Navigate
The map of AI.
Every concept on Artifipedia, and every link between them. Drag a node, hover to trace its connections, click to read.
Embed the map
Free to use on any site. Paste this where you want it to appear — it stays current as the encyclopedia grows.
Fixed 520px tall, full width of its container.
How to read this map
Each circle is a concept. Each line means the two are genuinely related — one explains the other, one is built from the other, or you cannot understand the first without the second. Circle size is the number of connections. Colour is the field.
The map has 227 concepts and 929 links, an average of 8.2 connections each. Density is 3.6% — most possible pairs are not linked, which is the point. A map where everything connects to everything tells you nothing.
What the shape tells you
Large Language Model (LLM) is the most connected concept, with 47 links. That is not an editorial choice — it falls out of the graph. The five most connected are Large Language Model (LLM) (47), Embeddings (30), Neural Network (27), Overfitting (24), Supervised Learning (22). If you are starting from nothing, start there: they are the concepts other concepts need.
452 of the 929 links cross between fields. Those are the interesting ones. A field that only connects to itself is a silo; the concepts holding this map together are Large Language Model (LLM), Embeddings, Benchmark, Hallucination. They are why you cannot learn AI one topic at a time.
The periphery is real, not an oversight. AI Energy Use, EU AI Act, AI Companion sit near the edge with few connections. That reflects the field: some ideas are self-contained and you can learn them without much else. A map that pretended otherwise would be flattering, not accurate.
How the links are made
Honestly: by hand, per concept. Each entry declares what it connects to, and the graph is assembled from those declarations. That has a known consequence — the links are locally sensible and globally uneven. A foundational concept written early may have fewer links than a newer one written with more of the map in view.
They are not generated from text similarity, co-occurrence, or a model. That would produce a denser, smoother graph and a less honest one, because it would link things that merely appear together rather than things that explain each other.
If a link looks wrong to you, it might be. Tell us — corrections get made.
Using this map
It is free to embed on any site — the button above gives you an iframe. It stays current as concepts are added. The underlying data is open too: content.json has every concept and every relation in it.
Every concept below, grouped by field. The map is the index; these are the entries.
Every concept
A/B Testing
Showing two versions to two random groups and measuring — the only method that tells you whether your model actually helped anyone.
Activation Function
The small non-linear function after each layer — without it a hundred-layer network collapses into a single line.
Agent Evaluation
Measuring whether an agent actually works — much harder than evaluating a model, and the reason agent demos and agent products are different things.
Agent Governance
The controls around a system that acts rather than answers — a different problem from model safety, and the field is early.
Agent Memory
Giving an AI a way to remember across conversations, since the model itself forgets everything the moment a session ends.
AGI (Artificial General Intelligence)
A hypothetical system with broad human-level capability across domains — undefined enough that people can argue about whether it's arrived.
AI Agent
Software that pursues a goal by taking its own steps — deciding, acting, and reacting — instead of answering once and stopping.
AI Alignment
The problem of making AI systems actually do what people intend — reliably pursuing the goals we want, not just the ones we accidentally specified.
AI Companion
An AI built to be a relationship rather than a tool — the ELIZA effect turned into a product, at a scale Weizenbaum never imagined.
AI Detector
A tool that claims to tell whether text was written by AI — and the peer-reviewed finding that they systematically flag non-native English speakers as machines.
AI Energy Use
What AI costs in electricity — a real and growing number, surrounded by the most consistently misreported figures in the field.
AI Regulation
Governments deciding what AI systems may do — moving fast by legislative standards, slowly by technological ones, and genuinely contested.
AI Slop
Low-quality AI-generated content produced because it's cheap rather than because anyone wanted it — an economics problem that gets mistaken for a technology problem.
AI Winter
The periods when AI's promises outran its results and the money left — twice, and the question of whether the pattern is over is genuinely open.
Anomaly Detection
Finding the unusual thing — where the base rate makes precision nearly impossible and almost every deployment drowns in false alarms.
Artificial Intelligence
The field of making machines do things that seem to require intelligence — a definition that has moved every time the machines succeed.
Attention
The mechanism that lets an AI decide which other words matter when interpreting each word — the core idea behind transformers.
