What categories of AI terms are defined in the ZealVector AI glossary?

The ZealVector AI glossary defines terms covering machine learning foundations, large language models, generative AI, AI agents, retrieval, evaluation, hardware, and AI governance.

Who is the ZealVector AI glossary for?

Product managers evaluating AI vendors, engineers integrating AI APIs, founders shipping AI features, journalists covering AI, and anyone building AI literacy who wants plain-English artificial intelligence definitions.

AI Glossary

Q: What AI terms should beginners learn first?

Beginners should start with AI, large language model (LLM), generative AI, prompt, token, hallucination, AI agent, and RAG. Those eight AI terms unlock most product demos and AI news headlines.

At a glance: A free AI glossary with artificial intelligence terms from A to Z — each AI definition written in plain English. Use it to follow AI model releases, AI agent launches, generative AI product news, AI policy fights, and research papers without drowning in jargon. Same no-noise standard as our weekly AI news digest.

AI news is full of words that get used more often than they get explained. This AI glossary defines the artificial intelligence vocabulary that actually comes up in 2025–2026: AI agents and agentic AI, large language models (LLMs), generative AI, RAG, reasoning models, multimodal AI, prompt engineering, fine-tuning, AI safety, and hallucination. Every AI term links to related concepts, and every definition has a stable anchor you can share or cite.

Whether you are building AI products, evaluating AI vendors, leveling up AI literacy on your team, or just trying to read the week's AI headlines, start here. Search below, jump by letter, or browse the full AI terminology list. If you need help putting AI into production, our custom AI solutions page shows how models, retrieval, agents, and evals fit together in real systems.

A ↑ top

Agentic AI: An umbrella term for AI systems that don't just answer a single question but pursue goals over multiple steps — planning, calling tools, checking their own work, and adjusting course. Where a classic chatbot waits for your next message, agentic AI decides what to do next on its own. The shift from single-turn chat to agentic workflows has been one of the defining AI trends since 2024.
AI Agent: A program built around a model — usually a large language model — that can perceive its environment, make decisions, and take actions toward a goal with limited human supervision: researching a topic, fixing a bug, booking a meeting. Agents typically combine a model with tool use, memory, and a loop that keeps them working until the task is done or a limit is reached.
AI Alignment: The research problem of making AI systems reliably pursue the goals their operators and society actually intend, rather than a literal-but-wrong reading of their instructions. Alignment work ranges from training models to refuse harmful requests today to long-term questions about steering systems smarter than their designers. Closely related to AI safety and RLHF.
AI Copilot: A product pattern where an AI assistant works beside a human inside an existing tool — a code editor, spreadsheet, CRM, or design app — suggesting, drafting, and completing work rather than replacing the user. AI copilots are a common form of generative AI productization: the model is embedded in the workflow, not a separate chatbot tab. Closely related to AI agents, but copilots usually stay under continuous human control.
AI Governance: The policies, processes, and accountability structures organizations use to develop, deploy, and oversee AI systems responsibly. AI governance covers risk tiers, approval gates, model inventory, documentation, and who signs off when AI touches customers or regulated data. It sits next to AI safety and responsible AI: safety is largely technical; governance is how the organization actually runs AI.
AI Literacy: The practical ability to understand what AI can and cannot do, how to use large language models productively, and how to spot risks like hallucination, bias, and over-trust. AI literacy is now a workplace skill — not just for engineers — because AI tools are appearing in every knowledge job. This AI glossary is one building block of AI literacy: shared definitions so teams can talk about AI clearly.
AI Observability: Instrumenting production AI systems so you can see what models did, why, and at what cost: prompts, retrieved docs, tool calls, latency, token use, refusals, and user feedback. AI observability is the operations counterpart to evals — evals check quality offline; observability watches AI behavior live. Essential once AI moves from demos into products.
AI Safety: The broader field concerned with preventing harm from AI systems. It spans technical work like alignment, evals, and red teaming, plus governance, policy, and deployment practices. "Safety" covers both near-term issues — bias, misuse, hallucinations — and longer-term concerns about highly capable systems.
Artificial General Intelligence (AGI): A hypothetical AI system able to match or exceed capable humans across most economically valuable cognitive work, rather than excelling at one narrow task. There is no agreed test for having reached AGI, which is why lab leaders can disagree in the same week about whether it is imminent or distant. Beyond it lies superintelligence.
Artificial Intelligence (AI): The umbrella discipline of building computer systems that perform tasks normally requiring human intelligence: understanding language, recognizing images, reasoning, planning, and learning from experience. Modern AI is dominated by machine learning, and in particular deep learning.
Attention Mechanism: The technique at the heart of the transformer architecture that lets a model weigh which parts of the input matter most to each other part — for example, which earlier words in a sentence a given word refers to. Self-attention is the reason modern models handle long-range context far better than earlier architectures did.
Autoregressive Model: A model that generates output one unit at a time, each new token conditioned on everything already produced. Nearly all modern large language models are autoregressive: they predict the next token, append it, and repeat. That sequential nature is why AI text streams token by token and why long generations cost more inference compute.

