111 terms
A
-
A/B test
Half the users see the old version, half see the new one; whichever gets better results wins.
Read article #59: How to measure AI product quality: sample questions and A/B tests
-
A2A Agent to Agent
A shared standard that lets one company's agent hand work to another company's agent.
Read article #77: How do multi-agent systems work? Orchestrator, subagents and reviewer
-
Adaptive learning
A machine learns how you learn: get it right and it gets harder, get it wrong and it goes back to the basics you're missing, like a private tutor.
Read article #04: What is adaptive learning? Turn AI into your own private tutorSee also #06
-
AI agent
AI that doesn't just answer, but is given tools to do work for you over many steps, and stops to ask you before anything important.
Read article #16: What is an AI agent? When AI does multi-step work for youSee also #68#72#98
-
Alert
Each metric has a red line; when the needle crosses it, an alarm rings so the person on call checks right away.
Read article #85: What is AI system monitoring? Dashboards, alerts and driftSee also #79
-
API
The door apps use to send questions over the internet to an AI model and get answers back; most AI apps rent their model through this door.
Read article #80: Where does AI run? Cloud, data centers, APIs and on-device AISee also #114
-
Artifact
A finished piece that opens right next to Claude's chat: a document, slides, a small spreadsheet or a tool you can click.
Read article #64: What are Artifacts in Claude? Turn answers into things you can use
-
Artificial neuron
Takes a few numbers, multiplies them by weights, adds them up and decides; surprisingly simple.
Read article #38: How does AI learn? Weights, neurons and neural networks
-
Attention
Each word "looks" at every other word in the sentence and scores which ones matter most to it; this is the heart of today's chatbots.
Read article #21: What is attention in AI? Where AI focuses when it reads
-
Autoscaling
At peak hours, the system turns on more machines by itself to keep up; if it can't, you see "overloaded, try again later".
Read article #80: Where does AI run? Cloud, data centers, APIs and on-device AI
B
-
Backpropagation
The mistake at the output is passed back through each layer of the neural network; knobs that caused more of the mistake get turned more.
-
Benchmark
AI's exam: thousands of questions with answers ready, and every AI takes the same test; each test only measures one skill.
Read article #45: What is an AI benchmark? What "scored 90%" really meansSee also #46#83
-
Branching scenario
An exercise where each choice you make opens a different outcome, so you see the consequences for yourself.
Read article #89: What is a branching scenario? Learning with AI choose-your-path stories
C
-
Chain of thought
Telling AI to "think step by step" so it writes out each step before answering; it guesses better and you can check it more easily.
Read article #103: What is chain of thought? Tell AI to think step by stepSee also #07
-
Citation
The answer says which document it came from, so you can open it and check.
Read article #82: What are output guardrails? How chatbots check answers before sendingSee also #30
-
Claude Code
Claude's coding agent: describe a job in plain words, and it writes the code, tests it and fixes bugs.
Read article #70: What is Claude Code? Build your own tools without knowing how to code
-
Claude Design
Claude's design tool: chat on the left, canvas on the right; describe it in words and get slides, a landing page or a prototype.
Read article #65: What is Claude Design? Make slides and web pages without design skills
-
Cloud
Big AI doesn't live in your phone; it runs in faraway data centers, reached over the internet.
Read article #80: Where does AI run? Cloud, data centers, APIs and on-device AISee also #79
-
Connector
A bridge that lets AI into Gmail, Google Drive, your calendar or other tools; many connectors run on MCP.
Read article #67: What are Skills in Claude? Teach Claude a process once, use it foreverSee also #73#32
-
Context window
AI's "desk": it sees whatever is on the desk, and when the desk is full, old sheets get pushed off and AI stops noticing them.
Read article #09: Why does AI forget in long chats? What context is, and 2 tips to fix itSee also #76#98
-
Cowork
Claude's AI agent: hand it a whole long task, and it makes a plan and works through it step by step in the cloud.
Read article #68: What is Claude Cowork? Hand long tasks to an AI agent
-
Custom instructions
A note stuck on your assistant's desk: who you are, whether you want long or short answers, how to talk to you; AI reads it first every time.
