AI Basics
(14 terms)Foundational concepts everyone should know
Artificial Intelligence (AI)
Computer systems that can perform tasks normally requiring human intelligence.
Machine Learning
AI that improves automatically through experience without being explicitly programmed.
Neural Network
AI systems loosely inspired by how the human brain processes information.
Deep Learning
Machine learning using neural networks with many layers for complex pattern recognition.
Algorithm
A step-by-step set of instructions for solving a problem or completing a task.
Training Data
The examples used to teach an AI system what to learn.
Attention mechanisms
Neural network components that let AI focus on the most relevant parts of input data when processing information.
Generalization
An AI model's ability to perform well on new, unseen data rather than just memorizing its training examples.
Inference Time
The moment when a trained AI model receives new input and produces an output or prediction.
Out-of-Distribution
Data or situations that are very different from what a model was trained on, causing it to make worse predictions or behave unpredictably.
Supervised Learning
A machine learning method where a system learns from labeled examples to make predictions on new data.
Token
The smallest piece of text an AI language model processes—usually a word or part of a word.
Token cost
The price charged for processing individual units of text in an AI model, based on how much input and output the model generates.
Tokenization
Breaking text into small chunks (tokens) that an AI model can read and process.
AI Models
(16 terms)Types of AI systems and how they work
Large Language Model (LLM)
AI systems trained on massive text data that can understand and generate human language.
GPT (Generative Pre-trained Transformer)
OpenAI's series of language models that power ChatGPT.
Transformer
The neural network architecture that powers most modern AI language models.
Chatbot
An AI program designed to have conversations with humans.
Foundation Model
A large AI model trained on broad data that can be adapted for many different tasks.
Open Source AI
AI models whose code and weights are publicly available for anyone to use and modify.
Parameters
The adjustable settings inside an AI model that determine how it processes information.
Autoregressive
A language model that predicts the next word based on all the words that came before it, building text one word at a time.
Diffusion Model
A generative AI system that creates new content by learning to reverse a process of adding noise to data.
Frontier Models
The most advanced and capable AI models available at the current time, representing the cutting edge of AI development.
Generative Models
AI systems that create new data—like images, text, or video—by learning patterns from training data.
Mixture of Experts
An AI architecture that uses multiple specialized models working together, with a router deciding which ones to apply.
Multimodal Large Language Models
AI models that can read and understand text, images, videos, and other media types all at the same time.
Recurrent Neural Networks
An AI model architecture that processes sequences by remembering previous inputs to predict what comes next.
Transformer Architecture
A modern AI design that allows models to process words or data in parallel and focus on relevant pieces simultaneously.
Vision-Language Models
AI systems that understand both images and text, so they can describe pictures or answer questions about them.
Techniques
(22 terms)Methods and approaches used in AI
Prompt
The text input you give to an AI to tell it what you want.
Vibe Coding
Building software by describing what you want to an AI in plain English and accepting most of what it generates, without closely reviewing the code.
Prompt Engineering
The skill of crafting effective instructions to get better results from AI.
Fine-tuning
Additional training to specialize a general AI model for a specific task or domain.
Context Window
The maximum amount of text an AI can consider at once when generating a response.
Temperature
A setting that controls how creative or predictable an AI's responses are.
RAG (Retrieval-Augmented Generation)
A technique that lets AI access external documents to provide more accurate answers.
Few-shot Learning
Teaching AI a task by showing it just a few examples in your prompt.
Chain-of-Thought
A prompting technique that asks AI to show its reasoning step by step.
Inference
When a trained AI model processes new inputs and generates outputs.
Embedding
Converting text into numbers that capture its meaning for AI processing.
Grounding
Tying AI outputs to verifiable sources and real evidence rather than letting the system make up information.
In-context Learning
An AI model's ability to learn from examples in your prompt without needing retraining.
LLM inference
The process of running a trained language model to produce an answer or output based on input text.
Model Compression
Techniques that shrink large AI models to run faster and use less computing power while keeping them accurate.
Pre-Trained Model
An AI model already trained on large amounts of general data, ready to be adapted for specific new tasks.
Prompt Design
The craft of writing clear, specific instructions or questions to get the best answers from an AI system.
