Skip to content

mtr7x/AI-Crash-Course

 
 

Repository files navigation

AI Crash Course, Decoded

Understand the 8 papers that define modern AI.

Most AI courses overwhelm you with everything. This one focuses on what matters: the papers that practitioners actually reference, explained from multiple perspectives so concepts click.


Start Here

New to AI papers? Start with the plain English guide — no jargon, just intuition.

Ready to dive in? These 8 papers are the foundation. Everything else builds on them:

# Paper What You'll Understand Analysis
1 Transformers The architecture behind every modern AI Read →
2 GPT-3 Why scale changes everything Read →
3 RLHF How ChatGPT learned to be helpful Read →
4 Chain of Thought Teaching models to reason step-by-step Read →
5 ReAct Models that think and act Read →
6 LoRA Fine-tuning without the cost Read →
7 Llama 3 Inside a frontier model Read →
8 DeepSeek R1 Pure RL for reasoning Read →

Want the full curriculum? See the complete 2-week learning path with 37 papers.


How the Analyses Work

Each paper is broken down from 5 perspectives:

Perspective What it gives you
Precision Key insights, surprising findings, quotable moments
Karpathy-style First-principles technical explanation
Swyx-style What it means for builders shipping products
Elad Gil-style Strategic and business implications
Pseudocode The core algorithm, readable

Pick the perspective that matches how you learn. Or read all five to fully internalize a paper.


Browse All Papers

View all 37 paper analyses →

Organized by category: Foundations, Reasoning, Agents, Benchmarks, and more.


Quick Links


For contributors: regenerating analyses
pip install -r requirements.txt
python tools/paper_processor.py  # Re-analyze papers with your API key
python web_server.py             # Optional local viewer

Fork of Henry Shi's AI Crash Course. See the original thread.

About

AI Crash Course, decoded. 37 papers × 5 perspectives × plain English explanations.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • HTML 51.1%
  • Python 48.9%