B ↑ top

Backpropagation: The algorithm that teaches most neural networks: compute the error of a prediction, then push that error backward through the layers to update parameters (weights). Backpropagation plus gradient descent is the workhorse of deep learning and of training frontier AI models. When people say a model is "learning," this loop is usually what they mean.
Batch Size: How many examples a model processes together in one training or inference step. Larger batch sizes can use GPUs more efficiently but need more memory; smaller batches can generalize differently. Batch size is a core hyperparameter in AI training and a practical lever for AI serving cost and throughput.
Benchmark: A standardized test used to compare AI models: a fixed set of questions or tasks with a scoring method — math problem sets, coding challenges, graduate-level Q&A. Benchmark numbers headline every model release, but they can saturate (top models all near 100%) or leak into training data, so a single score deserves healthy skepticism. See also evals.
Bias: Systematic skew in a model's outputs, usually inherited from patterns in its training data — for instance, associating certain professions with certain demographics. Bias can also be introduced or amplified by fine-tuning choices. Measuring and mitigating it is a core part of model evaluation and of AI safety.

C ↑ top

Catastrophic Forgetting: When a model trained on a new task loses performance on earlier tasks — it "forgets" old skills as new ones are added. Catastrophic forgetting is a central challenge in continual learning and multi-stage AI training. Techniques like replay buffers, regularization, and careful fine-tuning schedules try to keep prior AI capabilities intact while teaching something new.
Chain of Thought (CoT): A technique in which a model works through a problem step by step before giving its final answer — either because the prompt asks it to ("think step by step") or because the model was trained to reason that way. Reasoning models industrialize this idea, spending extra inference-time compute thinking before they answer.
Chatbot: A conversational interface to an AI system: you type (or speak), it answers, and the exchange continues across turns. Early chatbots followed scripts; modern ones are built on large language models and can hold open-ended conversations, write code, and analyze documents. The chatbot is the interface; the model behind it is what varies.
Chunking: Splitting long documents into smaller pieces before embedding and storing them for RAG. Chunk size and overlap decide what the AI can retrieve: too large and retrieval is noisy; too small and context is lost. Good chunking is often the highest-leverage fix when an AI knowledge system returns weak answers.
Compute: Shorthand for the processing power used to train and run AI models, measured in chip-hours or FLOPs (floating-point operations). Compute is one of the three ingredients of model capability alongside data and algorithms — see scaling laws — and access to it (chips, data centers, electricity) has become an industrial and geopolitical story of its own.
Computer Vision: The branch of AI that enables machines to interpret images and video — detect objects, read text (OCR), track motion, and describe scenes. Modern computer vision is powered by CNNs, vision transformers, and increasingly multimodal AI models that combine vision with language. It underpins self-driving perception, medical imaging, and visual search.
Constitutional AI: A training approach (popularized by Anthropic) where an AI model is steered by a written set of principles — a "constitution" — and critiques or revises its own outputs against those rules, often with less human labeling than classic RLHF. Constitutional AI is one strategy for AI alignment and safer assistant behavior at scale.
Context Engineering: The practice of assembling everything a model sees at inference time — instructions, retrieved documents, tool results, conversation history — so the model has exactly the information it needs inside its context window. Increasingly used as the successor term to prompt engineering now that agentic systems juggle far more than a single prompt.
Context Window: The maximum amount of text — measured in tokens — that a model can consider at once: its working memory. Everything you send plus everything the model writes back must fit inside it. Windows have grown from a few thousand tokens to millions, though long context reduces rather than eliminates the need for retrieval techniques like RAG.
Convolutional Neural Network (CNN): A neural network architecture designed for grid-like data such as images. Convolution filters scan for local patterns — edges, textures — and deeper layers compose them into higher-level features like faces or objects. CNNs powered the 2012–2020 computer-vision boom; transformers now dominate many vision tasks, but CNNs remain widely deployed.

