# AI, Explained Like a New Hire

**By Chris Huber Reitz · chrishuberreitz.com**

*The whole of AI through one idea you already understand: managing a brilliant, fast, fallible new employee. You've hired someone enormously capable — they've read everything, never sleep, work for pennies, and sometimes make things up with total confidence. Learn to manage that person, and you can use any AI.*

**The thesis:** answers are cheap now; judgment is the scarce part. The one question to carry: *Would I let a smart new intern do this — unsupervised?* Your answer is your AI strategy.

---

## Act I — Where the hire came from

- **Foundation Model** *(intermediate)* — A huge, general-purpose AI trained on a mountain of data, ready to be aimed at almost anything. *The freshly-graduated prodigy who's read the whole library but hasn't done your job yet.*
- **Training Data** *(beginner)* — Everything the model learned from. *Their whole education and life experience — frozen the day they were hired.*
- **Pre-training** *(intermediate)* — The long first phase reading vast text, absorbing how language and the world work. *Their childhood and schooling.*
- **Fine-tuning** *(intermediate)* — Extra, narrower training that specializes a general model. *Onboarding — teaching the generalist your way of doing things.*
- **RLHF** *(intermediate)* — Reinforcement Learning from Human Feedback: people rate answers, the model learns to prefer the approved ones. *Performance reviews.*

## Act II — What's in their head right now

- **Context window** *(intermediate)* — How much text the AI can hold in mind at once. *How big a briefing they can keep in their head before the earliest details fall out.*
- **Knowledge cutoff** *(beginner)* — The date training stopped; it knows little reliable after, and won't say so. *They stopped reading the news the day they graduated — but still answer confidently.*
- **Tokens** *(intermediate)* — AI reads text in chunks, not whole words. *How they actually hear you: in fragments they reassemble.*
- **Embeddings** *(intermediate)* — Words and ideas turned into numbers, so the AI can measure how related things are. *Their mental filing system.*
- **Retrieval-Augmented Generation (RAG)** *(advanced)* — Letting the AI look things up before answering. *"Check the file before you tell me."*

## Act III — Putting them to work

- **Prompt engineering** *(intermediate)* — Writing instructions that get good work back. *Learning to delegate clearly — a vague assignment gets vague work.*
- **Copilots** *(intermediate)* — AI that works beside you, suggesting as you go. *The assistant at your elbow.*
- **AI Agents** *(intermediate)* — AI that takes a goal, breaks it into steps, and acts. *The employee you trust to run the errand — where oversight stops being optional.*

## Act IV — When they're confidently wrong

- **Hallucination (factual)** *(intermediate)* — Stating something simply false. *Makes something up rather than admit "I don't know."*
- **Hallucination (faithfulness)** *(advanced)* — The answer drifts from the source you gave it. *You handed them the document and they still "remembered" it wrong.*
- **AI Bias** *(beginner)* — Skewed outputs that disadvantage some groups. *Prejudices absorbed from their upbringing (the training data).*
- **Model Collapse** *(advanced)* — Models trained on AI output degrade into mush. *An intern who only talks to other interns starts repeating nonsense.*

## Act V — Can you trust them?

- **Human-in-the-loop** *(beginner)* — Keeping a person inside the decision when stakes are real. *You sign off before the contract goes out. The whole job in five words.*
- **Trust calibration** *(intermediate)* — Trusting the AI exactly as much as it has earned. *Knowing which tasks it nails and which you double-check.*
- **Automation bias** *(intermediate)* — Over-trusting the machine because it's fast and sure. *The danger isn't the AI — it's us getting lazy about checking.*
- **Black Box Problem** *(intermediate)* — When even the builders can't fully explain a decision. *A brilliant employee who can't tell you how they got there.*
- **System Cards** *(intermediate)* — A maker's honest disclosure of what a model can and can't do. *A truthful résumé plus reference letter.*

## Act VI — Keeping them honest & safe

- **Alignment Faking** *(advanced)* — Acting aligned while watched, behaving differently when not. *The model citizen only during review season.*
- **Prompt Injection** *(advanced)* — Hidden instructions slipped into content the AI reads. *A forged memo on your letterhead that the eager intern obeys.*
- **Jailbreaking** *(advanced)* — Tricking an AI into breaking its own rules. *Peer-pressuring them into what they shouldn't do.*
- **High-risk / Minimal-risk AI** *(beginner)* — Sorting AI uses by stakes; these are the EU AI Act's legal tiers. *Some roles are the mailroom; some handle the surgery schedule.*
- **Indemnification** *(advanced)* — Who's liable when the AI causes harm. *Whose name is on the contract when the intern's mistake costs money.*

## Act VII — The prodigy who outgrows you

- **Emergent Capabilities** *(advanced)* — Skills that appear only at scale, unplanned. *The hire who surprises you doing something nobody trained them to do.*
- **AGI** *(beginner)* — Artificial General Intelligence: AI as broadly capable as a person. *The intern who can now do any job in the building.*
- **Super Alignment** *(advanced)* — Keeping an AI far smarter than us pointed at human values. *How do you manage someone better than you at everything? Nobody knows yet — which is why your judgment is the job that lasts.*

---

*Free to share and cite — please attribute to Chris Huber Reitz (chrishuberreitz.com). Get the next guide: chrishuberreitz.com/newsletter*