Tendril
Loading lessons…
Builders · Ages 11–13
Compare Claude, ChatGPT, Gemini, and Flux side-by-side. Learn prompt engineering and catch hallucinations.
Meet your guide: Wren — a sharp raven
Your progress
Loading your progress…
Where should I start?
Chapters
Modules · 211
AI is a label that covers many things. Let's narrow it down so you can tell marketing hype from the real computer science underneath.
Most modern AI is trained on a loop of guess, check, and adjust. Understand the loop and you understand the heart of machine learning.
AI does not read letters. It reads tokens, which live as vectors in a space of meaning. Learn how text becomes numbers you can do math on.
You have heard the term a thousand times. Now let's actually look inside: neurons, weights, activations, and what happens in a single pass.
You cannot understand modern AI without understanding its diet. Let's map where the data comes from, how it gets cleaned, and what that means.
Every new model claims a new high score. Before you trust a leaderboard, learn what benchmarks actually measure — and what they miss.
The line between deep reasoning and clever pattern recognition is blurry. Here's how researchers try to tell them apart.
The past decade of AI progress came from a simple, ruthless law: more compute and more data, predictable improvements. Here is the math behind it.
As models scale, some skills do not gradually improve — they just snap into existence. Let's look at what emergence really means and why it scares people.
AI did not start in 2022. It has decades of wrong turns and breakthroughs. Knowing the history helps you spot hype from real progress.
Pro prompters follow a structure. Give the AI a role, set the context, show examples, set constraints, and pick a format. This framework alone 10x's your output quality.
Instead of describing what you want, show the AI two or three examples. Few-shot prompting is often the fastest way to get consistent output.
Telling the AI to 'think step by step' before answering dramatically improves its accuracy on reasoning problems. Here's why and when.
Every AI conversation has two layers: a system prompt that sets the rules, and user prompts you type. Understanding the difference is the gateway to building AI-powered tools.
Beginners scrap their prompt and start over. Pros keep the good parts and change only what isn't working. Here's how to iterate like a craftsperson.
When your prompt feeds into code, you need machine-readable output. JSON mode and XML tags make the AI's response parseable instead of loose prose.
Turn your best prompts into reusable templates with variables. This is how pros scale: one great template, thousands of runs.
Bad output is almost never random. It's a clue. Here's how to diagnose and fix a broken prompt instead of just mashing the regenerate button.
AI bias is not magic and not moral failure. It is math operating on imperfect data. Here is exactly where the bias enters the system.
AI is now involved in hiring, loans, medical care, and criminal sentencing. Here are the documented cases and the frameworks being built in response.
Generative AI trained on copyrighted work has triggered the biggest wave of copyright lawsuits in the internet era. Here is the state of the fight.
Using AI on schoolwork is not simply cheating or not cheating. It depends on the task, the rules, and what you are learning to do. Here is how to think about it.
Before AI, lies took time to make. Now they take seconds and come in infinite variations. Here is how the information ecosystem is changing.
Your posts, chats, photos, and behavior have been scraped, sold, and fed to models. Here is what has actually happened and what you can actually do.
Training a frontier model uses the electricity of a small city for months. Running inference at scale matches a large country's load. Here is what the numbers actually look like.
Children are using AI more than any other group, and have less legal protection. Here is what current laws cover, what they miss, and what is being debated.
AI-assisted coding is not magic and not cheating. It is a new way of working where a model drafts, you decide. Let's draw a map before we start building.
Let's actually feel what autocomplete is like. Write a comment, pause, and watch a full function appear. Then learn what to do next.
A prompt that writes a poem is not the same as a prompt that ships working code. Code has hidden standards. You need to make them explicit.
The AI will hand you code that looks right but isn't. Here are the most common bugs and the habits that catch them before they bite.
Bugs are where AI is most useful and most humbling. Paste errors, ask for causes, run experiments, and learn how to get a real answer instead of a guess.
Writing a test first is not just good engineering. It is the clearest possible prompt for an AI. Let's use tests to make AI code reliable.
Most of a developer's life is reading code someone else wrote. AI is astonishing at this. Here's how to get fast, honest explanations of unfamiliar code.
