Artificial Intelligence (AI) noun
The simulation of human intelligence in machines programmed to think, learn, and make decisions.
A customer service chatbot uses AI to understand queries and suggest solutions without human intervention.
Machine Learning noun
A subset of AI where systems learn from data to improve performance on a task without explicit programming.
A recommendation engine learns from user clicks to suggest relevant products, improving over time.
Deep Learning noun
A subset of ML using multi‑layered neural networks to model complex patterns in large datasets.
Deep learning powers facial recognition in smartphones, identifying the user even with glasses.
Neural Network noun
A computing system inspired by biological neurons, consisting of layers of interconnected nodes that process data.
A neural network with 50 layers (deep neural net) analyzes medical images to detect tumors.
Large Language Model (LLM) noun
A transformer‑based model trained on massive text data to generate, summarize, and understand human language.
GPT‑4o, Claude 3.5, and Gemini 1.5 are LLMs used for writing, coding, and analysis.
Transformer Architecture noun
A neural network design that uses self‑attention to process all tokens in parallel, enabling modern LLMs.
The “T” in GPT stands for Transformer; it allows the model to weigh the importance of each word in context.
Attention Mechanism noun
A technique that lets a model focus on relevant parts of the input when generating output, mimicking human attention.
In translation, attention helps the model focus on “chat” when generating the French “chat” from “the black cat.”
Token noun
A unit of text (word, subword, or character) that an LLM processes; pricing and context limits are often token‑based.
“I love AI” is 3 tokens in most tokenizers; the model generates one token at a time.
Context Window noun
The maximum number of tokens an LLM can consider at once; larger windows allow processing of entire books.
Gemini 1.5 has a 2M token context window, enough to analyze all three books of The Lord of the Rings at once.
Prompt Engineering noun
The practice of designing input prompts to elicit desired outputs from LLMs, often using techniques like few‑shot or chain‑of‑thought.
A marketer crafts a prompt: “Write a 200‑word product description for a eco‑water bottle in a friendly tone, include three benefits.”
Few-Shot / Zero-Shot adj
Zero‑shot: model performs a task without examples. Few‑shot: model is given 2‑5 examples in the prompt to guide output.
Zero‑shot: “Translate to Spanish: Hello.” Few‑shot: Provide two English‑Spanish examples, then the target sentence.
Chain of Thought noun
A prompting technique that encourages the model to reason step‑by‑step, improving accuracy on complex tasks.
“If a store has 120 apples and sells 15 per day, how many days until sold out? Let’s think step by step.”
RAG noun
A framework that retrieves relevant information from an external knowledge base and feeds it to an LLM to generate grounded, up‑to‑date answers.
A customer support bot first searches a vector database of product manuals, then uses an LLM to summarize the answer.
Fine-Tuning noun
Further training a pre‑trained model on a specific dataset to adapt it for specialized tasks.
A legal tech company fine‑tunes Llama 3 on 10,000 court documents to create an AI that drafts legal briefs.
RLHF noun
A training method where human feedback is used to reward desirable model outputs, aligning AI with human preferences.
ChatGPT was fine‑tuned using RLHF to make responses more helpful and less harmful.
Hallucination noun
When an AI generates confident but factually incorrect or nonsensical information.
An LLM might invent a citation for a non‑existent research paper – a hallucination.
Temperature noun
A parameter controlling randomness in LLM outputs: low temp = more deterministic, high temp = more creative/varied.
For code generation, use temperature 0.2; for creative story writing, use 0.8.
Top-p (nucleus) sampling noun
Decoding strategies where the model considers only the top p probability mass (or top k tokens) when choosing the next token.
Top‑p=0.9 means the model selects from the smallest set of tokens whose cumulative probability exceeds 90%.
Embeddings noun
Numerical vector representations of data (text, images) that capture semantic meaning, used for similarity search.
Text embeddings allow semantic search: “dog” and “puppy” have similar vectors, so searching “puppy food” returns dog food products.
Vector Database noun
A database optimized for storing and querying high‑dimensional vectors, enabling fast similarity search.
Pinecone and Weaviate store embeddings of millions of documents for RAG applications.
Semantic Search noun
Search that understands the meaning and intent behind a query, rather than just keyword matching.
Searching “treats for senior dogs” returns relevant products even if the product description doesn't contain “senior.”
Cosine Similarity noun
A metric measuring the cosine of the angle between two vectors; values near 1 indicate high similarity.
In a vector DB, you find the top‑k most similar items by highest cosine similarity.
Computer Vision noun
A field of AI that enables machines to interpret and make decisions based on visual data (images, video).
Self‑driving cars use computer vision to detect pedestrians, lanes, and traffic signs.
Object Detection noun
A computer vision task that identifies and locates objects within an image, often with bounding boxes.