D ↑ top

Data Labeling: The human (and increasingly automated) work of annotating training examples — tagging what's in an image, ranking chatbot answers, marking toxic content — so models can learn from them. It is a large, mostly invisible industry, and the quality of the labels caps the quality of the model. RLHF depends on it directly.
Dataset: A collected set of examples used to train, validate, or test an AI model — text corpora, image banks, preference rankings, or tool-use traces. Dataset quality often matters more than clever architecture: biased, duplicated, or contaminated data produces biased or overstated AI results. See also training data and synthetic data.
Deep Learning: The branch of machine learning built on neural networks with many layers ("deep"), which learn useful features directly from raw data instead of needing hand-engineered rules. Essentially every headline AI system of the last decade — image recognition, speech, LLMs — is deep learning.
Deepfake: Synthetic media — video, audio, or images — that convincingly mimics a real person's likeness or voice, usually via generative AI or diffusion models. Deepfakes are useful for film and accessibility, and risky for fraud, disinformation, and non-consensual content. AI detection, watermarking, and provenance standards are active responses.
Diffusion Model: A generative model that creates images — and increasingly video and audio — by learning to reverse a noising process: start from pure random noise and denoise it step by step toward an output that matches the prompt. Diffusion is the technique behind most AI image and video generators. Compare the transformer, which dominates text.
Direct Preference Optimization (DPO): A method for aligning language models to human preferences without training a separate reward model as in classic RLHF. DPO optimizes the policy model directly on preferred vs. rejected response pairs. It has become a popular post-training recipe for open and commercial AI assistants because it is simpler to run while still improving helpfulness and harmlessness.
Distillation: Training a smaller "student" model to imitate a larger "teacher" model's outputs, preserving much of the capability at a fraction of the size and inference cost. It's how labs produce fast, cheap variants of their flagship models — and a controversy when done against a competitor's model without permission.

E ↑ top

Edge AI: Running AI models on devices near the user — phones, laptops, cars, factory sensors — instead of only in the cloud. Edge AI cuts latency, protects privacy by keeping data local, and works offline, at the cost of smaller models and tighter memory budgets. Often paired with quantization, small language models, and on-device AI optimizations.
Embedding: A list of numbers (a vector) that represents the meaning of a piece of text, an image, or other data, arranged so that similar things sit near each other in the vector space. Embeddings power semantic search, recommendations, and the retrieval step of RAG; they are stored and searched in a vector database.
Emergent Capabilities: Abilities that show up in larger models without being explicitly trained for — arithmetic, translation, code — seemingly appearing once scale crosses a threshold. Researchers debate whether emergence is real or an artifact of how capabilities are measured, but the term captures why each jump in scale manages to surprise even the people building the models.
Epoch: One full pass through the training dataset during model training. AI training runs for many epochs (or fractions of an epoch on huge web-scale corpora). Too few epochs underfit; too many risk overfitting. Epoch count is a basic training hyperparameter.
Evals: Industry shorthand for systematic evaluations of a model's behavior: capability benchmarks, safety tests, and regression suites run before and after release. Frontier labs publish eval results with launches, and "we need better evals" is the standard refrain whenever models behave unexpectedly in the wild.
Explainability (XAI): Techniques for understanding why a model produced a given output — from visualizing attention to mechanistic interpretability, which tries to reverse-engineer the circuits inside a network. Explainability matters for debugging, regulation, and deciding whether to trust models in high-stakes settings.

F ↑ top

Federated Learning: A training setup where many devices or organizations improve a shared AI model without centralizing raw data: each site trains locally and only model updates are aggregated. Federated learning is used when privacy, regulation, or logistics block classic data lakes — for example hospitals collaborating on medical AI. It is one privacy-preserving approach in the broader AI toolbox.
Few-Shot Learning: Getting a model to perform a task by showing it a handful of examples inside the prompt itself — no retraining involved. The ability to learn a task from examples on the fly is a signature capability of large language models. With no examples at all, it's zero-shot learning.
Fine-Tuning: Continuing a pre-trained model's training on a smaller, task-specific dataset to specialize it — a support bot trained on your tickets, a coding model trained on your codebase's conventions. Far cheaper than training from scratch, and cheaper still with parameter-efficient methods like LoRA.
Foundation Model: A large model pre-trained on broad data that serves as the base — the foundation — for many downstream uses via prompting or fine-tuning. GPT, Claude, Gemini, and Llama are all foundation models; most AI products are layers built on top of one.
Frontier Model: The most capable models at the current state of the art — the ones pushing the frontier of what AI can do. The term carries policy weight: "frontier AI" safety frameworks and commitments target models whose novel capabilities may pose novel risks, and several labs gate releases on frontier-capability evals.
Function Calling: A model API feature where the model can reply with a structured request to run a function the developer defined — get_weather(city) — instead of plain text. The application executes the function and hands the result back. Function calling (also called tool calling) is the plumbing that makes tool use and agents reliable.