Read article #27: What are custom instructions? Teach AI who you areSee also #102
D
-
Dashboard
A screen showing an AI system's metrics: response speed, cost, refusal rate, and how often customers are unhappy.
Read article #85: What is AI system monitoring? Dashboards, alerts and drift
-
Data center
Buildings full of server racks with GPUs, where big AI models run; they use as much power as a whole neighborhood.
Read article #80: Where does AI run? Cloud, data centers, APIs and on-device AISee also #43
-
Data contamination
AI's test questions slip into its training data, so it gets answers right by remembering, not understanding.
Read article #46: What is data contamination? When AI crams for benchmark tests
-
Debrief
After each simulation practice, looking back at where you went wrong and what you'll do differently next time.
Read article #91: Traps when practicing with AI simulations, and how to avoid them
-
Deepfake
AI copying a person's voice and face, even live during a video call.
Read article #14: What is a deepfake? How to spot a fake call in a loved one's voiceSee also #109#110
-
Design system
Save your colors, fonts and logo once, and everything you make afterwards matches your brand automatically.
Read article #65: What is Claude Design? Make slides and web pages without design skills
-
Distillation
A big model is the teacher, and a small model learns from how it answers; the result is a small, fast, cheap model that's still pretty smart.
Read article #115: What is distillation in AI? A big model teaches a small one
-
Drift
The world changes and users ask new kinds of questions, but the AI system stays the same.
Read article #85: What is AI system monitoring? Dashboards, alerts and drift
E
-
Elo
A chess-style scoring system: winners gain points, losers lose some, and beating a strong opponent earns more.
Read article #47: What is LMArena? Ranking AI with blind matches and Elo scores
-
Embedding
AI places every word on a giant map where words with similar meanings sit close together; that position is the meaning.
Read article #20: What is an embedding? How AI understands word meaning with a mapSee also #25
-
Error compounding
Each step being 95% right sounds great, but chain ten steps together and the whole thing is only about 60% right.
Read article #78: When not to use multi-agent: 3 ways AI teams go wrong
-
Eval set
A set of real customer questions with the answers you want, used to grade an AI product every time you change it or switch models.
Read article #83: What is evaluation? Why you test again before switching AI modelsSee also #59#85
-
Evaluation
Rerunning the test set and scoring it before switching models or editing a prompt; it only goes live once it passes.
Read article #83: What is evaluation? Why you test again before switching AI modelsSee also #79
F
-
Feedback loop
Incidents become lessons, and lessons become test questions: questions that caused problems are added to the eval set.
Read article #85: What is AI system monitoring? Dashboards, alerts and drift
-
Few-shot
Giving AI a few sample examples so it picks up the tone, length and layout you want.
Read article #101: What is few-shot prompting? Give examples so AI writes in the right toneSee also #10
-
Fine-tuning
Extra training for a model that has already learned, using a special set of examples, like taking a job course after high school.
Read article #113: What is fine-tuning? Explained in 60 seconds
-
Flashcard
Cards with a question on one side and the answer on the other for review; AI can make them from your videos or documents.
Read article #93: How to turn YouTube videos into lessons with questions using AISee also #94
G
-
Goodhart's law
When a number becomes the goal, it stops measuring well.
Read article #46: What is data contamination? When AI crams for benchmark tests
-
GPU graphics chip
A special kind of chip for running and training AI, lined up by the tens of thousands in data centers.
Read article #43: Why do bigger AI models get smarter? Scaling laws explainedSee also #80#114
-
Gradient descent
Like walking down a mountain blindfolded: feel which way lowers the error most, take a small step, and repeat millions of times.
Read article #40: What is gradient descent? How AI reduces error, like walking down a foggy mountain
-
Guardrails
A separate guard layer around the model: it blocks off-topic questions, hides sensitive information and checks answers before they reach you.
Read article #81: What are input guardrails? A chatbot's gate on the way inSee also #82#56#79
H
-
Hallucination
AI gives a wrong answer that sounds true, because when it doesn't know, it still guesses whatever sounds most likely.
