/ Stage 1: Foundations
Stage 1 of 7
Before you can use AI effectively, you need to understand what it actually is. This stage builds the mental models and vocabulary that make everything else click. How large language models work, the key players shaping the industry, and the difference between generative, agentic, and physical AI.
Large Language Models like ChatGPT and Claude are sophisticated pattern-recognition systems trained on vast amounts of text, learning to predict what comes next in a sequence. Understanding this fundamental mechanism --- prediction, not reasoning --- is key to using these tools effectively and understanding their limitations.
The most comprehensive and accessible explanation of how LLMs work, from one of the field's pioneers. Covers pretraining, fine-tuning, reinforcement learning, and practical "LLM psychology." Watch at 1.5x if pressed for time.
The condensed version. Covers the core technical components behind ChatGPT, Claude, and similar systems. Part 1 explains how they work; Part 2 covers where they're heading.
Grant Sanderson's gift is making complex math intuitive through animation. This visual walkthrough explains LLMs in a way that genuinely builds understanding, not just familiarity.
Takes you inside the transformer architecture (the "T" in GPT) with beautiful visualizations. You'll understand attention, tokens, and the core mechanics that power modern AI.
A free MIT course covering the history of AI, what makes foundation models different, and practical applications. More academic but extremely thorough.
AI is layers of technology working together. Understanding the stack --- models, infrastructure, applications --- helps you see how different pieces fit together and where value is being created. This mental model is essential for evaluating AI tools and understanding industry dynamics.
The definitive visual map of the AI ecosystem. Over 2,000 companies organized by category --- from foundation models to applications to data infrastructure. Bookmark this; you'll return to it often.
The most comprehensive annual report on AI progress. Covers research breakthroughs, industry applications, geopolitics, and safety. Free, open-access, and updated every October.
Stanford's authoritative tracking of AI trends across technical performance, adoption, investment, and policy. Data-rich and balanced.
A venture capital perspective on where AI infrastructure is maturing and where opportunities remain. Good for understanding the business landscape.
Analysis of which AI products are actually gaining traction with users. Useful reality check on hype vs. adoption.
AI has its own language. Tokens, context windows, fine-tuning, RAG, embeddings --- these terms come up constantly. A working vocabulary makes everything else more accessible. Think of this as learning the menu before ordering.
Beginner-friendly definitions of essential terms like LLMs, context windows, chain-of-thought, and prompting. Clear, jargon-free, and designed for learners.
The most comprehensive glossary available. Covers foundational ML concepts through advanced topics. Actively maintained by Google's ML team. More technical but authoritative.
Business-focused glossary that helps you understand AI terminology in a commercial context. Well-written and accessible.
Innovation-focused terminology resource. Good for business leaders who need to communicate about AI with their teams.
A quick-reference PDF covering core AI terms from the makers of Claude. Handy to save and print.
Not all AI is created equal. "Generative AI" creates content. "Agentic AI" takes actions autonomously. "Physical AI" operates in the real world through robots and sensors. Understanding these categories helps you map capabilities to use cases and separate genuine innovation from marketing buzzwords.
Clear academic explanation of what makes generative AI different from traditional rule-based systems. Foundational distinction.
MIT's explainer on autonomous AI systems that can plan, reason, and take actions. This is where AI is heading --- understand it now.
Balanced VC perspective on what's actually working in agentic AI vs. the hype. Good reality check.
Major analysis of AI moving into the physical world through robotics. Projections, use cases, and implications.
How organizations are transitioning from old AI approaches to new ones. Strategic context for business leaders.
The AI industry is consolidating around a few major players while also fragmenting into specialized applications. OpenAI, Anthropic, Google, Meta, and others are competing fiercely. Understanding who's who --- and their different approaches --- helps you make informed choices about which tools to use and trust.
Head-to-head comparison of the four dominant model families. Covers strengths, weaknesses, and use cases for each. Updated for 2025.
Real-time benchmark comparison showing capabilities, pricing, and context windows across all major models. Bookmark this for reference.
Report card evaluating each major player's 2025 performance. Useful for understanding competitive dynamics.
Analysis of recent breakthroughs from the top three competitive players.
Extended comparison including GPT-5, Claude, Gemini, Perplexity, and emerging models.
For those who want to understand the technical foundations --- not just what AI does, but why it works. Transformer architecture, the attention mechanism, and the intellectual history from neural networks to today's foundation models. More technical but still accessible.
The gold standard for visual explanations of transformer architecture. Referenced in courses at Stanford, Harvard, MIT, and beyond. If you only read one technical explainer, make it this one.
Interactive visualization with a live GPT-2 model running in your browser. Input text and watch attention weights, computations, and predictions in real-time.
The deepest 3Blue1Brown dive into how attention actually computes. Beautiful animations make the math intuitive.
Companion to The Illustrated Transformer, showing how GPT-2 specifically implements these concepts.
Removes the mystery around larger models by showing training and inference visually.
Full course building neural networks from scratch. For those who want to truly understand by building. Requires basic Python.
Beautifully illustrated O'Reilly book with working code. The best resource for practitioners who want to build with LLMs.
Theory is great, but doing is better. These exercises reinforce what you've learned and take 10-30 minutes each.
After watching at least one Karpathy or 3Blue1Brown video, explain how LLMs work to a friend or colleague without using jargon. Can you explain why LLMs sometimes "hallucinate"? If you can teach it, you understand it.
Open the Matt Turck MAD Landscape. Find three companies you've heard of, three you haven't, and identify where in the stack the AI tools you currently use sit. Where is the landscape most crowded? Where are opportunities?
Using the MIT Sloan Glossary, write brief definitions in your own words for five essential terms: Token, Context window, Hallucination, Fine-tuning, and Prompt. Keep these somewhere you can reference them.
Not as magical thinking machines, but as sophisticated prediction systems trained on text. You can explain the basics to others.
Generative AI creates content, agentic AI takes actions, physical AI operates in the real world. You can identify which category a given application falls into.
You know the major players, understand how the technology stack fits together, and can reference authoritative sources for staying current.
Tokens, context windows, fine-tuning, embeddings, hallucinations. These are tools in your toolkit now, not mysterious terms.
Understanding the fundamentals means you can evaluate AI claims more critically. You know what these systems can and can't do.