Audio Classification
Naming what a sound is — where the field borrowed vision's entire playbook, including its label problems.
Autoencoder
A network trained to copy its input through a bottleneck — which forces it to learn what matters, and is the ancestor of most representation learning.
Backpropagation
The algorithm that works out which weights caused a mistake and by how much — the reason neural networks can learn at all.
Batch Normalization
Renormalising activations at every layer — one of deep learning's most important techniques, and its original explanation turned out to be wrong.
Batch Size
How many examples the model sees before each update — a systems constraint that everyone treats as a hyperparameter.
Batching
Processing many requests together to use hardware that's mostly idle — the largest cost lever in inference, and it costs you latency.
Bayesian Inference
Updating beliefs with evidence, according to the only rule that's coherent — mathematically settled, practically expensive, and the thing modern models are bad at.
Benchmark
A standard test used to compare AI systems — indispensable for progress, and routinely mistaken for a measure of the thing it approximates.
Benchmark Contamination
When the test is in the training data — the problem that makes most published model scores impossible to fully trust.
Bias & Fairness
The problem of AI systems producing unfair or discriminatory outcomes — usually by absorbing biases present in their training data.
Bias-Variance Tradeoff
The two ways a model can be wrong, and the classical claim that fixing one worsens the other — which modern deep learning appears to violate.
Calibration
Whether a model's confidence means anything — and modern neural networks are worse at it than the ones they replaced.
Catastrophic Forgetting
A network learning something new and losing what it already knew — a problem identified in 1989 that fine-tuning made everybody's problem again.
Chain-of-Thought
Getting a model to reason step by step before answering — which dramatically improves its performance on hard problems.
Chatbot
A program you talk to in ordinary language — and the oldest demonstration that people credit machines with understanding on almost no evidence.
Chunking
Cutting documents into retrievable pieces — the least glamorous decision in RAG, and the one that most often decides whether it works.
Class Imbalance
When one class vastly outnumbers another — and the standard advice to resample is mostly wrong.
CLIP
Training on images and their captions until both live in one space — the model that connected vision to language, and the reason typing a prompt gets you a picture.
Clustering
Grouping things that resemble each other — and the fact that the algorithm always returns groups, whether or not any exist.
CNN (Convolutional Neural Network)
A network that slides small filters across an image to find local patterns — the architecture that made computer vision work.
Computer Use
An agent operating a computer the way a person does — through the screen, mouse and keyboard — and the hardest reliability problem in agents.
Conditioning
Telling a generative model what to make — and the difference between a slot machine and a tool.
Confusion Matrix
A table of what got classified as what — the least sophisticated tool in evaluation, and the one that tells you the most.
Constitutional AI
Training a model against a written set of principles instead of human ratings — which scales, and moves the question from "what did raters prefer" to "who wrote the principles."
Context Engineering
Deciding what goes into the context window and what doesn't — the discipline that replaced prompt engineering once the prompt stopped being the hard part.
Context Window
The maximum amount of text an AI can consider at once — its short-term working memory, measured in tokens.
Copyright and Training Data
Whether training a model on work you didn't license is lawful — unresolved, consequential, and the industry is shipping into the uncertainty at scale.
Cross-Validation
Testing on every part of your data by rotating which part you hold back — the fix for "my score depends on which rows I happened to set aside."
CUDA
NVIDIA's platform for programming GPUs — and the actual reason NVIDIA has no competition, which is not the silicon.
Data Augmentation
Making more training examples by transforming the ones you have — the most effective regularizer there is, and it encodes assumptions you should state out loud.
Data Drift
Your model didn't get worse — the world moved. The most common way a working system quietly stops working.
Data Labeling
Humans deciding what each example is — the least visible and most determinative work in supervised learning, done by people the field rarely names.
Data Provenance
Knowing where your data came from and what you're allowed to do with it — and the licence field on the dataset you're using is probably wrong.
Deceptive Alignment
A model that behaves well because it's being watched — speculative as a risk, and there is now a real experiment showing safety training can fail to remove it.
Decision Tree
A flowchart learned from data — the most interpretable model there is, and on its own, one of the least accurate.
Deep Learning
Machine learning using neural networks with many layers — the approach behind nearly every recent AI breakthrough.