Retail inventory cameras use object detection to count products on shelves automatically.
OCR noun
Technology that converts images of text (scanned documents, photos) into machine‑readable text.
A mobile banking app uses OCR to capture check details when a user takes a photo.
Speech Recognition noun
Technology that converts spoken language into text; also known as Automatic Speech Recognition (ASR).
Zoom’s live transcription uses speech recognition to display captions during meetings.
TTS noun
Synthetic generation of human‑like speech from text, used in audiobooks, voice assistants, and accessibility tools.
ElevenLabs TTS creates realistic voiceovers for YouTube videos from a script.
Sentiment Analysis noun
NLP technique to determine the emotional tone (positive, negative, neutral) in text.
A brand monitors Twitter mentions with sentiment analysis to gauge reaction to a new product launch.
NER noun
A subtask of NLP that identifies and classifies entities (people, organizations, locations) in text.
A news aggregator uses NER to tag articles with the companies and politicians mentioned.
Summarization noun
Automatically condensing a longer text into a concise summary while preserving key information.
A lawyer uses an LLM to summarize 200‑page deposition transcripts into a 2‑page brief.
Machine Translation noun
Automated translation of text or speech from one language to another using AI.
DeepL and Google Translate use neural networks to provide fluent cross‑language communication.
Multimodal AI noun
Models that process and generate multiple data types (text, image, audio, video) simultaneously.
Gemini 1.5 can analyze a video, understand its audio, and answer questions about the content.
Generative AI noun
AI that creates new content (text, images, music) by learning patterns from training data.
Midjourney generates photorealistic images from text prompts; ChatGPT writes blog posts.
Diffusion Model noun
A generative model that learns to reverse a noising process, creating high‑quality images from random noise.
Stable Diffusion and DALL‑E 3 are diffusion models that turn text prompts into images.
Agentic AI / AI Agent noun
Autonomous AI systems that can pursue complex goals, use tools, and make decisions without step‑by‑step instructions.
An AI agent books a flight: it searches airline sites, checks dates, enters payment, all without human intervention.
Multi‑Agent System noun
A system where multiple AI agents collaborate or compete to achieve goals, often used in simulations or complex automation.
In a supply chain simulation, agents represent suppliers, warehouses, and retailers to optimize inventory.
Orchestration noun
Managing and coordinating multiple AI agents, APIs, and workflows to execute a complex task.
LangChain orchestrates a sequence: retrieve docs → summarize → translate → email report.
Workflow Automation noun
Using software to automate business processes, often integrating AI to handle decisions and data extraction.
Zapier automates: when a new email arrives (trigger), save attachment to Google Drive, then send Slack notification.
RPA noun
Automation of repetitive, rule‑based tasks by software “robots” that mimic human interactions with applications.
An RPA bot logs into an accounting system, downloads invoices, and enters them into Excel daily.
No‑Code / Low‑Code adj
Development platforms that allow users to build applications via visual interfaces with minimal or no programming.
Bubble.io lets a non‑technical founder build a marketplace app without writing code.
API noun
A set of protocols that allows different software applications to communicate and share data.
A weather app calls the OpenWeatherMap API to get the forecast for your location.
Webhook noun
An HTTP callback that sends real‑time data from one application to another when an event occurs.
When a payment succeeds, Stripe sends a webhook to your server to trigger order fulfillment.
Trigger / Action noun
In automation, a trigger is an event that starts a workflow; an action is the task performed in response.
Trigger: new row in Google Sheets → Action: send personalized Gmail message.
Predictive Analytics noun
Using historical data and ML to forecast future events, such as customer churn or inventory demand.
A retailer predicts which products will be best‑sellers next month to optimize stock levels.
Anomaly Detection noun
Identifying rare events or outliers in data that may indicate fraud, system failures, or significant changes.
A bank uses anomaly detection to flag unusual credit card transactions for fraud review.
Recommendation Engine noun
An AI system that suggests products, content, or services to users based on behavior, preferences, or similarity.
Netflix's recommendation engine suggests movies you might like based on your viewing history.
Personalization noun
Tailoring experiences, content, or offers to individual users using AI and data.
Amazon shows personalized homepage banners based on your browsing and purchase history.
GAN noun
Two neural networks (generator and discriminator) contest to create realistic synthetic data; used for deepfakes and art.
A fashion brand uses GANs to generate new clothing designs for inspiration.
Autoencoder noun
A neural network that learns efficient data representations (encoding) for tasks like denoising or dimensionality reduction.
An autoencoder removes noise from old photos, reconstructing a clean version.
Alignment noun
The challenge of ensuring AI systems act in accordance with human values and intentions.
RLHF is a technique to improve alignment by training models based on human preferences.