Curious to AI-Fluent

Replace science fiction mental models with accurate ones. Learn how ML, neural networks, LLMs, and data interact - without writing a single line of code.

1. 1

   Module 1BEGINNER5 min

   What Is AI, Really? Stripping Away the Hype

   The first barrier to AI literacy is not technical - it is conceptual. Most people carry a mix of science fiction imagery and tech-marketing claims. Neither is useful. This module replaces those with an accurate, honest picture of what AI systems are.

   Topics, prerequisites, and deliverables

   Builds on None. This is the starting point.

   Depth 2-3 hours reading + reflection

   What AI is (and is not); the difference between narrow AI and general AI; why AI systems fail; the relationship between models, data, and predictions; what "training" means mechanistically.

   • Write a one-page explanation of AI for a family member who has never worked in tech. It should accurately describe what AI can and cannot do.

   Open module
2. 2

   Module 2BEGINNER4 min

   How Machine Learning Differs from Normal Software

   The fundamental difference between ML and traditional software determines almost everything about how AI systems are built, debugged, evaluated, and operated. Without this mental model, every downstream concept is harder to absorb.

   Topics, prerequisites, and deliverables

   Builds on Module 1 (What Is AI, Really?).

   Depth 2-3 hours

   Explicit rules vs. learned patterns; why ML requires labeled examples; what a "model" is; training vs. inference; why ML systems can be correct on average but wrong in individual cases.

   • Describe, in writing, why you would use ML instead of rule-based software for two specific real-world problems. Explain what data you would need.

   Open module
3. 3

   Module 3BEGINNER5 min

   What Neural Networks Are (and Aren't)

   Neural networks are the architecture behind most modern AI systems. The sci-fi framing ("artificial brains") is misleading and causes people to over-anthropomorphize AI behavior. The accurate model - a mathematical function approximator - leads to better intuitions.

   Topics, prerequisites, and deliverables

   Builds on Modules 1 and 2.

   Depth 3-4 hours

   What a neuron does mathematically; layers as transformations; why depth helps; what training adjusts; why neural networks are not like biological brains; failure modes unique to neural networks.

   • Explain to a colleague why a neural network that achieved 99% accuracy in testing could still fail badly on a specific type of input. Root your explanation in how networks learn.

   Open module
4. 4

   Module 4BEGINNER8 min

   Large Language Models: Under the Hood, in Plain English

   LLMs are the AI systems most people interact with daily. The "autocomplete on steroids" mental model is accurate enough to reason about failure modes, calibration issues, and appropriate use cases - without requiring any math.

   Topics, prerequisites, and deliverables

   Builds on Modules 1-3.

   Depth 3-4 hours

   Tokenization; next-token prediction; why LLMs hallucinate; temperature and determinism; context windows; why longer prompts cost more; what fine-tuning does and does not do.

   • Find three examples of ChatGPT, Claude, or Gemini producing an incorrect or misleading output. For each, explain which property of LLMs caused the failure.

   Open module
5. 5

   Module 5BEGINNER5 min

   What Data Actually Does in an AI System

   Every AI system is a function of its training data. Understanding this is the prerequisite for reasoning about AI quality, bias, generalization, and failure - for any AI system you encounter professionally.

   Topics, prerequisites, and deliverables

   Builds on Modules 1-4.

   Depth 2-3 hours

   What training data is; why data quality determines model quality; bias and representation gaps in training data; why models fail on out-of-distribution inputs; what "ground truth" means and why labeling is hard.

   • Find a publicly documented case where an AI system behaved poorly because of training data issues. Write a two-paragraph analysis: what was the data problem, and what was the consequence.

   Open module

Learn how to prompt, check outputs, use coding assistants, and spot when an AI tool is wrong.

1. 6

   Module 6BEGINNER5 min

   Prompt Engineering That Actually Works

   Using AI tools effectively is a skill with a learning curve. Most people plateau at basic prompting. This module gives you the structured approach that produces consistently better results.