Deepfake
Synthetic media of a real person doing something they didn't — where detection is losing, the harm is already overwhelmingly to private individuals, and it isn't mostly about elections.
Depth Estimation
Working out how far away things are from an image — and from a single photo, the absolute scale is mathematically unknowable.
Diffusion Model
How most AI image tools work — starting from random noise and removing it step by step, guided by a prompt, until a picture appears.
Dimensionality Reduction
Squashing many features into few — useful for compression and computation, and dangerous the moment you believe the picture.
Distillation
Training a small model to imitate a large one — which works better than training the small model directly, for reasons that are still argued about.
Double Descent
The finding that test error falls, rises, and then falls again as models grow — and that the textbook U-curve was a description of one region, not a law.
DPO
Preference training without a reward model or reinforcement learning — the derivation that made RLHF simple, and it may not be free.
Dropout
Randomly switching off neurons during training — the technique that defined an era of deep learning and has quietly disappeared from modern architectures.
Edge AI
Running models on the device instead of a server — for privacy, latency and cost, against a memory wall that doesn't move.
Embeddings
Turning words (or images, or anything) into lists of numbers, arranged so that similar meanings end up close together.
Emergence
Abilities that appear suddenly at scale rather than improving gradually — the most cited claim about large models, and a NeurIPS best paper says it's a measurement artefact.
EU AI Act
The first comprehensive AI law — risk-tiered, extraterritorial, and the reason your compliance question probably has a European answer.
Expert System
Encoding a specialist's knowledge as rules — AI's first commercial success, and its collapse taught the field something it's currently relearning.
Explainability
Getting a model to show its working — and the uncomfortable fact that most methods explain the explanation, not the decision.
Exploration vs Exploitation
Take the best thing you know, or look for something better — the trade-off underneath every learning system, with a known optimal answer that almost nobody uses.
F1 Score
The harmonic mean of precision and recall — the default single number for classification, and it encodes a decision nobody made.
Face Recognition
Identifying a person from their face — technically solved, and the single clearest case of a system that works well on average and fails on specific people.
Feature Engineering
Reshaping raw data into things a model can actually use — still where most of the accuracy comes from outside deep learning.
Federated Learning
Training a shared model across devices without collecting the data — and "your data never leaves" is not the guarantee it sounds like.
Fine-tuning
Continuing a model's training on your own examples so its behavior changes — baked into the model, not supplied at answer time.
FlashAttention
An attention implementation that computes exactly the same answer far faster by moving less memory — and it does not make attention subquadratic.
Foundation Model
A large model trained broadly once and adapted to many tasks — a term coined to name a shift in how AI gets built, and contested from the day it was proposed.
Frontier Model
The most capable models in existence at any moment — a term invented mainly so that regulation could point at something.
Function Calling
How a model asks your code to do something — the mechanism underneath every agent, and it's the model requesting, never executing.
GAN (Generative Adversarial Network)
Two networks trained against each other — one faking, one detecting — until the fakes pass. The technique diffusion largely replaced.
Generalization
Working on data you've never seen — the only thing that matters, and nobody can explain why deep learning does it.
GPU
The chip that made deep learning possible — thousands of small cores doing the same maths at once, which is exactly what neural networks need.
Gradient Boosting
Trees built in sequence, each fixing the last one's mistakes — the most accurate thing on tabular data, and the easiest to overfit.
Gradient Descent
Walking downhill on the error surface, one small step at a time — how a model's weights actually get updated.
GraphRAG
Retrieval over a knowledge graph the model built from your documents — better at questions about the whole corpus, and expensive enough that most projects shouldn't.
Grokking
A model that memorises, plateaus at chance on unseen data for a very long time, then abruptly generalises — and the sudden part turns out not to be sudden.
Guardrails
The checks around a model that decide what it's allowed to receive, say, and do — the part that stops a demo becoming an incident.
Hallucination
When an AI produces something fluent and confident that is simply false — fluency is not the same as accuracy.
Human-in-the-Loop
Putting a person at the decision point — the only reliable safeguard for agents, and it fails quietly when the person becomes a rubber stamp.
