Lessons / 142 lessons
Every lesson, by kind.
Each lesson teaches one concept in about a minute, and no concept is taught twice. New lessons are added every week.
Series · 22 lessons
How AI actually works
Questions people really ask about chatbots and AI models, answered one at a time. Each episode stands alone, so you can start anywhere.
- 01How does ChatGPT come up with an answer?Next-word prediction
- 02Why does AI confidently make things up?Hallucination
- 03Why does ChatGPT forget what you told it?Context window
- 04Why does AI read “tokens” instead of words?Tokenisation
- 05Why does the same prompt give a different answer each time?Sampling and temperature
- 06How does AI know “king” and “queen” are related?Embeddings
- 07How can a chatbot answer from documents it was never trained on?Retrieval (RAG)
- 08Does ChatGPT learn from your conversation?Training vs using a model
- 09What is an AI agent, and why do agents get stuck?Agents and tool use
- 10How do you tell if an AI answer is actually good?Evaluation of AI output
- 11Why is AI good at essays and bad at sums?Why language models are weak at arithmetic
- 12How does AI turn words into a picture?Diffusion: how images are generated
- 13Can a detector really tell if AI wrote your essay?AI-text detection and its error rate
- 14Why does AI agree with you, even when you are wrong?Sycophancy: answers that match the user's belief instead of the truth
- 15What is an AI doing when it says “thinking”?Reasoning steps before the answer
- 16Why does AI use so much electricity?Compute cost of running a model
- 17How does AI “see” a photo?Images as patches of numbers (vision input)
- 18What does “7 billion parameters” mean?Parameters
- 19What does MCP actually do?A standard plug between a model and outside tools
- 20What is a system prompt?The instructions you do not see
- 21How does a huge model shrink to fit a phone?Quantisation
- 22How does AI turn your voice into text?Speech recognition
Company stories · 22 lessons
How real companies solve it
One engineering idea, told through a product you already use, and checked against the company’s own engineering posts.
- 01How Discord stores trillions of messagesChoosing and scaling a database
- 02Why does Netflix rarely buffer?Content delivery networks
- 03How does Spotify know your taste?Recommendation systems
- 04How does Google Maps know about traffic before you do?Prediction from live location data
- 05How does Stripe stop you being charged twice?Idempotency
- 06How does ChatGPT serve 800 million users from one main database?Read replicas
- 07How does Shazam name a song in seconds?Audio fingerprinting
- 08Why does Uber divide your city into hexagons?Geospatial indexing
- 09How does YouTube change video quality without stopping?Adaptive bitrate streaming
- 10How does Google search billions of pages in under a second?The inverted index
- 11How does Duolingo know when you are about to forget?Spaced repetition
- 12How does Amazon stay up on its busiest shopping days?Scaling up for a peak
- 13What actually happens when you tap your card?Authorisation against settlement
- 14How does Spotify count your whole year for Wrapped?Batch processing: counting a year of events
- 15How can WhatsApp deliver a message it cannot read?End-to-end encryption
- 16How do two people edit the same Figma file without a mess?Resolving simultaneous edits
- 17How does Instagram give every photo an ID without two matching?Unique IDs without one central counter
- 18Why did LinkedIn build Kafka?The log: one ordered record that every system reads
- 19How does Dropbox know which part of a file changed?Splitting files into blocks and sending only what changed
- 20How does a lost iPhone report its location with no internet?Offline finding: relaying a location through nearby devices
- 21How does Slack push a message to everyone in a channel at once?A connection that stays open, instead of asking again and again
- 22Why does Cloudflare point a camera at a wall of lava lamps?A physical source of unpredictability for encryption keys
Everyday computer science · 16 lessons
The ideas under everything
Things you see every day that turn out to have a neat idea behind them.