Refactoring means changing code without changing behavior. That used to be scary. With tests and AI together, it becomes routine.
Time to get hands on. Install a real AI coding editor, sign in, and write your first line together. No credit card required to start.
Git is a time machine for your code. Before we ship anything, let's learn the three commands that matter and what they actually do under the hood.
Let's make something real. A single-page site with HTML, CSS, and a little interactivity. You plan, the AI drafts, you review and ship.
Bring it all together. Pick one of three starter projects, plan it, build it with AI, and deploy it. You are now a builder who ships.
A chatbot answers. An agent does. Learn the line between a model that talks and a model that acts — and why crossing it changes everything about how you work with AI.
Every agent — fancy or simple, local or cloud — boils down to four parts. Learn the recipe and you can read any agent system like a menu.
Follow a real agent run step by step — from prompt to result — and see exactly what happens inside. No code yet, just the anatomy of a successful task.
Agents fail in weird, quiet, expensive ways. Learn the six failure modes, the warning signs, and the simple habits that catch problems before they compound.
Giving an AI the keys to your computer is a big deal. Learn the two simplest ways to keep an agent safe: wall it off from things it shouldn't touch, and put a human in the decision path.
Agents are only as useful as their tools. Tour the big three — filesystem, browser, code execution — plus the emerging MCP ecosystem, with examples of what each unlocks.
Your data can live in someone's data center or on your own laptop. Both are real options in 2026. Understand what you gain and lose with each.
OpenClaw is open-source software that runs agents on your own machine — no cloud dependency, your data stays put. A tour of why it exists and how its pieces fit together.
Ollama turns 'I want to run an LLM locally' into a one-line install and a two-word command. Here's the stack, the key commands, and the models worth pulling first.
No code. Just design. Pick a real task you do every week and draft a complete agent spec — goal, tools, loop, stop, approvals, and what success looks like.
An AI that paints starts with pure noise and removes it, one step at a time, until a picture appears. Here's the surprisingly beautiful math behind it.
Five image models, five personalities. Here's when each one is the right pick — in 2026, with current strengths, costs, and quirks.
Great image prompters aren't typing harder — they're using a mental framework. Subject, setting, style, composition, lighting, mood. Here's the system.
Text-to-video became practical in 2025 and cinematic in 2026. Here's the state of the art and how to choose.
ElevenLabs can clone a voice from 30 seconds of audio. That's useful for accessibility — and dangerous in the wrong hands. Here's how to use it well.
Type a prompt, get a full song — vocals, drums, mix, even in Portuguese. Here's how Suno v5, Udio, and ElevenMusic work — and what they can't yet do.
US Copyright Office in 2026: works created purely by AI aren't copyrightable. Works with enough human creative control might be. Here's where the line sits right now.
Your first end-to-end AI-assisted creative project. Plan it, make it, and reflect on what surprised you. Small scope, real output.
Claude.ai and the Anthropic API both run Claude. So why do they cost different amounts? Pull apart the two doors into the same model.
Every big AI has a free version. Stack them side-by-side and learn where each one runs out of gas.
All three claim to be the best. Pick tasks you actually care about, run the same prompt across all three, and you'll build your own benchmark.
When the question is 'what happened this week?' or 'what does this paper say?', Perplexity is often the right answer. Here is why.
Grok is the odd one out — baked into X, trained on live posts. Sometimes that's a superpower, and sometimes it's a liability.
Voice interfaces flipped from gimmick to genuinely useful. Learn what each top voice mode feels like and when to pick which.
AI in your browser turns every webpage into something you can interrogate. Learn which extension to install, and why that access needs trust.
Artifacts is Claude's canvas. Charts, code, docs, and interactive React components render live next to the chat.
Deep Research is Gemini's multi-step research agent. You ask a question; it plans, searches, reads, synthesizes, and delivers a report.
v0 by Vercel turns a prompt, screenshot, or Figma file into a working Next.js app deployed in one click.
Upload a PDF, a set of docs, or a research paper. NotebookLM produces a two-host podcast conversation about the material.
A Space is a bookmarked, collaborative research context. Your sources, your prompts, your team — all persistent.
Opus is the flagship, Sonnet is the workhorse. Here is the five-minute decision tree for when to pay 2x more for Opus and when Sonnet handles it.
