Summary

This repository provides a comprehensive tutorial series on Generative Artificial Intelligence, curated from Professor Hung-yi Lee's university lectures. It targets beginners and practitioners seeking to understand cutting-edge AI, offering a structured curriculum that demystifies complex topics, explains core technologies, and highlights practical applications and ethical considerations.

How It Works

The tutorial adopts a modular, lecture-based approach, systematically covering Generative AI's landscape from foundational concepts to advanced applications. Core components include an introduction to GenAI, in-depth exploration of Large Language Model (LLM) training methodologies across three stages (self-learning, fine-tuning with expert guidance, and practical application via RLHF), and the architecture of AI agents. It meticulously explains the pivotal Transformer model and diverse image generation techniques like VAE, Flow, Diffusion, and GANs. This pedagogical structure prioritizes conceptual clarity and the underlying principles driving AI advancements, bridging theory with emerging practices.

Quick Start & Requirements

Access is primarily through downloadable PDF lecture materials, available via GitHub Releases and alternative links. The repository does not offer a direct codebase for execution but provides links to supplementary Deep Learning and Reinforcement Learning tutorials for extended study.

Highlighted Details

• Comprehensive coverage spanning introductory GenAI, multi-stage LLM training, AI agents, Transformer architecture, and image/video generation.
• Detailed explanations of key technologies including prompt engineering, RLHF, diffusion models, GANs, and Transformer mechanics, with specific modules dedicated to their principles.
• Inclusion of contemporary topics such as AI safety, ethical considerations, model evaluation metrics, and advanced techniques like speculative decoding.
• Analysis of emerging models and potential underlying technologies, referencing GPT-4o's voice capabilities and discussing model fusion.

Maintenance & Community

Developed collaboratively by researchers from leading academic institutions, including National Taiwan University, Shanghai Jiao Tong University, and the University of Oxford. A dedicated WeChat group facilitates reader interaction and community support.

Licensing & Compatibility

Licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0). This license permits non-commercial sharing and adaptation, requiring attribution and reciprocal sharing of derivative works. Commercial use is restricted.

Limitations & Caveats

Primarily an educational resource focused on conceptual understanding rather than a deployable software project. Direct code implementations are not the focus. Content is predominantly in Chinese, potentially limiting accessibility for non-Chinese speakers.