Category · AI
AI vocabulary
Core terms for talking to and about AI. Models, tokens, prompting, behavior, RAG, agents, controls, safety, image generation. The vocabulary that turns hand-waving into instructions a model can act on.
Category · AI
Core terms for talking to and about AI. Models, tokens, prompting, behavior, RAG, agents, controls, safety, image generation. The vocabulary that turns hand-waving into instructions a model can act on.
The foundations. Get these and the rest of the vocabulary clicks into place.
A model trained on vast text that generates language by predicting the next token. Claude, GPT, Gemini are LLMs.
The chunks a model reads/writes, roughly ¾ of a word. Pricing and limits are counted in tokens.
The learned numbers that store what a model "knows." More isn't always better.
Layers of interconnected "neurons" that transform inputs to outputs, the underlying structure.
The architecture behind modern LLMs; its "attention" lets it weigh relationships across text.
Training = learning from data (slow, expensive, one-time); inference = using it to answer (each request).
Pre-training builds broad ability; fine-tuning adapts a model to a specific domain or style.
The date past which the model has no built-in knowledge, needs search for newer facts.
How you talk to a model. Small wording changes produce big quality changes.
The text you send the model, instruction, question, and any context.
A behind-the-scenes instruction setting the model's role, rules, and persona for the whole chat.
Zero-shot = no examples; few-shot = a few examples to show the pattern you want.
Asking the model to reason in steps before answering, improves hard/logic problems.
Telling the model who to be, which shifts its vocabulary, depth, and defaults.
How much text (prompt + history + output) the model can consider at once. Overflow gets dropped.
Asking for a specific machine-readable shape (JSON, table, XML) so output is parseable.
A reusable prompt with variable slots filled in per request.
What models can do, and the quirks to watch for.
When a model states false information fluently and confidently. Always verify facts.
Tying answers to provided/retrieved evidence so they're factual, not invented.
A model working through a problem internally before answering, better on complex tasks.
Handling more than text, images, audio, video, as input and/or output.
Shaping a model to behave according to human intentions and values.
Capabilities that show up in big models but not small ones, not explicitly programmed.
The same prompt can give different outputs each time (unless temperature is 0).
How fast a model responds and how many tokens it streams per second.
How models get information beyond their training, the basis of most real AI apps.
Fetching relevant documents and feeding them to the model so answers use your data.
A list of numbers representing text's meaning, so similar ideas sit close together.
A store optimized to find the embeddings closest to a query (Pinecone, pgvector, Weaviate).
Finding results by meaning rather than exact words, "car" matches "automobile."
Splitting long documents into smaller pieces that fit the context and retrieve precisely.
Storing facts/preferences so the AI recalls them in future conversations.
Deliberately assembling the right instructions, examples, and data into the prompt.
A second pass that re-scores retrieved results so the most relevant reach the model.
When a model doesn't just talk, but acts, calling tools and taking steps toward a goal.
An AI that loops, plans, takes actions, checks results, to complete a multi-step goal autonomously.
Letting a model call external functions/APIs, search, code, databases, to do real work.
An open standard for connecting AI to tools and data sources (Slack, GitHub, files) in a uniform way.
Coordinating multiple steps, tools, or sub-agents into a reliable workflow.
A separate agent spun up to handle a focused subtask (research, verification) in parallel.
How much an agent does alone vs pausing for human approval at key steps.
A workflow follows predefined steps; an agent decides its own steps dynamically.
Constraints (allowed tools, scopes, checks) that keep an agent safe and on-task.
The knobs that tune how a model generates. Naming these gives you fine control.
Randomness. Low = focused/repeatable; high = varied/creative.
Limits choices to the most probable tokens summing to p, another diversity control.
A ceiling on how long the response can be.
Text that, when generated, halts the output, useful for clean formatting.
Sending the response token-by-token as it's produced, so the user sees it type out.
A fixed starting value that makes random generation repeatable.
Measuring whether AI works, and keeping it from misbehaving.
Tests that measure model quality on a task. Build your own to know if changes help.
Systematic skew inherited from training data that can produce unfair outputs.
Hidden instructions in input data that trick a model into ignoring its real task.
Crafted prompts that try to bypass a model's safety guidelines.
Deliberately probing a system for failures and harms before users find them.
Checks that block unsafe inputs/outputs around the model.
Training a model using human ratings of its responses to align behavior.
Whether a model's stated certainty matches how often it's actually right.
The vocabulary for generating images, audio, and video.
Generates images by starting from noise and progressively refining (Stable Diffusion, Midjourney, DALL·E).
Creating an image from a text description (the prompt).
Emphasizing or de-emphasizing parts of an image prompt with weights.
Telling the model what to avoid in the image.
Fixes the starting randomness so you can reproduce or vary an image consistently.
Regenerating a masked region (inpaint) or extending the canvas (outpaint).
Increasing an image's resolution while adding plausible detail.
Steering generation with an extra input, a sketch, pose, or depth map, for precise control.