Read article #08: What is AI hallucination? Why AI makes things up and still sounds confidentSee also #98#106
-
Human in the loop
For real actions like sending money, AI only gets things ready, and a person has to click confirm.
Read article #82: What are output guardrails? How chatbots check answers before sendingSee also #58
I
-
Input guardrails
Checking questions before they reach AI: hiding card numbers, staying on topic, blocking trick prompts.
Read article #81: What are input guardrails? A chatbot's gate on the way inSee also #79
-
Integrated learning
Learning several subjects at once through a real project, like planning a trip.
-
Interactive learning
Turning you from a watcher into a doer: answering questions mid-lesson, asking about hard parts, trying your own examples.
Read article #92: Why do you forget after watching a lesson? What interactive learning is
-
Interactive simulation
A small page with buttons, sliders and charts so you can drag, try and see for yourself; AI can build one in seconds.
Read article #90: How to get AI to build an interactive simulation with sliders
-
Interpretability
The research field that shines a light into the "black box" to find the concepts inside AI.
Read article #44: Why is AI a black box? How interpretability looks inside
K
-
Knowledge cutoff
The day AI stopped reading its training data; it knows nothing about what happened after that day.
Read article #22: Where does AI learn from? Training and AI's knowledge cutoff dateSee also #98
L
-
Leaderboard
A table ranking AIs by score; for example, LMArena collects millions of users' votes from blind matches.
Read article #47: What is LMArena? Ranking AI with blind matches and Elo scoresSee also #59
-
LiDAR
Measuring distance with light: send a laser flash, time the trip there and back, and build a 3D map for AI to read.
Read article #05: What is LiDAR? How self-driving cars measure distance with light
-
LLM as a judge AI grading AI
Using an AI as the judge to grade answers against written criteria; fast, but people still need to double-check.
Read article #83: What is evaluation? Why you test again before switching AI modelsSee also #49
-
Logging
Every question leaves a line in the log: the time, the question with private details hidden, the answer, how long it took and what it cost.
Read article #84: What are logging and tracing in AI systems? AI's black box recorderSee also #79
-
Loss error, loss function
A number measuring how wrong AI is compared with the right answer; the whole learning process is about pushing it down.
Read article #39: What is loss? How AI knows how wrong it isSee also #40
M
-
Machine learning
Instead of writing rules by hand, you give the machine lots of examples and let it find the patterns itself.
Read article #10: What is machine learning? Explained with a spam filterSee also #11
-
MCP Model Context Protocol
A shared USB-C-style port for AI: any tool with an MCP server can plug into any AI app that supports MCP.
Read article #73: What is MCP? Model Context Protocol, a shared port for AISee also #74#75
-
Memory
General things about you that an AI app remembers between chats; you can view, edit or delete them.
Read article #66: What does Claude remember about you? Projects and Memory in ClaudeSee also #09
-
Mixture of Experts MoE
A model split into many small experts; each token only calls the few that fit best, so it's big but still fast.
Read article #117: What is Mixture of Experts (MoE)? Why big models can still be fast
-
Model
AI's "brain"; one company offers small tiers that are fast and cheap, middle tiers that are balanced, and large tiers that are powerful but pricey.
Read article #18: Which AI model should you choose? Haiku, Sonnet, Opus and Fable comparedSee also #43#98
-
Model collapse
When AI only learns from AI-made data, it gets blurrier over time, like a photocopy of a photocopy.
Read article #116: What is synthetic data? When AI makes data to teach AI
-
Monitoring
Tracking speed, cost and wrong-answer rates on a dashboard, so you know about problems before customers complain.
Read article #85: What is AI system monitoring? Dashboards, alerts and driftSee also #79
-
Multi-agent
Splitting a big job among several agents working together; only worth it when the job splits into independent parts.
Read article #76: What is multi-agent AI? Why several AIs work on one jobSee also #77#78
-
Multimodal
AI that takes in text, images and sound in one brain, because everything gets cut into tokens.
Read article #55: What is multimodal AI? One brain that listens, talks and seesSee also #13
N
-
Neural network
Thousands of artificial neurons stacked in layers, each layer passing its results to the next.