G ↑ top

Generative Adversarial Network (GAN): An earlier generative AI architecture with two networks: a generator that creates samples and a discriminator that tries to tell fakes from real data. The adversarial game produced sharp images and deepfakes before diffusion models became dominant. GANs remain part of AI history and still appear in specialized generation and compression tasks.
Generative AI: AI that produces new content — text, images, code, audio, video — rather than only classifying or ranking existing data. The post-2022 wave of chatbots and image generators that took AI mainstream is generative AI, built mostly on transformers and diffusion models.
GPT (Generative Pre-trained Transformer): A family of large language models built on the transformer architecture and trained with a two-stage recipe: broad pre-training then instruction/preference tuning. "GPT" started as OpenAI's product line and became shorthand in AI news for chatbot-class models. Technically it means generative, pre-trained, transformer-based AI — not one single model forever.
GPU (Graphics Processing Unit): The chip class that made deep learning practical: thousands of cores doing parallel matrix math, originally for rendering games. Training a frontier model takes tens of thousands of them running for months, which is why GPU supply shapes the entire industry. NVIDIA dominates; Google's TPU is the best-known alternative.
Gradient Descent: The optimization method that updates a model's parameters by stepping opposite the gradient of the loss — downhill on the error surface. Variants (SGD, Adam) power almost all modern AI training. Combined with backpropagation, gradient descent is how AI models actually learn from data at scale.
Grounding: Connecting a model's answers to authoritative, verifiable sources — retrieved documents, live search results, databases — so its output reflects checkable facts rather than only what's compressed in its parameters. RAG is the standard grounding technique, and grounded answers can cite their sources. The opposite failure mode is hallucination.
Guardrails: The layer of checks wrapped around a deployed model that constrains what it will do: refusal policies, content filters, input and output validation, rate limits. Guardrails are product-level defenses; alignment works at the training level. Jailbreaks and prompt injection are the attacks they exist to stop.

H ↑ top

Hallucination: When a model states something false with complete confidence — an invented citation, a nonexistent API, a wrong date. It happens because LLMs generate statistically plausible text rather than looking facts up. Mitigations include grounding answers in retrieved sources, better post-training, and models that say "I don't know" — but no current model is immune.
Human-in-the-Loop: System design in which a human reviews, approves, or corrects an AI system's decisions at key points instead of letting it run fully autonomously. Standard practice for high-stakes uses like medical or legal work, and the pattern behind the approval steps in agent workflows. Compare data labeling, where humans improve the model rather than police its output.
Hyperparameter: A training or serving choice set by humans rather than learned as model weights: learning rate, batch size, temperature, context length, or number of experts. Hyperparameters shape AI quality and cost; poor choices can waste millions in compute or tank accuracy. Tuning them is part art, part experiment design.

I ↑ top

In-Context Learning: When a large language model adapts to a task from examples or instructions inside the prompt alone — no weight updates. Few-shot prompting is the classic case of in-context learning. It is one reason modern AI feels flexible: the same model can switch roles mid-conversation by changing context, not by re-fine-tuning.
Inference: Running a trained model to get outputs — every chatbot reply, every generated image is inference, as distinct from training. Inference cost and speed now drive much of AI engineering (see quantization and distillation), and "inference-time compute" — extra thinking while answering — is the axis reasoning models scale on.
Instruction Tuning: Fine-tuning a raw pre-trained model on instruction-and-response examples so it learns to follow directions like an assistant instead of merely continuing text. It's the step that turns a base model into a chat model, typically followed by preference training such as RLHF.

J ↑ top

Jailbreak: A prompt crafted to bypass a model's safety training and make it produce output it would normally refuse — via role-play framing, encoded text, or gradual multi-turn manipulation. Jailbreaking is done by the user; prompt injection comes from third-party content. Defending against both is a core job of red teams and guardrails.

K ↑ top

Knowledge Cutoff: The date after which a model's training data ends. Ask about events past the cutoff and the model either admits ignorance or hallucinates. Products work around it by giving models live search or retrieval via tool use — which is how a chatbot can discuss yesterday's news despite a months-old cutoff.
Knowledge Graph: A structured network of entities and relationships (people, products, concepts) that AI systems can query for precise facts. Knowledge graphs complement LLMs: models write fluently; graphs store verifiable links. Many enterprise AI stacks combine both — graph or database facts plus generative wording — to reduce hallucination.
KV Cache: Key-value cache: stored attention states from previous tokens so an autoregressive AI model does not recompute them on every new token. The KV cache is a major driver of memory use and speed during inference, especially for long context windows and multi-turn chat. Optimizing it is central to cheap, fast AI serving.

L ↑ top

Large Language Model (LLM): A neural network — in practice always a transformer — trained on vast text corpora to predict the next token, and which as a byproduct learns to write, summarize, translate, reason, and code. LLMs are the core technology behind modern chatbots, agents, and most of the AI news cycle.
Latency: The time between sending a request and getting a response. For chat products, time-to-first-token matters most; for agents, whole-task latency does. Latency trades off against model size and reasoning effort — the reason providers ship both fast-and-cheap and slow-and-smart tiers of the same model family.
Latent Space: The internal multi-dimensional map where a model represents meaning after encoding inputs as vectors. Nearby points in latent space tend to mean similar things — the geometry behind embeddings, image morphs, and semantic search. When AI people say a model "understands" structure, they often mean it has organized a useful latent space.
Long Context: Models and products that accept very large context windows — hundreds of thousands or millions of tokens — so AI can read whole codebases, contracts, or multi-hour transcripts in one go. Long context reduces some need for RAG, but retrieval and memory still matter for cost, freshness, and focus.
LoRA (Low-Rank Adaptation): A parameter-efficient fine-tuning method that freezes the base model and trains only small "adapter" matrices — often under 1% of the parameters — making customization drastically cheaper and the resulting adapters easy to swap in and out. Ubiquitous in the open-weights ecosystem for both language and image models.