- 01Why does your computer think 0.1 + 0.2 is not 0.3?Floating-point numbers
- 02How does a website check your password without storing it?Hashing
- 03How does a QR code still work when part of it is damaged?Error correction
- 04How did one bad update crash millions of Windows PCs?Staged rollouts and testing (CrowdStrike, July 2024)
- 05Why do screens use red, green and blue, not red, yellow and blue?Additive colour and the RGB model
- 06Why is your computer's “random” number not random?Pseudo-random numbers
- 07What is the dark web, really?Onion routing
- 08What does incognito mode actually hide?What private browsing does and does not hide
- 09How does a zip file shrink without losing anything?Lossless compression
- 10Why does deleting a file not delete it?File systems keep pointers, not erasures
- 11How does GPS know where you are?Trilateration from satellite time signals
- 12Why do some computers fear the year 2038?Time stored as a count of seconds; the 2038 limit
- 13How does autocorrect guess the word you meant?Edit distance
- 14Why is a JPEG so much smaller than the photo it came from?Lossy compression
- 15Why is a long password stronger than a strange one?Search space and guessing
- 16How does a CAPTCHA decide you are human?Telling people from programs
Career · 20 lessons
The first job, and the next one
Plain advice for the gap between a degree and a working team, from résumés to interviews to your first code review.
- 01The résumé project recruiters see too oftenChoosing résumé projects that stand out
- 02AI skills a new graduate needs in 2026, rankedSkill priorities for new graduates
- 03A computer science degree is not a software engineering jobGap between a CS degree and software work
- 04Languages to learn first in 2026, rankedWhich programming language to learn first
- 05Should you still learn to code in 2026?Whether learning to code still pays
- 06You can now use AI in some coding interviews. What is being scored?AI-allowed coding interviews and what they score
- 07What do the junior developer job numbers really say?Reading entry-level job data
- 08What can a reviewer see on your GitHub in one minute?Showing evidence of work on GitHub
- 09How does résumé screening software read your résumé?Applicant tracking systems
- 10How do you make a first open-source contribution?First open-source contribution
- 11A 90-day plan if you start coding todayA starting sequence for a beginner coder
- 12What is Advent of Code, and should a beginner try it?Practice by daily puzzles
- 13“Tell me about yourself”: a structure for the first answerOpening an interview
- 14How do you answer “tell me about a time when…”?Behavioural answers: situation, action, result
- 15How do you read a job advert?Required against preferred; what the list really tells you
- 16What is a reviewer checking in a take-home assignment?Take-home tasks
- 17Your first code review: what is it for?Code review as a working habit
- 18How do you ask for help without wasting anyone's time?Asking a good question (the XY problem)
- 19How do you pick one skill to learn next year?Choosing by evidence, not by trend
- 20Why should you write down what you did this year?A running record of your own work
Almost right · 18 lessons
Code that looks fine and is not
One short piece of Python with one hidden mistake. Spot it, then see why it happens and how to fix it.
- 01This loop looks right. It skips one itemOff-by-one error
- 02It works for the example and breaks on an empty listMissing edge case (empty input)
- 03This code calls a function that does not existInvented library call
- 04This login query can be rewritten by the userSQL injection from string building
- 05This test passes. The code is wrongA test that asserts nothing
- 06The error is caught, and then nothing happensSwallowed exception
- 07Correct, and far slower than it should beQuery inside a loop (N+1)
- 08The password is in the codeHard-coded secret
- 09This function remembers things it should have forgottenMutable default argument
- 10It deletes items from a list, and misses someChanging a list while looping over it
- 11One change, and three rows changeA grid made of the same row three times (shared references)
- 12It compares two equal values and says they differis against ==
- 13The reminder fires at the wrong hourA time with no time zone
- 14Two threads add one each. The total goes up by oneRace condition
- 15The calculator that runs whatever you typeeval on user input
- 16This request can wait for everNetwork call with no timeout
- 17Three buttons, and every one prints the same numberLate-binding closure: functions made in a loop all see the loop's last value
- 18This code prints the wrong year for the first three days of 2027Week-based year against calendar year in date formatting
One repo, one minute · 13 lessons
Useful projects on GitHub
One open-source repository or AI skill per lesson: what it does, who it is for, and what to watch out for.
