Agent
AI software capable of acting autonomously to achieve a goal. It doesn't just respond to you (like ChatGPT) — it can 'do things': send emails, book flights, or browse the web on your behalf.
The Dictionary of
Artificial Intelligence
Decode the language of machines. Clear, curated definitions with no unnecessary jargon, for professionals.
94 terms
A
AGI (Artificial General Intelligence)
The 'Holy Grail' of AI. It's a hypothetical future AI that matches or surpasses human intelligence at any task, not just a specific one.
AI Act
The world's first comprehensive law on artificial intelligence, passed by the European Union, which classifies AI systems by risk level.
AI Ecosystem
The collection of hardware, software, data, researchers, and companies that allow AI to exist and develop.
AI Ethics
The branch that studies the moral implications of AI, such as privacy, bias, and the impact on jobs.
AI Hardware
Specific chips (NPUs, TPUs) designed exclusively to process artificial intelligence workloads.
Algorithm
The 'recipe'. It's a series of step-by-step instructions the computer follows to solve a problem. In AI, the algorithm is the method by which the machine learns from data.
Algorithmic Bias
Systematic errors in AI outputs caused by prejudices present in the training data (e.g. gender or racial stereotypes).
Alignment
The process of ensuring an AI's goals stay aligned with human values and intentions, to avoid unintended or harmful consequences.
Artificial Intelligence
A branch of computer science concerned with creating intelligent systems.
B
Benchmark
A standardized test used to measure how 'intelligent' or high-performing an AI model is compared to others.
Bias
Prejudices present in the training data that the model absorbs and can replicate (e.g. gender or racial stereotypes).
Big Data
Data sets so large and complex that traditional software can't handle them. They're the 'food' AI feeds on to learn.
Black Box
Refers to the fact that we often don't know why an AI made a certain decision. The internal reasoning is opaque even to its creators.
Bot
Short for robot. An automated program that performs repetitive tasks on the internet.
C
Canvas
An interface that pairs the chat with a persistent editor for collaborating on text or code without constant copy-pasting.
Chain of Thought
A prompting technique where you ask the AI to 'think out loud' or show its logical steps before giving the final answer.
Chatbot
A program designed to simulate a conversation with humans; it can be simple or based on advanced LLMs.
Computer Vision
The field of AI that teaches computers to 'see' and understand images and video (e.g. facial recognition).
Context Window
The AI's 'short-term memory' during a conversation. It indicates how much text the AI can remember and analyze at once.
Copilot
An AI assistant that works alongside you (writing code, emails, or documents), while leaving you in final control.
D
Data Mining
The process of digging through large amounts of data to uncover hidden patterns, trends, and useful information.
Dataset
The ordered collection of data (text, images, numbers) used to instruct the AI.
Deep Learning
An advanced Machine Learning technique inspired by the structure of the human brain (deep neural networks).
Deepfake
Videos, images, or audio created by AI that look real but are fake, often used to impersonate famous people.
Diffusion Model
The technology behind image generators like Midjourney. It creates images starting from 'noise' and cleaning it up until the requested subject takes shape.
E
Embeddings
The translation of words or images into lists of numbers (vectors) that let the computer understand the semantic closeness between concepts.
Epoch
One complete training cycle in which the AI has seen the entire dataset once.
Explainable AI (XAI)
An approach to building AI that can explain its decisions in a way humans can understand.
F
False Positive/Negative
Classification errors. False Positive: flags a problem that isn't there. False Negative: misses a problem that is.
Feedback Loop
The cycle in which the AI acts, receives feedback on the result, and uses that information to improve.
Few-shot Learning
A model's ability to learn a new task by seeing only a handful of examples.
Fine-tuning
Specializing a general-purpose model through further training on data specific to an industry.
Foundation Model
A huge, general-purpose model trained on massive amounts of data, on top of which specific applications are built.
G
GAN (Generative Adversarial Network)
Two neural networks competing against each other (one creates fakes, the other tries to spot them) to generate highly realistic data.
GenAI (Generative AI)
AI that creates new content (text, audio, video, code) instead of just analyzing existing content.
GPT
Stands for Generative Pre-trained Transformer. It refers to a generative model, pre-trained, based on the Transformer architecture.
GPU (Graphics Processing Unit)
Processors originally built for graphics but essential for AI thanks to their parallel computing power.
Grounding
The process of anchoring the AI's answers to real, verifiable data to prevent hallucinations.
H
Hallucination
When the AI confidently states things that are completely made up or false.
Hugging Face
A key platform where the community shares open-source models and datasets (the 'GitHub of AI').
Human-in-the-loop
A system where the AI does the work, but a human reviews or approves the final result.
Hyperparameter
The settings engineers configure before training to decide how the model should learn.
I
Inference
The moment when the AI 'works' to answer a question using what it learned during training.
Input
Everything provided to the AI: prompts, images, files, or data.