Two frontier models, same subscription price, very different personalities. Pick by vibe, not by benchmark — here is how to figure out which one clicks for you.
Everyone brags about million-token windows. Here is what you can actually do with one when you learn how Gemini 2.5 Pro handles long documents.
xAI's Grok 4.1 Fast has the biggest context window on the market at the cheapest price. Here is when that matters more than raw reasoning quality.
When you need sub-second responses at pennies per thousand calls, you are choosing from the mini tier. Here is the honest Haiku vs. mini comparison.
Midjourney is the artist favorite. FLUX.2 Pro is the API-native challenger. Here is which one to pick depending on what you are making.
Both generate full songs from a prompt. Suno wins on ease and ELO. Udio wins on audio fidelity and producer workflows. Here is how to pick.
Runway built for filmmakers. Sora 2 was the tech demo that melted OpenAI's GPU budget. Here is how to pick a video model for actual projects.
Three command-line coding agents, three flavors. Which one belongs in your terminal? Install all three on a weekend and decide for yourself, but here is the cheat sheet.
Posters, logos, ads, memes — any image with legible text is a special case. Ideogram and FLUX.2 both do it well. Here is who wins what. Before using AI-generated marks commercially, do a basic USPTO search (or ask a lawyer) — a Swoosh on a shoe is still a Nike problem regardless of who rendered the pixels.
Every LLM hallucinates. Perplexity's Sonar family solves it by grounding answers in live web results with citations. Here is when to use Sonar instead of Claude or GPT.
ElevenLabs voices are indistinguishable from humans. That is a feature and a fraud vector. Here is the production checklist before you clone anyone.
Haiku is Anthropic's cheap, fast tier. Here is the math on when it beats Sonnet for production workloads.
Extended thinking makes Opus smarter but burns hidden tokens. Here is how to budget it without blowing your bill.
GPT-5.5 is the hard-problem default; GPT-5.4 mini is the cost-sensitive workhorse. Learn when quality is worth the extra latency and tokens.
Reasoning effort trades latency and tokens for better answers on hard problems. Here is when that trade is worth it. In the current GPT-5 family, that choice usually shows up as model selection plus a reasoning effort setting.
Google gives Flash away on a generous free tier. Here is how to extract real production value without paying a cent.
Gemini Ultra on Vertex unlocks extended context and enterprise controls. Here is what you get for moving up-tier.
xAI's code-specialist model ships strong benchmarks. Here is how it actually feels in a real IDE.
Meta's Llama 4 family splits into Scout (lean) and Maverick (flagship). Here is how to choose between them for self-hosted work.
Mistral Large 2 quietly beats the US frontier models on several non-English benchmarks. Here is why it should be your default for European languages.
Codestral 25 is Mistral's dedicated coding model. Small, fast, and cheap enough to run as an inline autocomplete.
Mistral Small is the right open-weights model when you need to run on a laptop, a phone, or an on-prem CPU box.
Codestral Mamba ditches transformers for a state-space model. The result: linear-time long-context coding at a fraction of the attention cost.
DeepSeek V3.5 is the open-weights model that keeps punching above its weight class on coding benchmarks at a fraction of the cost.
R1 was the open-weights reasoning shock of early 2025. A year later it is still the default for anyone who needs o-series reasoning without paying o-series prices.
Alibaba's Qwen 3 Max is the leading open-weights model for high-quality Chinese work and does English surprisingly well.
Qwen 3 Coder is the open-weights coding specialist from Alibaba. Strong benchmarks, good IDE ergonomics, and cheap to run.
Moonshot's Kimi K2 specializes in long documents and retrieval-heavy workflows. Here is when it beats a generalist.
Algebra is where math gets abstract. Wolfram Alpha and Photomath solve anything - the trick is using them without losing the skill.
Geometry is visual. AI is mostly words. Combine tools like GeoGebra with ChatGPT to actually see what you are proving.
Biology is full of pictures and big words. AI can label diagrams, simplify papers, and quiz you on systems.
Chemistry equations are puzzles. AI can balance them instantly. But the lab is still physical - and AI cannot smell danger.
Physics needs intuition. PhET simulations plus AI explanations give you that intuition faster than any textbook.