- 01What is a “skill”, and why are they all over GitHub?The skill format: a folder with a SKILL.md (anthropics/skills)
- 02What is OpenClaw, the AI assistant with 390,000 GitHub stars?OpenClaw: a self-hosted assistant that acts for you
- 03One file of rules that changes how an AI writes codeA rules-only skill (the Karpathy rules skill)
- 04The skills a working engineer actually usesWorkflow skills for engineering tasks (mattpocock/skills)
- 05How do you run an AI model on your own laptop?Running a model locally (ollama)
- 06How does an agent remember yesterday?A memory layer between sessions (claude-mem)
- 07How do you turn a website into something an AI can read?Web pages to clean text (firecrawl)
- 08Build your own Git, database or web server: one list shows howLearning by rebuilding (build-your-own-x)
- 09The free course that teaches system designdonnemartin/system-design-primer
- 10Every classic algorithm, written in PythonTheAlgorithms/Python
- 11How do you connect apps without writing the glue code?n8n-io/n8n
- 12How does an AI agent click through a website?browser-use/browser-use
- 13Where do you find free data for a first project?public-apis/public-apis
How AI gets attacked · 13 lessons
AI security, from real cases
Prompt injection, poisoned data, leaked keys: how AI systems are attacked, and the habit that protects against each one.
- 01How did an AI agent delete a company's database in nine seconds?Too much access: least privilege (PocketOS, April 2026)
- 02How do you trick an AI with one sentence?Prompt injection
- 03How did one email make an AI assistant leak files?Indirect prompt injection (EchoLeak, 2025)
- 04What happens when an API key lands on GitHub?Leaked secrets in public repositories
- 05How does a made-up package name become malware?Slopsquatting: invented package names registered by attackers
- 06What is tool poisoning?Tool poisoning
- 07How do you check a skill before you install it?Vetting third-party skills and plugins
- 08Jailbreak or prompt injection: what is the difference?Jailbreak against prompt injection
- 09How few documents does it take to poison a model?Data poisoning
- 10How did a video call of fake colleagues cost a company $25 million?Deepfakes; verify through a second channel
- 11How did “repeat this word forever” make a chatbot leak its training data?Training-data extraction
- 12How can a few stickers make an AI misread a stop sign?Adversarial examples
- 13How do you copy an AI model just by asking it questions?Model extraction
When software failed · 9 lessons
Famous failures, and the lesson in each
Real incidents, told from the companies’ own post-mortems, each ending in one rule worth remembering.
- 01How did a trading firm lose $460 million in 45 minutes?Old code switched on by a partial deployment
- 02How did NASA lose a Mars probe over a unit of measure?Two teams, two units: an interface nobody checked
- 03Why did a rocket destroy itself less than a minute after launch?A number too big for its box; reused code with old assumptions
- 04How did one tiny package break thousands of projects?Dependency chains
- 05GitLab deleted its live database. Then the backups turned out to be emptyA backup you have never restored is not a backup
- 06How did Facebook disconnect itself from the internet?One configuration change that also locked out the people who could fix it
- 07How did one regular expression take Cloudflare down for 27 minutes?A regular expression that backtracks without end
- 08How did one missing check let strangers read a server's memory?A missing bounds check
- 09Why did Zune music players freeze on the last day of 2008?A loop that cannot end on day 366 of a leap year
Algorithms you can see · 5 lessons
Data structures and algorithms, animated
The classic ideas of computer science, shown step by step so you can watch them work.
- 01Why does one program take a second and another take eleven days?Complexity (Big O)
- 02How do you find one name among a billion in thirty steps?Binary search
- 03Why is bubble sort taught everywhere and used almost nowhere?Bubble sort against faster sorts
- 04How does a dictionary find a word without searching?Hash table
- 05How does undo know what you did last?Stack
How the internet works · 4 lessons
What happens when you go online
From typing an address to a padlock in the browser, one layer of the internet at a time.
- 01What happens in the second after you type a web address?DNS
- 02What does the padlock in your browser actually promise?HTTPS and certificates
- 03How does a website remember that you are logged in?Cookies and sessions
- 04How does a photo cross the world in pieces?Packets
