Part 1

The Model Under the Product

Trace the path from training data and token prediction to chat interfaces, hallucinations, and mitigation choices. This chapter gives product decisions a technical reason instead of a trend-driven guess.

19 notesOpen reading · no login
Part 119notes

THE QUESTION THIS PAGE ANSWERS

ANSWER FIRST

What will the “The Model Under the Product” AI learning path help you do?

Trace the path from training data and token prediction to chat interfaces, hallucinations, and mitigation choices. This chapter gives product decisions a technical reason instead of a trend-driven guess. The path contains 19 free notes, each centered on one question you can understand and test.

DECISION RULE

Core themes include Core Principles, From Completion to Chat, Hallucinations & Four Mitigations, Part Recap.

TRY NEXT

Begin with “AI's Food: Training Data,” then choose the next note by the task in front of you.

WATCH FOR

Do not optimize for finishing the list. Explaining one trade-off with your own example matters more than opening more titles.

What this route helps you practice

Open the first note

Each chapter follows a class of real decisions. Follow the sequence, or enter at the problem you are solving today.

19notes
01AI's Food: Training DataWhat does 15T Tokens look like? Corpus composition visualization + data-scale intuition sliderInteractive3 min02Training vs Inference: Two Different ProcessesConversations are not learning; parameters are frozen; billing is per Token — essentials every AI PM must knowConcept3 min03Vocabulary & TrainingFrom corpus to word-pair matrices: Tokenization + attention-weight interactive demoInteractive3 min04Base Model: A Token-Predicting MachineWhat do you get after training? Step-by-step generation with live probability distribution updatesAnimation3 min05The GPT Breakthrough: PreTraining Changes EverythingInteractive comparison of CNN / RNN / BERT / GPT, with memory-decay visualizationInteractive4 min06The chat/completions MysteryIt's clearly a conversation — so why is the API called "completions"? A typewriter animation explainsAnimation3 min07Faking a Chat HistoryOpenAI's original experiment: turning a completion machine into a chatbotAnimation3 min08Chat Template + SFTJinja formatting, instruction fine-tuning — LLMs finally learn to talkConcept3 min09The Context Window Is KeyWhy are prompts powerful enough? Token truncation visualization — no retraining neededInteractive3 min10LLM Hallucination DemoThree classic types: factual errors / confident fabrication / knowledge cutoffCase Study3 min11Mitigation 1: Prompt EngineeringConstraint instructions + limitations: the model doesn't know what it doesn't knowConcept4 min12Mitigation 2: RAG — Retrieval-Augmented GenerationInject real documents into the context; 5-step animated flow; compare with and without RAGInteractive4 min13The Cost & Optimization of RAGCost analysis table + four strategies: keyword trigger / model routing / semantic cache / precise chunkingPM Advanced5 min14Mitigation 3: Temperature & Top-PDrag the slider to see probability distributions and output changes in real timeInteractive3 min15Mitigation 4: Evaluation + Human ReviewExternal correction layer — a cold-start fallback strategy (HITL)Concept3 min16Recap (Part A) · What LLMs Are + HallucinationsTraining essence / Token / Base→SFT→Chat / four hallucination types and root causesRecap4 min17Recap (Part B) · Mitigation Strategies + Decision FrameworkFour mitigation strategies compared / common misconceptions / solution selection matrixRecap3 min18Lock That One Job DownFour-step demo of tightening a requirement; three tasks: write a four-line spec, try it on five real inputs, draw a human–AI boundaryHands-on7 min19LLM Fundamentals · 30 Tough QuestionsEach with intent, framework, and bonus points: probability prediction / message list / hallucination explanation / RAG vs retraining / Temperature / context windowQuiz18 min