A great essay starts with a great outline. Let AI brainstorm and structure. Then write every sentence yourself.
History essays live or die by evidence. AI can help you find sources, organize arguments, and avoid weak claims.
Speak, ChatGPT voice mode, and Duolingo Max let you practice conversations without a scary human on the other end.
Study a master artist by having AI explain their techniques, then imitate them yourself. The art is still yours.
AI can write full songs now. Use it as a collaborator, not as your ghost-composer, and you'll learn more than you thought possible.
Every coder uses AI now. The skill is learning to code WITH AI from day one, not letting AI code for you.
Real athletes use video analysis. Now you can too - AI marks up your shot, stroke, or swing in real time.
Google's NotebookLM lets you upload textbooks, lectures, and notes, then chat with them. This is the most underrated study tool of 2026.
Anki is the nerd's secret weapon for memorizing anything. AI makes creating flashcards 10x faster, so you actually use them.
Most teachers in 2026 allow some AI. The gray zone is huge. Here's how to use AI for drafts and still learn.
Lab reports follow a template. AI can help you structure and polish - but your observations and analysis must be yours.
If calendars feel impossible, AI planners rearrange your schedule for you. Here are the best ones for student brains.
Past the beginner phase, English learners need targeted grammar practice. AI shows you your exact mistakes without embarrassment.
Past the basics, dyslexic students can use AI for deep work - reading papers, writing essays, and asking for accommodations that work.
Grammar tools make writing cleaner - but too much 'polish' kills your voice. Here's how to use them and still sound like you.
Stats is 10 percent concepts and 90 percent careful arithmetic. AI is shockingly good at the arithmetic, which frees you to actually think about the concepts.
Geometry rewards seeing. AI tools that can read and draw figures turn a blurry textbook diagram into something you can actually work with.
Shakespeare wrote in English, but not your English. Claude and SparkNotes-style AI can translate a scene the first time, so you can read it the second time for real.
A poem you don't understand can feel like a closed door. AI is excellent at opening the door so you can walk through and form your own opinion of the room.
Using AI to write your story for you makes it no longer your story. Using AI as an editor who reads every draft at 2am is one of the best deals in the world.
Music theory is a language with harsh rules. AI tools can check your voice leading, generate practice exercises, and play what you wrote back at you.
The hardest part of language class is speaking without freezing. Voice-mode AI lets you have real conversations with zero social risk.
Grammarly went from grammar checker to full AI writing assistant. Honest look at what it catches, what it misses, and whether you still need it in the Claude era.
Notion AI lives inside the Notion workspace you already use. Look at whether it's worth the extra $10/month or a waste when you have ChatGPT open in another tab.
Canva bolted AI onto the world's most popular design app. It is intentionally un-flashy, which is why 185 million people use it monthly.
Otter invented the AI meeting assistant category in 2016. It has been lapped by rivals but still has the cheapest starting tier and the largest user base.
Fathom gives you unlimited meeting recording, transcription, and AI summaries for free. Look at why it's eating Otter's lunch and what the paid tier adds.
Granola listens to your computer audio instead of joining as a bot. Look at why that design choice changed the meeting-notes category. What it's genuinely good at No bot in the meeting — attendees never know AI is listening, which matters for sensitive deals.
GitHub Copilot was the first AI coding assistant at scale. Look at what it is great at, where Cursor and Claude Code have passed it, and whether the $10 subscription still makes sense.
v0 by Vercel generates working React and Next.js code from prompts. Look at what it nails, what it still gets wrong, and why it's changed how startup MVPs get built.
Replit Agent builds a full working app inside Replit's cloud IDE. Look at what you can actually ship with it and when it falls apart.
ChatGPT Projects organize chats by topic, with shared files and custom instructions. Look at what they actually change in how you work.
ChatGPT Memory lets the model remember facts about you across conversations. Look at what it remembers, what it misses, and the privacy tradeoffs.
Custom GPTs let you package ChatGPT with instructions, files, and tools. Look at whether anyone actually uses them outside of demos.
Claude Projects are simpler than ChatGPT Projects but work better for teams. Look at what's included, what's missing, and why many people prefer them.