M ↑ top

Machine Learning (ML): The approach of having computers learn patterns from data rather than following hand-written rules — show a system enough examples and it works out the rules itself. Its main paradigms are supervised, unsupervised, and reinforcement learning; deep learning is its dominant modern branch.
Memory (Agent Memory): Mechanisms that let an AI agent retain information across steps or sessions: short-term scratchpads in the prompt, long-term stores in a vector database, or structured profiles. Without memory, AI agents restart cold every turn. With it, they can learn user preferences, track multi-day projects, and avoid repeating work.
Mixture of Experts (MoE): An architecture in which a model contains many "expert" sub-networks but activates only a few of them for each token, so total parameter count can be huge while the compute paid per token stays modest. Many frontier and open-weights models use MoE to buy capability at lower inference cost.
MLOps: Machine learning operations: the engineering practice of shipping and running AI/ML systems reliably — data pipelines, training jobs, model registry, deployment, monitoring, and rollback. MLOps applies DevOps ideas to AI so models are versioned and tested like software. Closely related to AI observability and production evals.
Model Collapse: A degradation mode where AI models trained mostly on other models' outputs lose diversity and start amplifying errors — a risk of recursive synthetic data loops. Model collapse is an active research concern as the web fills with AI-generated text and images. Mixes of real data, filtering, and careful curation are proposed mitigations.
Model Context Protocol (MCP): An open standard, introduced by Anthropic in late 2024, for connecting AI assistants to external tools and data sources through one common interface — so any MCP-compatible model can use any MCP server, instead of every integration being custom-built. Often described as "USB-C for AI tools," it was adopted across the industry through 2025 and underpins much of today's tool use.
Multi-Agent System: Multiple AI agents coordinating on one task — a planner delegating to workers, agents debating to check each other, or swarms of specialists handling research, coding, and review. Multi-agent setups can beat a single agent on complex work, at the price of orchestration complexity and errors that compound between agents.
Multimodal AI: Models that work across more than one kind of data — text, images, audio, video — as input, output, or both: describing a screenshot, holding a voice conversation, generating video from a script. Frontier models are now natively multimodal, trained on several modalities at once rather than having vision bolted onto a text model.

N ↑ top

Natural Language Processing (NLP): The field of getting computers to understand and generate human language. For decades it was a collection of separate specialized tasks — parsing, sentiment analysis, translation — each with its own models. The rise of LLMs largely unified it: one general model now handles most NLP tasks through plain prompting.
Neural Network: A mathematical function loosely inspired by the brain: layers of simple units pass signals forward, and the connection weights between them are adjusted during training until inputs map to correct outputs. Neural networks are the substrate of deep learning; an LLM is one with up to trillions of weights.

O ↑ top

On-Device AI: AI inference that runs entirely on the user's hardware (phone NPU, laptop GPU, browser WASM) without sending prompts to a remote server. On-device AI improves privacy and offline reliability; it usually means smaller models, quantization, and hardware-aware runtimes. A major product trend for personal AI assistants and OS-level features.
Open-Weights Model: A model whose trained weights are published for download, so anyone can run it locally, inspect it, or fine-tune it — Llama, Mistral, and DeepSeek are prominent examples. Often loosely called "open source," though full open source would also include training data and code, and licenses vary from permissive to restrictive.
Orchestration: The control layer that sequences AI calls, tools, retrieval, and business logic into a working product workflow. Orchestration frameworks decide when to call an LLM, which tools to expose, how to retry, and when to hand off to a human. In multi-step agentic AI, orchestration is as important as the model itself.
Overfitting: When a model memorizes its training data instead of learning patterns that generalize — acing training examples, failing in the real world. The classic machine-learning failure mode. Its modern echo in LLMs is benchmark contamination: a model that has seen the test in training looks smarter than it is.