   Topics, prerequisites, and deliverables

   Builds on Phase 1 (How AI Actually Works). You should understand why LLMs produce the outputs they do before trying to control them.

   Depth 3-4 hours + project time

   System prompts and task definition; few-shot examples; chain-of-thought for complex tasks; specifying output format; iterative refinement; when prompting is the wrong tool.

   • Write a prompt chain for a real task in your work. Test it on 10 different inputs, document where it succeeds and fails, and propose one improvement based on the failures.

   Open module
2. 7

   Module 7BEGINNER5 min

   When to Trust an AI Output (and When Not To)

   Uncritical use of AI outputs is the most common way AI tools cause harm in professional settings. This module gives you a reliable framework for deciding when to trust, verify, or reject.

   Topics, prerequisites, and deliverables

   Builds on Phase 1 and Module 6 (Prompt Engineering).

   Depth 2-3 hours

   Types of AI errors (hallucination vs. reasoning errors vs. outdated knowledge); when AI confidence does not correlate with accuracy; domain-specific risks; verification workflows; when not to use AI.

   • Take an AI-generated output from your recent work and systematically evaluate it: what claims require verification, what sources would you check, and what is the risk of acting on it without verification?

   Open module
3. 8

   Module 8BEGINNER8 min

   Working With AI Coding Agents Effectively

   AI coding tools became agents: they take a task, edit files, run commands, and report back. The habits from the autocomplete era (read every suggestion) do not transfer. This module teaches delegation, review, and permission boundaries so non-programmers can get real work from a coding agent without losing control.

   Topics, prerequisites, and deliverables

   Builds on Module 7 (When to Trust AI Output). Some familiarity with reading code is helpful but not required.

   Depth 3-4 hours

   The agent loop (gather context, act, verify); writing a task with a definition of done; CLAUDE.md and AGENTS.md context files; the four-step review routine; permission settings and irreversible actions; how agents fail (overconfidence, over-reach, faked verification, loops); building small tools without writing code.

   • Delegate one real repetitive task to a coding agent: write the task spec and a CLAUDE.md, run it with restrictive permissions, review with the four-step routine, and document what it got right, what you corrected, and how you would specify it differently.

   Open module
4. 9

   Module 9BEGINNER5 min

   AI Agents: What They Are, What They Can Do, and How They Go Wrong

   Agents act, and most advice about using AI safely was written for chatbots. Non-technical people now meet agents that read their email and act on their behalf; they need a plain-English model of what an agent is, why it can be tricked by what it reads, and the one question to ask before delegating.

   Topics, prerequisites, and deliverables

   Builds on Modules 4 (LLMs Under the Hood) and 8 (Working With AI Coding Agents).

   Depth 2-3 hours

   What an agent is (model + tools + loop) and why the wrapper matters more than the model; how agents connect to tools (MCP as the common plug); what agents are good at; how they go wrong (acting on bad information, prompt injection, over-reach, loops, bad memory); the one question: what can it do without asking; what good agent products look like.

   • Audit one agent product you can access: what it can reach, what it can do without asking, whether it logs, whether you can see its memory. Run one low-risk and one side-effect task, record where it asked for approval, and write down what you would and would not delegate.

   Open module
5. 10

   Module 10BEGINNER4 min

   AI for Research, Writing, and Analysis

   Research, writing, and analysis are the tasks where AI tools provide the largest leverage for non-technical professionals. This module covers the specific workflows that work and the risks to manage.

   Topics, prerequisites, and deliverables

   Builds on Module 7 (When to Trust AI Output).

   Depth 3-4 hours

   AI-assisted literature synthesis; first-draft generation and editing; data summarization; structured analysis prompting; citation and factual accuracy risks; what AI cannot replace in knowledge work.

   • Use AI tools to assist with a real research or writing task in your work. Document your workflow, identify two points where AI could have misled you if you had not verified, and assess the quality of the final output.

