learning.
This talk introduces engineering students to the fundamental principles behind large language models (LLMs) — without assuming any prior AI background. It begins with vector representations of words, builds intuition for how relationships and context arise, and culminates with generative text demos.
The goal: To make AI feel approachable, logical, and rooted in the same math and data principles every engineer learns.
MIT License — freely available for educational use and adaptation. Please credit the author if you reuse the slides or code examples.