Read article #38: How does AI learn? Weights, neurons and neural networks
O
-
Observability
Being able to see clearly inside an AI system, thanks to logs and traces.
Read article #84: What are logging and tracing in AI systems? AI's black box recorder
-
On-device AI
Small models that run right on your phone: no internet needed, your data stays on the device, but they're less smart.
Read article #80: Where does AI run? Cloud, data centers, APIs and on-device AI
-
Open-weight open model
A model you can download with all its weights; fully open-source models also share their data and how they were trained.
Read article #114: What are open-source and open-weight AI? How they differ
-
Orchestrator lead agent
The agent that makes the plan, splits work among subagents, then puts the results together for you.
Read article #76: What is multi-agent AI? Why several AIs work on one jobSee also #77
-
Output format
Saying exactly how AI should answer: a table, a checklist, three bullet points, or JSON or CSV.
Read article #104: How to make AI answer in tables, checklists or JSON (output format)
-
Output guardrails
Checking answers before they're sent: any careless promises, someone else's information leaked, or missing sources?
Read article #82: What are output guardrails? How chatbots check answers before sendingSee also #79
-
Overfitting
AI memorizes each training example: nearly perfect on familiar questions, wrong on new data.
Read article #42: What is overfitting? When AI memorizes instead of understanding
P
-
PII masking
Card numbers, phone numbers and ID numbers are replaced with labels before they reach AI.
Read article #81: What are input guardrails? A chatbot's gate on the way inSee also #15
-
Pretraining
The stage where AI reads a huge amount of text and learns by guessing the next word, repeated billions of times.
Read article #22: Where does AI learn from? Training and AI's knowledge cutoff dateSee also #07
-
Project
A private workroom for one task: its own instructions, documents and every related chat, all in one place.
Read article #29: What are Projects for in AI apps?See also #66#30
-
Prompt
The message you send AI; include all four parts (role, context, task, format) and AI gets closer to what you want.
Read article #03: What is a prompt? The 4-part formula for getting the answer you wantSee also #28#107
-
Prompt chaining
Splitting a big task into small steps, with each step's result feeding into the next.
Read article #105: What is prompt chaining? Break a big task into a chain of small prompts
-
Prompt injection
Strange instructions hidden in data, like an email telling AI to send out your contacts, that trick AI into thinking they came from you.
Read article #75: Is MCP safe to use? Prompt injection and 3 safe habitsSee also #81
-
Prompt template
A prompt with blanks to fill in, reused every week and improved after each use.
Read article #107: What is a prompt template? Build your own set of prompt templatesSee also #28
R
-
RAG
Letting AI find the right passages in your documents and read them before answering, so it makes things up far less.
Read article #25: What is RAG? How to let AI read your own documentsSee also #30#113
-
Reasoning thinking mode
AI writes a draft first: it breaks the problem into steps and checks itself before answering; slower, but better at hard problems.
Read article #26: What is AI's thinking mode? When to turn on reasoningSee also #103
-
Recommendation algorithm
Machine learning that uses how long you stop to watch, and what people like you enjoy, to guess which clip will keep you watching longest.
Read article #11: How does TikTok's recommendation algorithm work? Why it seems to know you
-
Red team
People who play the bad guys and try to trick AI into doing things it shouldn't, so holes get fixed early.
Read article #49: How is AI tested for safety and honesty? Red teams and AI judges
-
Regression test
A new model has to redo the whole old test and be compared with the current model before it can replace it.
Read article #83: What is evaluation? Why you test again before switching AI models
-
Retrieval practice
Answering questions yourself helps you remember longer than rereading many times.
Read article #92: Why do you forget after watching a lesson? What interactive learning isSee also #94
-
RLHF
AI writes a few answers and people score which is better; helpful, honest answers get rewarded, made-up ones lose points.
Read article #23: How do chatbots learn to answer and say no? From parrot to assistant
-
Role
Telling AI to play a specific role so it picks the right knowledge and tone; the clearer the role, the better.
Read article #102: What is a system prompt? Give AI a role and set the rulesSee also #03
-
Role-play
AI plays an interviewer, customer or partner so you can practice conversations like they're real.