Instruction Tuning
A training phase where the AI is taught to follow specific instructions, beyond just predicting text.
Interpretability
The ability to understand the logical cause behind a decision made by an AI model.
L
Labeling
The human work of classifying data (e.g. 'this is a cat') so the AI can learn from it.
Latency
The wait time between sending a request and the AI starting its response.
Latent Space
A multidimensional mathematical representation where the AI organizes concepts based on their similarity.
LLM (Large Language Model)
Huge language models trained on vast chunks of the internet — the foundation of chatbots like ChatGPT.
Loss Function
The mathematical value that tells the AI how wrong it was during training; the goal is to minimize it.
M
Machine Learning
The field of computer science that teaches computers to learn from data without being explicitly programmed with fixed rules.
Machine Translation
Advanced automatic translation that uses neural networks to understand context and linguistic nuance.
Metadata
Additional information about data (date, location, author) that helps the AI organize knowledge.
Model
The final software file that contains the 'intelligence' learned after training.
Multimodal
An AI capable of understanding and generating different types of media at once: text, images, audio, and video.
N
Narrow AI (ANI)
Today's AI — extremely capable at specific tasks but without consciousness or general intelligence.
Natural Language Processing (NLP)
The branch of AI concerned with helping computers understand human language.
Neural Network
A layered software system that loosely mimics how neurons work in the human brain.
NLG (Natural Language Generation)
The technology that lets software write text that sounds like it was written by a human.
Noise
Useless or corrupted data that can confuse the AI during training.
NPU
Neural Processing Unit, a processor dedicated to speeding up AI operations on devices.
O
OCR
Technology that reads text inside images or scanned documents, making it editable.
One-shot Learning
When a model learns to do something after seeing a single example.
Open Source AI
AI models whose code and structure are public and freely usable or modifiable.
Optimization
The process of making a model more efficient, faster, or lighter.
Overfitting
When the AI memorizes the data instead of learning from it, and fails to generalize to new examples.
P
Parameters
The model's internal variables; their number roughly indicates the AI's complexity and capability.
Pre-training
The first massive phase of training on generic data, before specialization.
Predictive AI
AI used to forecast future events based on historical data (e.g. sales forecasts).
Prompt
The text or command given to the AI to get a response.
Prompt Engineering
The skill of crafting precise, effective prompts to guide the AI toward the best possible result.
R
RAG (Retrieval-Augmented Generation)
A technique that lets the AI consult specific external documents before answering, improving accuracy.
Reasoning
The ability of advanced models to break complex problems down into logical steps.
Reinforcement Learning
A learning method based on rewards and penalties to steer the AI toward its goal.
RLHF
Reinforcement Learning from Human Feedback: training based on human evaluations to make the AI more helpful and safe.
Robotics
The application of artificial intelligence to physical machines in the real world.
S
Scalability
The ability of an AI system to handle growing workloads without losing performance.
Semantic Search
Search based on meaning and intent, not just exact keywords.
Sentiment Analysis
The use of AI to identify the emotional tone (positive, negative, neutral) in a text.
Singularity
The hypothetical moment when AI becomes capable of self-improvement beyond human control.
Supervised Learning
Training where the AI learns from pairs of inputs and correct answers provided by humans.
Synthetic Data
Data artificially generated by algorithms to make up for a lack of real data or for privacy reasons.
T
Temperature
A parameter that controls the AI's creativity: low is precise and deterministic, high is creative and varied.
Text-to-Image / Video
Models that generate images or videos starting from a text description.
Token
The AI's fundamental unit of text; it corresponds to fragments of words.
Training Data
The set of information used to train the model.
Transformer
The neural network architecture that revolutionized AI by enabling efficient handling of language and context.
U
Unsupervised Learning
Learning in which the AI independently finds patterns in data without labels or human guidance.
V
Validation
A testing phase during training to verify that the AI is learning correctly on new data.
Vector Database
A database optimized to store vectors (mathematical representations of concepts) and run fast semantic searches.
Vision Transformer (ViT)
The application of the Transformer architecture to image analysis.
Z
Zero-shot Learning
The AI's ability to perform a task without having seen specific examples of it beforehand.
No terms found
Try adjusting your search term or selecting another letter.
Keep exploring
You might also like
Guides, courses and tools picked for you, starting from this page.
Guide
Google Search Console inside Claude Code
Search Console data inside your terminal.
Course
Deploy From Scratch: From Code to Link
Deploy on Vercel with no jargon: turn the project sitting on GitHub into a public link you can send anyone, then add your own domain on top.
Asset
+7 Clear Brief
The sheet to bring to the call
Guide
From a Confusing Brief to a Complete UX/UI
A vague brief becomes a navigable app.
Course
Git and GitHub for Vibe Coders
Built a project with Claude Code or Codex? Learn to put it on GitHub, roll back changes when things break, and bring other people in to help.
Asset
+3 Design Skills
AI skills for designers