   Open module
6. 11

   Module 11BEGINNER5 min

   Evaluating AI Tool Output: A Practical Framework

   Systematic evaluation is what separates experts from casual users of any tool. This module gives you a repeatable framework for assessing AI output quality across different task types.

   Topics, prerequisites, and deliverables

   Builds on Modules 6-10.

   Depth 4-5 hours

   Evaluation dimensions (accuracy, completeness, tone, consistency); building a test set for a specific task; automated vs. human evaluation; evaluating for specific failure modes; reporting evaluation results.

   • Define an evaluation framework for an AI tool you use regularly. Run it on 20 samples, produce a summary of findings, and make a recommendation about when to use vs. not use the tool.

   Open module

Go from zero to being able to load, explore, and describe a dataset in Python. No prior programming experience assumed.

1. 12

   Module 12BEGINNER4 min

   Python for Non-Programmers: The Absolute Minimum

   Python is the universal language of data work and AI. You do not need to become a programmer - you need to be able to read, run, and lightly modify Python code. That is achievable in a week.

   Topics, prerequisites, and deliverables

   Builds on No programming knowledge required. Basic comfort with computers is sufficient.

   Depth 4-6 hours

   Python syntax basics: variables, strings, numbers, lists, dictionaries, functions; running a script; installing packages with pip; reading error messages; knowing when to ask for help vs. googling.

   • Write a Python script that loads a CSV file, prints the number of rows and columns, and prints the top 5 rows. You may use AI assistance, but you must understand every line.

   Open module
2. 13

   Module 13BEGINNER5 min

   What Is Data? Rows, Columns, Types, and Missing Values

   Before you can work with data, you need a precise vocabulary for it. Most data literacy failures are vocabulary failures - misusing terms like "feature," "label," "schema," or "grain" leads to miscommunication and incorrect analysis.

   Topics, prerequisites, and deliverables

   Builds on None. This module is conceptual and accessible to anyone.

   Depth 2-3 hours

   Rows vs. columns; data types (numeric, categorical, datetime, boolean, text); grain (what a row represents); schemas; missing values and why they matter; wide vs. long format; tidy data principles.

   • Download any public dataset and answer in writing: what does each row represent? what are the data types of each column? which columns have missing values, and what might cause them?

   Open module
3. 14

   Module 14BEGINNER4 min

   Reading and Exploring a Dataset in Python

   Reading, exploring, and summarizing a dataset is the first step in almost every data project. This module takes you through the complete process on a real dataset, step by step.

   Topics, prerequisites, and deliverables

   Builds on Module 12 (Python for Non-Programmers) and Module 13 (What Is Data?).

   Depth 4-6 hours

   Loading CSV and JSON data with Pandas; df.head(), df.info(), df.describe(); counting missing values; filtering rows; sorting; groupby aggregations; saving results.

   • Load a public dataset of your choice, produce a summary that answers five specific questions you define, and document any data quality issues you discovered.

   Open module
4. 15

   Module 15BEGINNER4 min

   What Makes Data Useful for ML?

   Not all data trains good models. Understanding what makes data useful for ML enables you to evaluate AI project proposals, identify risks early, and communicate data requirements to ML teams.

   Topics, prerequisites, and deliverables

   Builds on Modules 1-5 (Phase 1) and Module 13.

   Depth 3-4 hours

   What makes a good label; class imbalance and its effects; the importance of data volume; why diversity of examples matters; historical bias in data; what "representative" means; when data is not the bottleneck.

   • Find a public dataset and assess it for ML usefulness: is the label clear? is there enough data? are there obvious bias risks? write a one-page assessment with a recommendation.

   Open module
5. 16

   Module 16BEGINNER5 minYou are here

   Visualization: Turning Numbers Into Decisions

   Data visualization is the bridge between numbers and decisions. A chart that reveals the right pattern at the right moment is more valuable than any model. This module covers the charts that matter and how to choose among them.