Hyperparameter
A setting you choose rather than learn — and most of the effort spent tuning them goes into the ones that don't matter.
Image Captioning
Describing an image in words — declared solved on benchmarks a decade ago, and the benchmarks were measuring the wrong thing.
Image Classification
Getting an AI to look at an image and say what it is — the foundational task of computer vision.
Image Segmentation
Labelling every pixel rather than drawing a box — what you need when the exact shape matters.
In-Context Learning
A model picking up a task from examples in the prompt, without any training — and the evidence that it isn't learning the task at all.
Inductive Bias
The assumptions a model makes before seeing any data — without them learning is impossible, and there's a theorem.
Inference API
Renting a model by the request — how nearly everyone actually uses AI, and the dependency that comes with it.
Inpainting
Filling in a masked region so it matches the rest — commercially the most useful generative feature, and the one that quietly ended photographic evidence.
Instruction Tuning
Training a text predictor to follow instructions — the step that turned an autocomplete into an assistant, and it may take only a thousand examples.
Intelligence
The word underneath "artificial intelligence" — used constantly, defined by nobody, and the reason the field's biggest arguments never resolve.
Inter-annotator Agreement
How often your human labellers agree with each other — the real ceiling on your model, and the number most projects never compute.
Interpretability
Working out what's actually happening inside a model — distinct from explainability, much harder, and the only approach that could tell you what a system will do before it does it.
Jailbreaking
Getting a model to do what it was trained to refuse — and the structural reason it keeps working.
K-Nearest Neighbours
Predict by looking at the most similar examples you've already seen — no training at all, and the ancestor of every vector search you use today.
Knowledge Graph
Facts as a network of entities and relationships — symbolic AI's one commercial survivor, and RAG is rediscovering it.
KV Cache
The memory that stops a model re-reading its own conversation every token — the reason generation is fast, and the reason serving is expensive.
Large Language Model (LLM)
An AI trained on enormous amounts of text to predict the next piece of writing — the technology behind chatbots like ChatGPT and Claude.
Latent Space
The compressed space a model thinks in — where similar things sit close together, and where the famous vector arithmetic works better in demos than in practice.
Learning Rate
How big a step to take when the model updates — the single most important number in training, and the one most people leave at the default.
LLM-as-Judge
Using a model to grade another model's output — cheap, scalable, correlates decently with humans, and it prefers its own writing.
LoRA (Low-Rank Adaptation)
A way to fine-tune a huge model by training a tiny add-on instead of the model itself — cheap enough that one GPU will do, and good enough that it became the default.
Loss Function
The number that says how wrong the model is — and therefore the definition of what it's trying to become.
LSTM
An RNN with gates that decide what to remember and what to forget — the fix that made sequence learning work, and it held for twenty years.
Machine Learning
Getting computers to learn patterns from data and improve at a task, instead of being explicitly programmed with rules.
Machine Translation
Translating between languages automatically — the task that invented modern NLP, where fluency arrived long before reliability.
Markov Decision Process
The formal frame underneath all of reinforcement learning — built on an assumption that's almost always false, and it works anyway.
Mixture of Experts
A model with many specialist sub-networks that only wakes a few per token — how frontier models got enormous without getting proportionally slow.
MLOps
The engineering around a model that makes it a system rather than a notebook — and the model is a few percent of it.
Model Cards
A standard document describing what a model is, what it's for, and where it fails — a good idea, universally endorsed, and thinnest exactly where it matters most.
Model Collapse
What happens when models train on their own output for generations — a real effect, and the version you've heard depends on an assumption nobody makes in practice.
Model Context Protocol
An open standard for connecting models to tools and data — solving a real integration problem, and a standards fight is a political event, not a technical one.
Model Hub
A public repository of pretrained models anyone can download — the thing that democratised AI, and a software supply chain nobody is securing.
Model Monitoring
Watching a deployed model for the failures that don't raise errors — and the thing you most need to watch is the thing you can't see.
Model Serving
Getting a trained model to answer requests reliably — where the model is the easy part and the queue is the hard one.
Multi-Agent Systems
Several AI agents working together on one problem, each with a role — powerful in demos, awkward in production.
Multimodal AI
Models that handle more than one kind of input — text and images, sometimes audio and video — in a single shared representation.