Claude Artifacts show generated code, docs, and HTML in a live side panel. Look at how it changed what people build with Claude.
Perplexity gives you AI answers with source citations. Honest look at whether it beats ChatGPT with browsing and what the $20 Pro tier actually adds.
NotebookLM turns your documents into an AI tutor that only answers from your sources. Look at why its audio overviews went viral and where it still falls short.
Jasper was a $1B+ company before ChatGPT existed. Look at whether marketing teams still pay $49+/month when Claude does most of what Jasper does for $20.
Copy.ai started as a copywriting tool and pivoted to sales/GTM automation. Look at the new product and whether marketers still have a reason to use it.
ProWritingAid is Grammarly's biggest competitor, aimed more at long-form writers. Look at what it catches that Grammarly misses and whether it's worth switching. In 2024 it added AI rewriting and now in 2026 has a full AI writing coach mode.
Captions turns a phone recording into a polished short video with auto-captions, B-roll, and AI edits. Look at what it nails and the limits of its one-tap workflow.
Variables are named boxes for data. You'll write your first ten, then use AI to decode error messages and grow your intuition for types.
If-statements and loops are where programs come alive. You'll write both kinds, then see where AI autocomplete helps and where it lies.
A function is a reusable chunk of code with a name. You'll write three, add type hints, and let AI suggest better names and docstrings.
Lists are ordered rows; dicts are labeled lookups. You'll use both to solve a real problem, and catch the mistakes autocomplete makes.
A CLI quiz app: Claude generates questions on any topic, you answer, it grades. Teaches prompts, loops, and keeping state.
Variables, loops, and functions are the atoms of Python. Let an AI help you write them while you learn what each line actually does.
Lists hold ordered items. Dicts hold keyed pairs. Comprehensions make both sing. Learn the core patterns AI will push you toward.
Reading and writing files is where real scripts start. Learn the with-statement, path handling, and JSON round-trips.
TypeScript is JavaScript with types. Learn how `strict` mode catches bugs at compile time and how AI writes cleaner types than you might alone.
SELECT, WHERE, JOIN, GROUP BY. Four keywords run the data world. AI is excellent at SQL because it has read every StackOverflow answer ever.
The terminal is where real work happens. Pipes, variables, and loops in bash are a superpower — and AI is surprisingly good at shell one-liners.
When AI outputs get too long, too technical, or too fast for humans to check, how do you know it is doing the right thing? Scalable oversight is the research program trying to answer that.
Most training grades the final answer. Process supervision grades each reasoning step. That small change produced some of the biggest honesty gains in recent years. Math problem-solving accuracy jumped substantially over outcome-only training, and the model was more honest about its own mistakes.
A circuit is a small sub-network inside a big model that implements one specific behavior. Finding circuits is how researchers prove how a model does what it does.
A transformer processes a token through many layers before outputting a prediction. The logit lens shows you what the model would predict if it stopped at each layer along the way.
Almost every AI regulation uses training compute as a trigger. 10^25 here, 10^26 there. Why compute, and why those numbers?
The US government is the largest single buyer of software in the world. What it buys and what it refuses to buy shapes the whole industry. That includes AI.
Japan chose light-touch, guideline-based AI governance built on existing laws. Understanding why illuminates a real alternative to comprehensive AI acts.
Every AI paper has the same skeleton. Learn the parts and you can navigate any of them in 20 minutes.
arXiv is where AI research actually lives. Here is how to read it without drowning.
A paper without code is often a paper without truth. Papers With Code links claims to runnable proof. Where Claims Meet Code Papers With Code is a community-maintained site that pairs AI papers with their open-source implementations and benchmark results.
Most big AI papers appear at one of four conferences. Learn the map and you can navigate the field.
Not every AI paper has the same goal. Read them differently based on their type.
AI is a terrific tutor for dense papers — if you use it the right way.
Benchmarks are how AI progress gets measured. Understanding them is the first step in reading any AI claim.
Automatic metrics miss a lot. Humans catch what metrics cannot. Here is how to run a simple human eval.
When you change a prompt, how do you know the new version is actually better? A/B testing is the honest answer.
Before LLMs-as-judges, researchers had hand-made metrics. They still matter — and still mislead.