   Topics, prerequisites, and deliverables

   Builds on Module 14 (Reading and Exploring a Dataset in Python).

   Depth 3-4 hours

   Histograms and distributions; bar charts for categorical comparison; scatter plots for relationships; line charts for time series; heatmaps for correlation; how to choose the right chart; what makes a visualization misleading.

   • Using a dataset you have explored, produce three visualizations that each answer a different question. For each, write one sentence explaining the insight and one sentence explaining the decision it would support.

   Resume

Build real AI-powered tools using APIs and no-code or low-code platforms - without needing to train any models or write large amounts of code.

1. 17

   Module 17INTERMEDIATE5 min

   Building Your First AI Application With APIs

   Building something real is the fastest way to go from understanding AI to having leverage with it. LLM APIs make it possible to build working AI tools without training any models or writing traditional software.

   Topics, prerequisites, and deliverables

   Builds on Phase 2 (Using AI Tools With Judgment) and Module 12 (Python for Non-Programmers). Basic Python comfort is needed.

   Depth 6-8 hours

   What an API is; getting API credentials; making your first LLM API call in Python; passing context and instructions; parsing responses; basic error handling; building a simple interactive tool.

   • Build a working AI tool that solves a real problem: a document summarizer, a FAQ responder, a tone checker, or similar. It must accept user input and produce useful output. Deploy it locally or on a free hosting platform.

   Open module
2. 18

   Module 18BEGINNER3 min

   RAG Without the Code: What It Is and How to Use It

   RAG is the pattern behind most enterprise AI tools: knowledge bases, customer support bots, document assistants. Understanding it well enough to use, configure, and evaluate RAG tools is practically valuable without building one from scratch.

   Topics, prerequisites, and deliverables

   Builds on Modules 4 (LLMs Under the Hood) and 17 (Building First AI App).

   Depth 4-5 hours

   What retrieval-augmented generation is; why RAG reduces hallucinations; what a vector database does conceptually; how chunking affects results; no-code RAG tools (Notion AI, Glean, Guru, etc.); when RAG is the right approach.

   • Use a no-code or low-code RAG tool to build a knowledge assistant over 10-20 documents of your choice. Test it on 10 queries, identify 3 failure modes, and document one thing you would change.

   Open module
3. 19

   Module 19INTERMEDIATE4 min

   Automation With AI: Connecting Tools and Workflows

   AI tools become dramatically more powerful when connected to each other and to existing workflows. This module covers the automation patterns that allow non-engineers to build multi-step AI pipelines.

   Topics, prerequisites, and deliverables

   Builds on Module 17 (Building First AI App).

   Depth 5-7 hours

   AI workflow platforms (Zapier, Make, n8n); LLM API chaining; trigger-based automation; structured output as workflow glue; where automation fails; building vs. buying automation.

   • Build an automated workflow that uses an AI tool as one step in a multi-step process. Document: what triggers it, what it does, what it produces, and where it could fail.

   Open module
4. 20

   Module 20INTERMEDIATE3 min

   Evaluating What You Built: Does It Actually Work?

   Most people who build AI tools do not evaluate them. Evaluation is what separates a prototype from something worth using. This module covers the lightweight evaluation discipline that every builder should practice.

   Topics, prerequisites, and deliverables

   Builds on Module 17 (Building First AI App) or Module 18 (RAG Without the Code).

   Depth 4-5 hours

   Defining success criteria before building; building a test set for your specific use case; measuring against the criteria; classifying failure modes; deciding when to iterate vs. ship; communicating results.

   • Evaluate the tool you built in Module 16, 17, or 18. Define success criteria, build a 20-case test set, measure performance, classify failures, and write a one-page evaluation report with a recommendation.

   Open module

Connect AI understanding to your career. Learn to work effectively with ML/AI teams, read technical documents, contribute to product decisions, and understand career paths that involve AI.