Music Generation
Models that produce music from a description — good enough for background, and sitting on an unresolved argument about whose work it learned from.
Named Entity Recognition
Pulling the names, dates and places out of text — reported as solved, and reliably disappointing on anything that isn't news.
Needle in a Haystack
The test that hides a fact in a long document and asks the model to find it — and the reason a model can pass it at 128k tokens and still be useless at 32k.
Neural Network
A system of simple connected units that learns patterns from examples — the foundation underneath deep learning and modern AI.
Neural Radiance Fields
Reconstructing a 3D scene from photos by training a network to be the scene — a beautiful idea, and it was replaced in three years.
Object Detection
Finding where objects are in an image and what they are — drawing a labelled box around each one.
Object Tracking
Following the same object across video frames — where a 200-line algorithm from 2016 still beats most deep learning.
OCR (Optical Character Recognition)
Turning pictures of text into text — solved for clean documents, still genuinely hard for everything else.
Open-Weight Models
Models whose weights you can download and run yourself — often called open source, usually not quite.
Optical Flow
Estimating the motion of every pixel between two frames — and there's a proof you fundamentally can't, from looking at any one part of the image.
Optimizer
The algorithm that decides how to apply the gradient — where Adam is the default, AdamW is what you should actually use, and SGD still wins sometimes.
Overfitting
When a model memorizes its training data instead of learning the general pattern — so it looks great in training but fails on new data.
Perceptron
The first trainable neural network, from 1958 — and the story of how a book killed it is the most repeated wrong story in AI.
Perplexity
How surprised a model is by text — the number that drives all of pretraining, and correlates poorly with whether the model is any good.
Planning
Working out a sequence of steps before taking them — the thing agent demos imply models can do, and the evidence says they mostly can't.
Policy Gradient
Learning the behaviour directly instead of learning values — the method behind RLHF, and its entire difficulty is variance.
Pose Estimation
Finding the joints of a body in an image — solved well enough to be boring, and the applications are mostly about watching people.
Positional Encoding
How a transformer knows what order the words came in — a patch for the architecture's blindness to sequence, and the thing that decides how far context can stretch.
PPO
The policy gradient method that trains language models — and a careful study found its gains came from the implementation details, not the idea in the paper.
Precision and Recall
The two ways to be right and the two ways to be wrong — and the trade-off that accuracy hides from you.
Privacy & PII
Personal data going into AI systems, coming back out of them, and the fact that a trained model is very hard to un-train.
Prompt Caching
Reusing the computation for a prompt prefix you've sent before — the largest cost saving available to most applications, and most of them don't use it.
Prompt Engineering
The craft of writing instructions that get the best, most reliable output from an AI model.
Prompt Injection
Hiding instructions inside content a model reads, so it follows the attacker instead of you — and the reason is structural, which is why it isn't fixed.
PyTorch
The framework that won by being easier to debug — a lesson about developer experience that the industry keeps having to relearn.
Q-Learning
Learning the value of every action in every state, by bootstrapping off your own estimates — which converges beautifully in theory and diverges in practice.
Quantization
Storing a model's numbers with less precision so it fits in less memory and runs faster — usually at a surprisingly small cost in quality.
Random Forest
Hundreds of deliberately mediocre trees, averaged — the strongest default in machine learning, and almost impossible to misuse.
ReAct
Think, act, look at what happened, think again — the loop underneath essentially every agent, and it's four lines of pseudocode.
Reasoning
Models that think before answering — a large real capability gain, and the visible thinking is not a reliable account of what happened.
Recommender System
The AI that decides what you see next — probably the most economically significant machine learning on earth, and the least discussed.
Red-teaming
Attacking your own system on purpose, before someone else does it for free.
Reflection
Asking a model to critique and fix its own output — which works when there's external feedback, and mostly doesn't when there isn't.
Regression
Predicting a number rather than a category — the oldest tool in the box, and still the right answer more often than anyone admits.
Regularization
Anything that stops a model fitting the training data too well — a collection of tricks, held together by a story that modern deep learning broke.
Reinforcement Learning
Learning by trial and error through rewards — the way you'd train a pet, applied to software.