P ↑ top

Parameters: The learned numeric weights inside a neural network — the model's knowledge, compressed into numbers. Counts run from millions to trillions, and more parameters generally means more capability and cost (see scaling laws), though data quality, MoE architectures, and distillation complicate the simple story.
PEFT (Parameter-Efficient Fine-Tuning): A family of methods — including LoRA — that adapt a large AI model by training only a small subset of parameters or lightweight adapters. PEFT makes custom AI cheaper: you keep a frozen foundation model and ship thin, task-specific layers. Standard practice for enterprise specialization of open-weight models.
Perplexity: A metric of how surprised a language model is by a stretch of text — lower perplexity means the model predicted the tokens more confidently. Used to evaluate language models during training and to compare AI systems on held-out data. (Not to be confused with the Perplexity AI search product; the metric predates the brand.)
Post-Training: Everything after raw pre-training that shapes a base model into a useful AI product: instruction tuning, preference optimization (RLHF, DPO), safety fine-tunes, and tool-use training. Post-training is where much of the "assistant personality," refusal behavior, and coding skill is added to modern AI models.
Pre-training: The first and most expensive phase of building a foundation model: training on trillions of tokens of broad data to predict the next token, which instills general knowledge of language, code, and the world. Post-training — instruction tuning, RLHF, reasoning training — then shapes that raw capability into a usable assistant.
Prompt: Whatever you send the model: a question, instructions, documents to analyze, examples to imitate, images to inspect. The model's output quality tracks prompt quality closely — the observation behind prompt engineering and its successor, context engineering. See also system prompt.
Prompt Engineering: The craft of writing prompts that reliably get the output you want: clear instructions, good examples (few-shot), role framing, explicit output formats, and structured reasoning. As models improved, the elaborate tricks mattered less and the discipline evolved toward context engineering for agentic systems.
Prompt Injection: An attack in which malicious instructions hidden inside content the model processes — a web page, an email, a document — hijack it into ignoring its real instructions: the LLM-era cousin of SQL injection. It is a top security concern for agents that browse and act on the world, and it differs from a jailbreak, which the user performs directly.

Q ↑ top

Quantization: Compressing a model by storing its weights at lower numeric precision — 16-bit floats down to 8- or 4-bit integers — which shrinks memory use and speeds up inference with only modest quality loss. Quantization is what makes large open-weights models runnable on laptops and phones.

R ↑ top

ReAct: A prompting and agent pattern that interleaves Reasoning and Acting: the AI thinks step by step, calls a tool, observes the result, and continues. ReAct-style loops power many AI agents that research, code, or browse. It combines chain of thought with tool use in one transparent cycle.
Reasoning Model: A model trained — typically with reinforcement learning — to produce a long internal chain of thought before answering, trading extra inference-time compute for much stronger math, code, and logic. Brought mainstream by OpenAI's o1 in 2024 and now standard across frontier labs, usually surfaced as a visible or hidden "thinking" phase.
Recurrent Neural Network (RNN): A neural architecture that processes sequences step by step with a hidden state carried forward — the pre-transformer workhorse of language and time-series AI. LSTMs and GRUs are popular RNN variants. Transformers largely replaced RNNs for large-scale language AI, but recurrent ideas still appear in efficient and streaming models.
Red Teaming: Deliberate adversarial testing of a model: jailbreaks, injection attacks, misuse scenarios, and probes for dangerous capabilities, run before and after release. Frontier labs use internal and external red teams, and findings feed guardrails, safety training, and sometimes release decisions.
Reinforcement Learning (RL): Training by trial and reward: an agent acts in an environment, receives a reward signal, and updates itself to earn more reward. RL produced the game-playing milestones (AlphaGo) and is now central to post-training language models — both RLHF and the RL on verifiable problems that builds reasoning models.
Responsible AI: The practice of building and deploying AI with attention to fairness, transparency, privacy, accountability, and societal impact. Responsible AI programs combine technical controls (evals, guardrails, red teams) with process (reviews, documentation, incident response). Overlaps AI governance and AI safety in enterprise settings.
Retrieval-Augmented Generation (RAG): An architecture in which relevant documents are fetched — usually via embeddings and a vector database — and inserted into the prompt, so the model answers from your data and can cite sources. RAG is the default enterprise pattern for grounding, freshness beyond the knowledge cutoff, and reducing hallucination.
RLHF (Reinforcement Learning from Human Feedback): The post-training recipe in which humans rank model outputs, a reward model learns those preferences, and reinforcement learning optimizes the model against that reward. RLHF is the technique that made chatbots helpful and polite rather than merely fluent; successors include RLAIF, where AI provides the feedback.