1. 21

   Module 21BEGINNER3 min

   How to Work With ML Engineers and Data Scientists

   Non-technical professionals who understand how ML engineers think and what they need are dramatically more effective collaborators. Miscommunication between technical and non-technical teams is one of the biggest sources of wasted effort in AI projects.

   Topics, prerequisites, and deliverables

   Builds on Phase 1 (How AI Actually Works) and Phase 2 (Using AI Tools With Judgment).

   Depth 3-4 hours

   What ML engineers and data scientists actually do day-to-day; what they need from product and business teams; common sources of miscommunication; how to write a useful AI feature spec; reading a technical proposal without needing to understand all the math.

   • Write a feature specification for an AI feature in your domain: what problem it solves, what data it needs, what success looks like, and what the constraints are. Have a technical colleague review it for clarity.

   Open module
2. 22

   Module 22INTERMEDIATE3 min

   Reading an ML System Design Document

   ML system design documents describe how AI systems are built, why choices were made, and what the tradeoffs are. Being able to read and engage with them meaningfully - without an engineering background - makes you a more informed decision maker.

   Topics, prerequisites, and deliverables

   Builds on Phase 1 and Module 21.

   Depth 4-5 hours

   Structure of an ML system design document; the data → features → model → serving → monitoring lifecycle; how to identify the key decisions and tradeoffs; questions to ask about evaluation and failure modes; what to look for in vendor proposals.

   • Find a public ML system design case study or paper (Google has published many). Read it and produce a one-page summary: what problem it solves, what key decisions were made, what tradeoffs were accepted, and one question you would ask the team.

   Open module
3. 23

   Module 23INTERMEDIATE3 min

   Product Decisions About AI: Quality, Cost, Latency, Risk

   Every AI feature involves tradeoffs that product and business teams must make alongside engineers. This module gives you a framework for reasoning through and communicating those decisions.

   Topics, prerequisites, and deliverables

   Builds on Phase 1, Module 21, and Module 22.

   Depth 3-4 hours

   The quality-cost-latency triangle; what "good enough" means for a specific use case; risk and failure mode assessment; build vs. buy vs. integrate; communicating tradeoffs to non-technical stakeholders; when not to build an AI feature.

   • For an AI feature you know well or have worked on, document the key product decisions made: what was the quality requirement, what was the latency constraint, what was the cost budget, and what was the risk accepted. Assess whether the decisions were correct.

   Open module
4. 24

   Module 24BEGINNER6 min

   Career Paths Into AI (Technical and Non-Technical)

   AI careers are not just for engineers. A wide range of roles - product, operations, policy, legal, research, and more - involve AI with different skill requirements. Knowing the landscape lets you position yourself for the right next step.

   Topics, prerequisites, and deliverables

   Builds on None. This module stands alone.

   Depth 3-4 hours

   ML Engineer vs. Applied AI Engineer vs. AI Product Manager vs. AI Researcher vs. Trust and Safety vs. AI Policy vs. Data Analyst; what each role requires; typical progression paths; how to evaluate a job description for fit; what signals seniority in each track.

   • Write a career plan: which AI role is your target, what does it require that you currently have, what does it require that you need to build, and what is your concrete 6-month plan to close the gap.

   Open module
5. 25

   Module 25CapstoneINTERMEDIATE4 min

   Capstone: Build, Document, and Present an AI-Powered Project

   The capstone is the evidence layer. It takes everything learned in this path and produces a single artifact that demonstrates AI fluency at the level of an informed practitioner.

   Topics, prerequisites, and deliverables

   Builds on All previous modules. At minimum: Phase 1, Phase 2, and one of Modules 17-20.

   Depth 15-20 hours over 2-3 weeks

   Project selection and scoping; building the project; evaluating it systematically; documenting for a non-technical audience; presenting it with appropriate confidence and appropriate honesty about its limits.

   • Complete a capstone project (see path recommendations). Produce: a working demo or documented output, a written evaluation with success criteria and failure modes, a non-technical summary of what you built and learned, and a 5-minute presentation you can give to any audience.

   Open module