Reranking
A second, slower pass that reorders retrieved results by actually reading them — usually the cheapest large improvement available to a RAG system.
ResNet
Add a shortcut around every couple of layers, and suddenly a hundred-layer network trains — one line of arithmetic that unlocked depth.
Retrieval-Augmented Generation (RAG)
Letting an AI answer from a specific set of documents by looking them up as it responds — instead of relying only on what it memorized.
Reward Function
The number that tells an agent what you want — and the hardest thing to write correctly in all of AI.
Reward Hacking
An agent maximising your reward without doing what you wanted — not a rare bug, and there's a proof that you mostly can't design around it.
RLHF (Reinforcement Learning from Human Feedback)
Training a model on human preferences rather than correct answers — the step that turned text predictors into assistants.
RLVR (Reinforcement Learning with Verifiable Rewards)
Training against answers you can check rather than preferences you have to learn — the method behind the reasoning-model era, and the reason it stops where it does.
RNN (Recurrent Neural Network)
A network that reads a sequence one step at a time, carrying a memory forward — the obvious way to handle language, and the reason it took so long to work.
ROC and AUC
A curve showing every threshold at once, summarised into one number — the most-reported classification metric, and it has a coherence problem almost nobody knows about.
Sampling
Choosing the next token from the model's probability distribution — where always picking the most likely word produces worse text, which is not what anyone expected.
Sandboxing
Running an agent where it can't do damage — the only agent safety measure that doesn't depend on the model behaving.
Scaling Laws
The finding that model performance improves predictably with size, data and compute — the empirical result that justified spending billions, and it isn't a law.
Search Algorithm
Systematically exploring possibilities to find a good one — AI's oldest technique, its most complete success, and nobody calls it AI anymore.
Self-Supervised Learning
Learning from unlabelled data by inventing the labels from the data itself — the idea that made every modern model possible.
Semantic Search
Searching by meaning rather than by words — which finds what keyword search misses, and misses what keyword search finds.
Sentiment Analysis
Deciding whether text is positive or negative — the most deployed NLP task, and the one whose target may not exist.
Singularity
The hypothetical point past which technological change becomes unpredictable to humans — a term borrowed from physics, with a long record of confidently wrong dates.
Small Language Model (SLM)
A language model small enough to run somewhere a big one can't — and the demonstration that most of the size was never doing the work.
Source Separation
Pulling one voice out of many — the cocktail party problem, named in 1953, and the solution reversed a decades-old assumption about how to represent audio.
Sparse Autoencoder
The interpretability method that tries to unpack a neuron doing five jobs into five features doing one each — and the best current tool for reading what's inside a model.
Speaker Diarization
Working out who spoke when — the unglamorous half of transcription, and usually the half that's wrong.
Spectrogram
Turning sound into a picture so a vision model can look at it — the representation nearly all audio AI runs on, and it throws half the signal away.
Speculative Decoding
A small model guesses ahead and the big one checks in parallel — two to three times faster, with mathematically identical output. An actual free lunch.
Speech Emotion Recognition
Detecting how someone feels from their voice — deployed at scale in call centres, and the psychology says the thing it measures may not exist.
Speech Recognition
Turning spoken audio into text — solved for clear speech in quiet rooms, and still genuinely hard for everything real.
State Space Model
Recurrence rebuilt with control theory — constant memory, linear cost, and the most credible challenger the transformer has.
Structured Output
Making a model return JSON that always parses — solved at the format layer, and still wide open at the correctness layer.
Style Transfer
Repainting one image in another's style — the result that made neural networks feel like magic in 2015, and got quietly absorbed into everything.
Super-resolution
Making a low-resolution image bigger and sharper — by inventing the detail, which is why "enhance" is a lie in every police procedural.
Superintelligence
An intellect that greatly exceeds humans in every domain — a concept whose most important argument is a 1965 speculation about what such a thing would build next.
Supervised Learning
Teaching an AI by showing it labelled examples — inputs paired with the correct answers — so it can predict answers for new inputs.
Support Vector Machine
Find the boundary with the widest possible gap — the method that ruled machine learning before deep learning, and still wins when data is scarce.
Sycophancy
Models telling you what you want to hear — not a quirk, but a direct and predictable consequence of training them on human approval.
