An evidence-based, historical journey demistifying how large language models (LLMs) work. We strip away the marketing hype and look into the "black boxes".
Main Website: skepticcto.com | YouTube Channel: @SkepticCTO
Series Overview
"Decoding the Language Machine" will be a 6-part series that traces the history of computer science, from Claude Shannon's 1948 work on the statistics of the English language to modern transformers.
The series and this repository were built during a 4-month sabbatical by Robert "Butch" Buccigrossi, Ph.D.. I’ve spent the last 21+ years as a CTO, earned my Ph.D. in Computer Science from the University of Pennsylvania in 1999 (focused on computer vision and machine learning), and currently serve as a Principal Investigator in the NIST AI Safety Initiative Consortium.
My goal is to apply rigorous scientific skepticism to AI and use historical breakthroughs to build intuition on LLM behavior.
Purpose of this Repository
In this repository we share the foundational resources used to create the series under a Creative Commons Licnese.
- Manim Source Code: Python scripts for the mathematical animations used in the videos.
- Media Assets: Video clips, audio, and images used in the episodes.
- LLM Prompts: The specific prompts used for research, scripting, and asset generation.
- Planning Documents: Research reports, production bibles, and style guides.
We encourage creatives, educators, and developers to explore these materials to learn about AI and to reuse these resources and approaches in their own educational or creative projects.
Repository Structure
- Series Planning/: High-level planning documents, channel strategies, and the series style guide.
- Episode 1/: "Shannon's N-grams: Precursor to LLMs" Video
- Episode 2/: "Symbolic AI and the AI Winter" Video
- Episode 3/: "The Learning Revolution" Video
License
All resources in this repository are released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license by SkepticCTO LLC / Robert Buccigrossi, Ph.D.