Alphabet Soup
This branch of AI teaches computers to improve at tasks over time by learning from data, rather than being explicitly programmed.
What is machine learning (ML)?
This is the input you give an AI model—whether a question, instruction, or example—to guide the response you get back.
What is a prompt?
Modeled loosely after the human brain, this structure of nodes and connections is the foundation of most modern AI.
What is a neural network?
This type of neural network scans images using layers of filters, making it especially good at spotting cats, dogs, or stop signs.
What are convolutional neural networks (CNNs)?
From customer service helpers to virtual assistants, this type of AI is designed to hold human-like conversations.
What is a chatbot?
From images to music to text, this kind of AI creates new content instead of just analyzing existing data.
What is generative AI?
AI models break text into these small chunks, which can be whole words, pieces of words, or even punctuation.
What are tokens?
Introduced by Google in 2017, this architecture uses self-attention to weigh relationships between all words in a sequence at once, enabling powerful models like GPT.
What is a transformer?
This technique lets you apply the artistic style of one image—like Van Gogh’s Starry Night—to another, like your selfie.
What is style transfer?
This technique improves AI answers by pulling in external information, so the model doesn’t have to rely only on what it was trained on.
What is retrieval-augmented generation (RAG)?
This field of AI focuses on enabling computers to understand, interpret, and respond to human language.
What is natural language processing (NLP)?
This term refers to the maximum number of tokens a model can handle in one conversation before it starts forgetting.
What is a context window?
This approach lets a model perform new tasks with little or no training examples, often relying only on clever prompting.
What is zero-shot learning?
In this framework, a generator and discriminator play a game of cat and mouse, producing hyper-realistic fake images.
What are generative adversarial networks (GANs)?
This emerging standard lets AI models securely connect to external tools, APIs, and data sources, giving them broader capabilities.
What is the Model Context Protocol (MCP)?
GPT-4 and Claude are examples of this kind of AI model, trained on massive amounts of text to generate human-like responses.
What is a large language model (LLM)?
This mathematical representation captures the meaning of words or text as numbers in a vector space, allowing AI to measure similarity.
What is an embedding?
Instead of training from scratch, this process adapts a pretrained model to perform better on a specific task or dataset
What is fine-tuning?
This popular AI system creates images from text prompts by gradually turning random noise into a coherent picture.
What is Stable Diffusion?
These AI systems can take actions toward goals on their own, chaining together reasoning and tools without step-by-step human instructions.
What is an agent?
Considered a future milestone, this type of AI would match or surpass humans at nearly any intellectual task.
What is artificial general intelligence (AGI)?
These millions—or even billions—of adjustable values in a neural network are learned during training and determine how the model behaves.
What are parameters (or weights)?
These specialized databases store embeddings and make similarity search possible, powering retrieval in systems like RAG.
What is a vector database?
Adapted from language tasks, these models split an image into small patches, embed them like tokens, and use self-attention to learn relationships across the whole picture.
What are transformer-based models?
This training method combines trial-and-error reinforcement learning with human preferences, helping align AI responses to what people actually want.
What is reinforcement learning with human feedback (RLHF)?