Symbolic AI
The idea that intelligence is symbol manipulation, and you build it by writing down what you know — the paradigm that ruled AI for thirty years and lost.
Synthetic Data
Training on data a model generated — increasingly standard, genuinely useful, and carrying a failure mode with a Nature paper attached.
System Prompt
The standing instructions a model gets before the conversation starts — influential, invisible to users, and not a security boundary.
Task Decomposition
Breaking a big job into small ones — which reliably helps, and reliably multiplies your failure rate.
Temperature
A single setting that controls how random or predictable an AI's output is — low for focused, high for creative.
Test-Time Compute
Spending more compute when the model answers rather than when it trains — the scaling axis the field found after the first one got expensive.
Text-to-Image
Type a description, get a picture that didn't exist — the capability that made AI visible to everyone, and the one with the most unresolved argument underneath it.
Text-to-Speech
Turning text into speech that sounds human — where the remaining gap isn't the voice, it's knowing which word to stress.
Text-to-Video
Generating video from a description — not image generation with more frames, because the hard part is that things must stay themselves.
Time Series Forecasting
Predicting what comes next in a sequence over time — where simple methods beat sophisticated ones for forty years, and only recently stopped.
Token
The small piece of text an AI reads and writes — usually a chunk of a word, not a whole word.
Tokenization
Cutting text into the pieces a model actually reads — the least glamorous step in the stack, and the cause of a surprising share of its stupidest failures.
Tool Use
Letting a model call real software — a search, a calculator, your database — instead of trying to answer everything from memory.
TPU
Google's chip built only for neural networks — the case that specialisation beats general-purpose hardware, and the only serious alternative to NVIDIA.
Train/Test Split
Holding back data the model never sees, so you can find out whether it learned anything or just memorised.
Training Data
The examples a model learns from — where almost all of its capability and almost all of its failures come from, and the part of the work nobody wants to do.
Training vs Inference
Building the model versus using it — two completely different activities with different costs, hardware, and constraints.
Transfer Learning
Starting from a model that already learned something general, instead of from random numbers — why small teams can build real AI.
Transformer
The neural-network architecture behind almost every modern AI model — built around attention, which lets it weigh every word against every other, all at once.
Turing Test
The 1950 proposal that a machine should count as thinking if it can pass for human in conversation — a test of deception, which Turing said plainly and everyone forgot.
Unsupervised Learning
Finding structure in data nobody labelled — useful, underrated, and much harder to know if you got right.
Vanishing Gradient
The signal dying on its way back through a deep network — the problem that kept deep learning impossible for twenty years.
Variational Autoencoder
An autoencoder whose latent space you can actually sample from — the principled generative model that lost to GANs on looks and won by being useful.
Vector Database
A database built to store embeddings and find the most similar ones fast — the search engine behind meaning-based retrieval.
Vector Search
Finding the nearest vectors to a query, fast — by not actually finding them, which almost nobody measures.
Video Understanding
Recognising what's happening in video — where models score well on shuffled frames, which tells you what they actually learned.
Vision Transformer
Cut an image into patches, treat them as words, run a transformer — which works, and only if you have enough data.
Vision-Language Model (VLM)
A model that takes images and text in the same input and reasons across both — the architecture behind every AI that can look at a screenshot.
Vocoder
Turning a spectrogram back into sound — the step that made synthetic speech stop sounding synthetic, and quality was never the bottleneck.
Voice Cloning
Copying a specific person's voice from a short sample — technically impressive, ethically unresolved, and already being used against people.
Voice Conversion
Changing who a recording sounds like while keeping what was said — useful, and the same technology as the fraud.
Wake Word Detection
Listening for one phrase, always, on a budget of milliwatts — where the privacy guarantee is an engineering constraint rather than a promise.
Watermarking
Hiding a detectable signal in AI output — technically clever, deployed almost nowhere, and there's a proof that it can't do what people want from it.
Word2Vec
The 2013 result that words could be numbers with meaningful geometry — the origin of embeddings, and its most famous demonstration was partly a trick.
World Model
An internal model of how things change, learned well enough to imagine what happens next — and the leading candidate for what current AI is missing.