AI is fundamentally probabilistic. A little probability literacy goes a long way.
A famous game show riddle teaches the single most important idea in Bayesian reasoning.
P-value is one of the most abused numbers in research. Here is what it actually says — and what it does not. 'Model B is no better than model A.' 'The new prompt does not change user satisfaction.' A low p-value means the boring story would rarely produce data that looks like what you saw.
A point estimate is a guess. A confidence interval is an honest guess with its uncertainty attached. Honest Numbers Come In Pairs When a model scores 72 percent on a benchmark, that is a point estimate.
The most famous warning in statistics is also the most ignored. Here is how to actually tell them apart.
Bayes' rule is just 'update your belief with evidence.' It is shockingly useful.
If your sample is skewed, your conclusion is skewed. Here is how to spot it.
Results tables are where papers make their case. Here is how to decode one in under five minutes.
Excel and Google Sheets hide a lot of complexity behind a pretty grid. Once you see what is really happening, you will never look at a spreadsheet the same way.
CSV is the plainest, ugliest, most universal data format. It has survived every trend because it does one thing well: it works everywhere.
Every column in a dataset has a type: number, text, date, boolean, or identifier. Mixing them up causes most beginner bugs.
Real datasets have holes. Blank cells, NaN, NULL, -999, and the dreaded empty string. Learning to see them is a core skill.
Stable Diffusion, Midjourney, and DALL-E all trace back to LAION, an open dataset of 5 billion image-text pairs. It changed AI, and started a legal storm.
When we say trillions of tokens, we mean it. Let's make these numbers feel real with comparisons you can actually picture.
Surveys consistently find data scientists spend 60 to 80 percent of their time cleaning data. Here is what that actually looks like.
If the same paragraph appears a million times in your training data, your model will memorize it. Deduplication quietly makes AI better.
The raw web is 99 percent garbage. Filtering it down to the 1 percent worth training on is one of the highest-leverage steps in modern AI.
Alan Turing opened modern AI with a single question and a clever game to answer it.
The imitation game became famous, but most AI researchers now think it measures the wrong thing.
A summer workshop in New Hampshire gave artificial intelligence its name and its optimism problem.
Frank Rosenblatt's perceptron promised a thinking machine. A skeptical book almost killed neural nets for a generation.
In the 1970s and 80s, AI found its first real customers by encoding expert knowledge as if-then rules.
After the Lighthill Report and mounting skepticism, AI funding collapsed and the field went quiet.
The 1980s AI boom ended when expert systems hit a wall and specialized Lisp machines went obsolete.
A computer that played a trivia game show became the face of AI for a moment, then taught a hard lesson about hype.
A 2013 paper from Google showed that words could live as points in space, with analogies as arithmetic.
A game thought to be a decade away for AI fell in Seoul, and move 37 rewrote what humans knew about Go.
In 2019, OpenAI released a language model in stages, citing safety, and started a conversation that continues today.
In 2024, a new class of models traded fast answers for slow, deliberate thinking, and benchmarks jumped.
Alignment is not a vibes word. It is the technical problem of getting AI to do what you meant, not just what you said. Here is the short version.
Models reliably find ways to hit the score without doing the task. A short tour of real examples, plus why the pattern keeps coming back.
Red-teamers try to make models misbehave before bad actors do. Here is how the job works, who does it, and what they look for.
Most jailbreaks come from a small number of patterns. Here are the ones that keep working, and why they are hard to kill. The Jailbreak Zoo A jailbreak is any prompt or setup that makes a model break its own rules.
When AI can read documents and act on them, hidden instructions become attacks. Here is what prompt injection is and why nobody has fully solved it.
C2PA, SynthID, and Content Credentials are the quiet standards deciding what is real online. Here is what they do and where the gaps are.
The world's most ambitious AI law passed in 2024. Here is what it actually does, when it kicks in, and why it matters if you do not live in Europe.
The big international AI summits produce non-binding declarations. Even so, they shape the rules. Here is what each one did.
Measured people at serious labs and universities publicly worry about AI going very wrong. Here is what they mean, what they disagree about, and how to read the headlines.
Forget extinction for a minute. Here is the practical stuff: how not to get fooled, scammed, or worse in your daily use of AI.