Chapter 11 · 5 min read
The Cheat Sheet
One page to reread in the car before any meeting.
The whole book on one page. Two minutes in the car, and you walk in as the calmest, most accurate person in the room.
20 concepts, one line each
How it works
- Machine learning: Software that learns patterns from examples instead of following hand-written rules.
- Generative AI: Machine learning that creates new content (text, images, audio, video, code) rather than just scoring or sorting.
- Large language model (LLM): A model trained on vast amounts of text to predict the next token. The engine inside ChatGPT, Claude and Gemini.
- Token: The chunk of text a model reads, writes and bills by; roughly three-quarters of an English word.
- Training: The huge up-front build. Pre-training on oceans of text, then post-training to turn it into a helpful assistant.
- Inference: Running the model on a request. It costs compute every single time, so AI has a real marginal cost.
- Context window: How much the model can hold in view at once. What's outside it, the model can't see.
- Hallucination: Fluent, confident and wrong. A by-product of prediction, reduced but never fully eliminated.
- Reasoning model: Spends extra compute "thinking" before it answers. Better on hard problems, slower and pricier.
- Multimodal: Takes in and produces images, audio and video, not just text.
Making it useful
- RAG: Short for retrieval-augmented generation. The model looks up your documents first, then answers from them. The standard way to make AI know your stuff.
- Fine-tuning: Extra training on your examples to shape style or specialist behavior. Rarely the first fix for missing knowledge.
- Agent: AI that pursues a goal by planning, using tools, checking results and repeating. It does, not just answers.
- MCP: The Model Context Protocol, a common plug that connects AI to tools and data. Think USB for AI.
- Evals: Structured tests of whether a system does your task well enough. The line between a demo and a deployment.
- Prompt injection: Hidden instructions in content an AI reads that trick it into misbehaving. The top security risk for agents.
The business and the big picture
- The stack: Energy → chips → cloud → models → tools → apps. Name the layer before you react to a headline.
- Compute: The processing power that trains and runs models, mostly Nvidia GPUs, often rented from hyperscalers like AWS, Azure and Google Cloud. The scarce input.
- Open-weight vs. closed: Models you can download and run yourself (such as DeepSeek or Alibaba's Qwen) vs. models you rent by the token (GPT, Claude, Gemini).
- AGI: AI that matches people across most thinking work. No agreed definition or date, so ask "whose definition?"
Sound smart: 10 lines that land
| Say this | Use it when |
|---|---|
| "Which layer of the stack is this, and who's paying whom?" | Any AI headline or partnership announcement. |
| "Is our edge the model or the data? Models get copied; proprietary data is the real moat." | Strategy and build-vs-buy discussions. |
| "What does it cost per task at our volume, not in the demo?" | Vendor pitches and budget reviews. |
| "What's the eval? How often is it right on our real cases, and what happens when it's wrong?" | Before any pilot goes live. |
| "Is it answering from our documents, or from memory?" | Someone trusts an AI answer about your business. |
| "Where's the human in the loop, and what can the agent do without asking?" | Any agent or automation proposal. |
| "Let's prove it on one painful, measurable workflow first." | Someone wants to buy an "AI platform." |
| "Speed is part of the product. What's the latency on our busiest day, not a quiet Tuesday?" | Anything real-time or customer-facing. |
| "Prices for this capability fall every year. Let's not lock in today's rate for three years." | Contract negotiations. |
| "The capability is real. The timeline is what's debated." | AGI and jobs conversations. |
Avoid saying: 10 traps and the better line
Each one costs you credibility: it sounds naive, overconfident or like AI-washing. Swap in the better line.
| Instead of… | Say… |
|---|---|
| "The AI knows our business." | "We've connected it to our documents, so it can look things up." It only knows what it was trained on or given. |
| "We should train our own model." | "Let's ground an existing model in our data first, and fine-tune only if that falls short." Training from scratch costs a fortune. |
| "It's 95% accurate." | "On [number] of our real cases, it was right [X]% of the time, and here's what happens with the misses." Accuracy means nothing without the test and the cost of errors. |
| "AI will replace [role]." | "AI will take over specific tasks in that role, and the job shifts toward judgment and relationships." |
| "AGI is two years away." / "AGI is pure hype." | "Serious people disagree on timing. What matters for us is what today's systems can do and how fast they're improving." |
| "ChatGPT told me…" | "I used AI to draft this and checked the key facts against the source." |
| "It's just autocomplete." | "It's trained to predict the next word, and at massive scale that produced surprisingly general skills." |
| "They'll fix hallucinations soon." | "Hallucinations are being reduced, not eliminated, so we build in verification." |
| "It's open source, so it's free." | "Open-weight saves license fees, not the cost of compute, hosting, security and people." |
| "Our product is AI-powered." | "It does [X] for the customer, using a model to [Y], and we measure it by [Z]." Say what it does, or it sounds like marketing. |
The 5 questions for any announcement or pitch
- Which layer is this? Energy, chips, cloud, models, tools or apps. The layer tells you who makes money and who gets squeezed.
- Real or demo? Is it shipping, to whom, at what price? In a pitch: "Show me on our data."
- Who checked? Independent results, or the company grading its own test? In a pitch: "Which customers can I call?"
- Who pays whom? Contract, investment or cloud credits? In a pitch: "What does it cost per task at our volume?"
- What changes for us? What would we do differently in the next 12 months? If nothing yet, file it and move on.
5 numbers worth knowing
| Number | What it means |
|---|---|
| 2017 → 2022 | The transformer paper (Google, 2017) made modern AI possible; ChatGPT launched in November 2022. The whole ChatGPT era fits inside a few years. |
| 1 billion+ | Active users of OpenAI's products, per the company in July 2026. Generative AI has spread faster than the PC or the internet did (Stanford AI Index, 2026). |
| ~13x a year | How fast the cost of a given level of AI performance has fallen since 2023 (Epoch AI estimate, September 2026). Uneven by task, but relentless. |
| ~$700 billion | Planned 2026 capital spending by Amazon, Microsoft, Alphabet and Meta combined, mostly on AI data centers (company guidance, some raised mid-year). |
| 4–7 months | How often the length of task AI agents can finish doubles (METR; mostly software tasks, at about 50% success). |
Date every number when you use it: "as of fall 2026." It keeps you accurate, and it signals that you know the numbers move.