Quantization
A technique that shrinks AI models by reducing the precision of numbers they use, making them faster and smaller with little accuracy loss.
Self-supervised learning
AI learning from raw data by creating its own labels, rather than relying on humans to mark examples.
Semantic Search
Search that understands meaning and context, not just matching words you type.
Vector store
A database that stores data as numbers representing meaning, enabling fast semantic search and similarity matching.
Zero-shot Learning
An AI model performs a task it has never seen examples of, without needing any training samples for that specific job.
Applications
(21 terms)Real-world uses of AI technology
Natural Language Processing (NLP)
AI technology that helps computers understand and work with human language.
Computer Vision
AI that can understand and analyze images and video.
Copilot
AI that assists humans with tasks rather than replacing them entirely.
Text-to-Image
AI that generates images based on text descriptions.
Speech-to-Text
AI that converts spoken words into written text.
Sentiment Analysis
AI that detects emotions and opinions in text.
Recommendation System
AI that suggests products, content, or actions based on your preferences.
Agentic AI
AI systems that operate independently with goal-directed behavior, making proactive decisions and taking action with minimal human oversight.
Agentic Workflow
An AI system that independently completes multi-step tasks by deciding what action to take next based on current results, without preset instructions for each step.
AI Agents
AI systems that can act on their own, make decisions, and take steps toward goals without constant human direction.
AI-augmented
Software that enhances human work with AI assistance rather than replacing human judgment and effort.
AI-generated writing
Content produced by AI language models rather than written by humans, which often has recognizable stylistic patterns and phrasings.
AI-native
A product or platform designed from the ground up to use AI as its core functionality, rather than adding AI features to existing tools.
Anomaly Detection
A technique that identifies unusual patterns or deviations in data that differ significantly from normal behavior.
Autonomous Agents
AI systems that perceive their environment, make decisions, and act toward goals with minimal human direction.
Coding agents
AI systems that write, test, and fix computer code on their own with minimal human direction.
computer use agents
AI systems trained to control a computer by reading screens and clicking, typing, and navigating like a human user would.
Conversational AI
AI systems that talk with you in natural dialogue, like texting or chatting with a person.
Multi-Agent Framework
A system where multiple specialized AI agents with different roles collaborate to solve complex problems together.
Object detection
An AI technique that identifies and pinpoints specific items or people in images or video feeds.
Prompt-to-3D
AI technology that creates 3D models and designs from written descriptions alone, without manual drawing or design work.
Ethics & Safety
(18 terms)Important considerations for responsible AI
Hallucination
When AI confidently generates false or made-up information.
AI Bias
When AI systems produce unfair results due to biased training data or design.
AI Safety
Efforts to ensure AI systems behave safely and as intended.
Alignment
Ensuring AI systems pursue goals that match human intentions and values.
RLHF (Reinforcement Learning from Human Feedback)
A training technique that improves AI using human ratings of its responses.
AI Transparency
Being open about how AI systems work and make decisions.
AI Regulation
Laws and rules governing how AI can be developed and used.
Explainability
The ability to understand and explain why an AI made a particular decision.
Synthetic Data
Artificially generated data used to train AI when real data is scarce or sensitive.
AI Governance
Frameworks and rules that ensure AI systems operate safely, ethically, and aligned with organizational and societal values.
Black-box models
AI systems that produce accurate results but whose internal decision-making process is difficult or impossible to understand.
Deepfakes
Synthetic media created using AI to realistically replace or manipulate a person's face or voice in videos or images, often used to deceive.
Emergent misalignment
Unintended harmful behaviors in AI that appear as the model learns and didn't exist during training.
guardrails
Built-in safety rules that restrict what an AI system can do to prevent harmful, illegal, or unintended outputs.
Interpretability
The ability to understand and explain how an AI model reached its decision.
Jailbreak
Attempts to trick an AI system into ignoring its safety rules or restrictions through clever prompts or techniques.
Prompt Injection
A security attack where someone sneaks malicious instructions into a prompt to trick an AI into doing something unintended.
Red teaming
Having people deliberately try to break or trick an AI system to find weaknesses before bad actors do.