S ↑ top

Scaling Laws: Empirical findings that model performance improves smoothly and predictably as parameters, data, and compute scale up together — the thesis that justified ever-larger frontier models. How far pre-training scaling can go is actively debated; reasoning models opened a second scaling axis at inference time.
Self-Supervised Learning: Training where the AI model creates its own labels from raw data — for example predicting the next token or a masked word — so humans do not hand-label every example. Most modern foundation model pre-training is self-supervised. It is why AI could scale to internet-sized corpora without internet-sized annotation budgets.
Semantic Search: Search that matches meaning rather than only keywords, usually by embedding queries and documents into a shared vector space and finding nearest neighbors. Semantic search is the retrieval engine inside many RAG systems and AI knowledge bases. It pairs with a vector database and good chunking.
Shadow AI: Employees using consumer AI tools (chatbots, image generators, coding assistants) for work without IT approval or data controls. Shadow AI spreads because AI is useful and easy, but it creates leakage and compliance risk. Organizations respond with approved AI copilots, training, and clearer AI governance.
Small Language Model (SLM): A compact language model — typically a few hundred million to a few billion parameters — good enough for many tasks while running cheaply, fast, on-device, or in privacy-sensitive settings. Built through careful data curation and distillation from larger teachers; phones and laptops increasingly ship with one.
Speculative Decoding: An inference speedup where a small draft model proposes several tokens and a larger AI model verifies them in parallel, accepting correct guesses. Speculative decoding can cut latency without changing the final distribution of the big model. One of several serving tricks (with KV caches and quantization) that make frontier AI feel snappier.
Structured Output: Constraining an AI model to return JSON, XML, or another schema instead of free prose — so downstream code can parse results reliably. Structured output is critical for production AI: agents, form-fillers, and pipelines need fields, not essays. Often implemented via tool schemas, grammar constraints, or function calling.
Superintelligence (ASI): Hypothetical AI decisively beyond the best humans at essentially every cognitive task — including AI research itself, which is what could make its arrival self-accelerating. Whether and when it is achievable is contested; it anchors the strongest safety arguments and, lately, the mission statements of several labs. One step beyond AGI.
Supervised Learning: Learning from labeled examples: each input comes with the correct answer — this email is spam, this photo is a cat — and the model learns the mapping. Most deployed machine learning historically worked this way, which is why data labeling became an industry. Compare unsupervised and reinforcement learning.
Sycophancy: When an AI model agrees with the user even if the user is wrong — flattering rather than truthful. Sycophancy is a known failure mode of preference-trained assistants and a target for better evals and alignment methods. Users who want accurate AI should watch for models that never push back.
Synthetic Data: Training data generated by models rather than collected from the world: model-written math problems and solutions, simulated dialogues, auto-labeled images. Labs lean on it as high-quality human data grows scarce and to target specific skills — with careful quality control, because training models heavily on model output can degrade them (the "model collapse" concern).
System Prompt: The standing instructions a developer gives a model before any user input: persona, rules, available tools, output format, guardrail policies. Users don't normally see it, and much of jailbreaking and prompt injection amounts to trying to override it.

T ↑ top

Temperature: A sampling setting that controls how random a model's output is. Low temperature makes it pick the most likely next token — focused, repeatable, right for code and facts. High temperature spreads the probability — varied and creative, better for brainstorming. One of the few dials exposed by most model APIs.
Test-Time Compute: Spending extra inference resources after training — longer chains of thought, search, self-critique, or multiple samples — to improve answers on hard problems. Test-time compute (also called inference-time compute) is the idea behind many reasoning models: more thinking tokens, better AI performance, higher cost per query.
Token: The unit a model actually reads and writes: a word piece averaging roughly four characters of English ("glossary" might be one token; "ZealVector" several). Tokenization is the process of splitting text into them. Pricing, context windows, and generation speed are all measured in tokens, which is why the word is everywhere in AI product news.
Tokenizer: The component that splits text into tokens the model can process (and maps tokens back to text). Tokenizer design affects cost, multilingual quality, and how AI handles code or rare words. Different AI models use different tokenizers, so the same paragraph can be a different token count across providers.
Tool Use: A model's ability to call external tools — web search, code execution, databases, browsers, other APIs — usually via function calling, to overcome its knowledge cutoff and act on the world rather than just describe it. Tool use is the defining capability of agents, and standards like MCP exist to make it interoperable.
Top-p (Nucleus Sampling): A decoding setting that samples the next token only from the smallest set of candidates whose cumulative probability exceeds p (for example 0.9). Together with temperature, top-p controls how random or focused AI text generation feels. Common defaults in creative vs. factual AI applications.
TPU (Tensor Processing Unit): Google's custom AI accelerator chip family — an alternative to the GPU — used internally to train and serve Gemini and rentable through Google Cloud. TPUs are the most established example of the industry-wide push toward custom AI silicon as compute demand explodes.
Training Data: The corpus a model learns from: web text, code, books, images, licensed datasets, and increasingly synthetic data. Its composition determines what a model can do and what biases it carries; its provenance is the subject of copyright lawsuits and a wave of licensing deals between AI labs and publishers.
Transfer Learning: Reusing what a model learned on one task or domain as the starting point for another, instead of training from zero. It is the principle that makes pre-train-then-fine-tune work at all — general knowledge transfers, so a small specialized dataset goes a long way.
Transformer: The neural architecture introduced in Google's 2017 paper "Attention Is All You Need," built on self-attention and highly parallelizable — which is what made training on internet-scale data feasible. It is the T in GPT and the basis of virtually every frontier model, across text, vision, and audio.
Tree of Thoughts: A reasoning technique where an AI explores multiple intermediate paths — a tree of partial solutions — instead of one linear chain of thought, then selects or votes on the best branch. Tree of Thoughts can improve hard planning and puzzle-style tasks at the cost of more test-time compute.
Turing Test: Alan Turing's 1950 proposal: if a machine's side of a conversation is indistinguishable from a human's, treat it as intelligent. Modern chatbots pass casual versions, which mostly revealed the test's limits — fluent conversation turned out not to equal general intelligence — so the field measures progress with benchmarks and evals instead.