Read article #88: How to role-play with AI to practice interviews and conversationsSee also #36
S
-
Scaling law
Push up the knobs, the data and the computing power together, in balance, and the error drops quite steadily.
Read article #43: Why do bigger AI models get smarter? Scaling laws explained
-
Scheduled task
Say it once, "every Monday at 7 a.m., do this", and AI does it on time and lets you know.
Read article #31: How to let AI do weekly tasks on its own with scheduled tasksSee also #68
-
Self-critique
Asking AI to point out the weaknesses in its own work, then write a better version.
Read article #106: What is self-critique? How to get AI to grade and fix its own work
-
Simulation-based learning
Practicing in made-up situations, seeing consequences right away without fear of mistakes, and repeating until it feels natural.
-
Skill
A recipe card for Claude: the steps and a layout template; when a task matches, Claude takes it out and uses it.
Read article #67: What are Skills in Claude? Teach Claude a process once, use it foreverSee also #107
-
Socratic method
Teaching by asking questions back, so learners work out the answers themselves.
Read article #95: What is study mode? Let AI ask you questions instead of giving answers
-
Speech-to-text
AI turns sound into a spectrogram picture and reads the shapes into words; double-check names and numbers.
Read article #50: How does AI understand speech? Speech-to-text explained simply
-
Study mode
A mode where AI gives hints and asks you step by step instead of handing over the answer.
Read article #95: What is study mode? Let AI ask you questions instead of giving answers
-
Subagent
A "worker" agent that takes one part of the job, has its own context window and tools, and sends back a short summary.
Read article #76: What is multi-agent AI? Why several AIs work on one jobSee also #77
-
Synthetic data
Data made by AI to teach AI; only useful when filtered carefully and mixed with real data.
Read article #116: What is synthetic data? When AI makes data to teach AI
-
System prompt
Instructions that sit above the whole conversation, like rules AI has to follow for every question.
Read article #102: What is a system prompt? Give AI a role and set the rulesSee also #81
T
-
Temperature
The knob for chance when AI draws the next word: low means steady, high means creative but more likely to drift off topic.
Read article #24: Why does AI give different answers to the same question? What temperature isSee also #98
-
Text-to-speech
AI turns text into sounds, adds intonation, then makes a sound wave; a few seconds of a voice sample is enough to clone a voice.
Read article #51: How does AI turn text into speech? Text-to-speech and voice cloning
-
Token
The pieces AI cuts a sentence into before reading, each turned into a number; length and price are both counted in tokens.
Read article #19: What is a token? Why AI doesn't read words the way people doSee also #55#60
-
Tool calling
AI writes a "request slip" (a tool call), the app does the job, then hands the result back for AI to answer.
Read article #72: What is tool calling? How AI uses toolsSee also #74
-
Tracing
Tying all the steps of one question into a single thread, to see what each step received and returned, and where the mistake was.
Read article #84: What are logging and tracing in AI systems? AI's black box recorder
-
Transcript
A video's speech as text; paste it into AI to turn the video into a lesson.
Read article #93: How to turn YouTube videos into lessons with questions using AISee also #53
V
-
Virtual lab
Doing experiments on a screen, even changing things you can't change in real life; for example, the free PhET library.
Read article #87: What is a virtual lab? Learning with free PhET simulations
-
Voice clone
AI copying someone's voice from just a short recording; scammers use it to call family members asking for money fast.
Read article #108: AI voice cloning scams: how to spot them and stay safeSee also #51#14
-
Voice mode
Talking with AI out loud like a phone call; it answers in about half a second, and you can interrupt.
Read article #52: How does voice chat with AI work? Voice mode explainedSee also #88#63
W
-
Weights
The "knobs" inside AI; learning means the machine turns billions of them itself, a little at a time.
Read article #38: How does AI learn? Weights, neurons and neural networksSee also #114
Z
-
Zero-shot
Asking with no examples at all, so AI has to guess what you want and gives you an average, safe tone.
Read article #101: What is few-shot prompting? Give examples so AI writes in the right tone
No term matches yet. Try another word, or look in the video list.