U ↑ top

Unsupervised Learning: Learning structure from unlabeled data — clustering similar customers, spotting anomalies — with no human-provided answers. The self-supervised pre-training of LLMs is its modern descendant: next-token prediction manufactures its own labels from raw text, which is what lets models learn from the whole internet without anyone annotating it.

V ↑ top

Vector Database: A database built to store embeddings and find nearest neighbors fast at scale — the retrieval layer behind RAG and semantic search. Examples include Pinecone, Weaviate, Milvus, and pgvector. (And no — ZealVector, despite the name, is an AI news digest, not a vector database.)
Vibe Coding: Building software by describing what you want to an AI coding agent and iterating on the results, without reading most of the generated code. Coined by Andrej Karpathy in early 2025, it went from half-joke to genuine workflow within months — and raised real questions about review, security, and maintainability of code nobody has read.
Vision-Language Model (VLM): A multimodal AI model that jointly understands images (or video frames) and text — captioning photos, answering questions about screenshots, or following UI instructions. VLMs power document AI, visual agents, and "see-and-talk" products. They combine computer vision encoders with language model decoders or fully unified multimodal transformers.
Vision Transformer (ViT): A transformer architecture applied to images by splitting them into patches treated like tokens. Vision transformers are a backbone of modern computer vision and of many multimodal AI systems. They largely closed the gap with (and often surpassed) pure CNN pipelines for large-scale visual AI.

W ↑ top

Weights: The learned numeric parameters inside a neural network — the "brain state" of an AI model after training. When labs release "open weights," they publish these numbers so others can run or fine-tune the model. Weight count is the informal size of an AI model (7B, 70B, etc.), though architecture and training data matter as much as the headline figure. See parameters and open-weights model.
World Model: A system's internal model of how an environment works, used to predict what happens next and plan actions — "if I drop this, it falls." Whether LLMs implicitly learn one from text alone is a live debate; building explicit world models is a major research bet for robotics, video generation, and game-like simulation.

Z ↑ top

Zero-Shot Learning: A model performing a task it was given no examples of — instructions only — relying purely on generalization from pre-training. "Translate this to German" with no sample translations is zero-shot. It is the everyday mode of using chat models; add examples to the prompt and it becomes few-shot learning.

AI glossary FAQ

Quick answers about this AI glossary, how AI terms are chosen, and how to use artificial intelligence vocabulary when you read AI news or ship AI features.

What is an AI glossary?

An AI glossary is a reference list of artificial intelligence terms with clear definitions. This page explains AI concepts — from AI agents and generative AI to LLMs, RAG, fine-tuning, and AI safety — in plain English so non-specialists and builders share the same language.

What categories of AI terms are defined here?

This ZealVector AI glossary defines AI terms covering foundational machine learning ideas, modern large language models, generative AI systems, AI agents, retrieval, evaluation, hardware, and AI governance. We add terms as they show up in weekly AI news.

Who is this AI terminology guide for?

Product managers evaluating AI vendors, engineers integrating AI APIs, founders shipping AI features, journalists and analysts covering AI, and anyone building AI literacy. If you can follow our weekly AI digest, you can use this glossary.

What is the difference between AI, machine learning, and deep learning?

Artificial intelligence (AI) is the broad field of machines performing intelligent tasks. Machine learning (ML) is the subset that learns patterns from data instead of only hand-written rules. Deep learning is ML with multi-layer neural networks — the approach behind most modern AI models including LLMs and image generators.

What AI terms should beginners learn first?

Start with AI, LLM, generative AI, prompt, token, hallucination, AI agent, and RAG. Those eight AI terms unlock most product demos and AI news headlines. Then branch into fine-tuning, embeddings, and AI safety as needed.

Can I suggest a new AI term?

Yes. Email admin@zealvector.com with the AI term and where you saw it. If it is showing up in AI news or product docs, we will prioritize adding a plain-English definition.

Missing an AI term you keep seeing in the news? Email admin@zealvector.com and we'll add it to this AI glossary. Prefer the week's artificial intelligence developments explained in the same plain-English style? The weekly AI news digest takes five minutes — or see what we build for AI products.