AI Fundamentals Glossary

AI conversations move fast. This course gives you the vocabulary to keep up, in videos short enough to watch before a call. The AI Fundamentals Glossary covers more than 90 of the terms you will actually hear in the field, explained in plain language across 30+ short videos. No technical background needed. What's inside: - Fundamentals: AI, machine learning, GPUs, neural networks, deep learning - Predictive AI: AutoML, forecasting, classification, tabular foundation models - Generative AI: LLMs, RAG, tokens, prompt engineering, fine-tuning - Agentic AI: agents, orchestration, MCP, agent builders, knowledge graphs - Observability: MLOps, monitoring, drift, evaluation, governance - Platform: on-prem, air-gapped, cloud, VPC, sovereign AI - Compliance: US, Europe, APAC and Latin America frameworks Watch it in order or jump to the terms you need. Finish the course and take the quiz to earn your H2O.ai University certification. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #AI #AIGlossary #MachineLearning #GenerativeAI #H2Oai

Agentic is in almost every AI conversation right now. Here is what it actually means, and how an agent differs from an assistant that only answers questions. An AI agent takes a goal and works out the steps itself. Instead of someone manually pulling reports, running analysis and flagging accounts, the agent accesses the data, identifies at-risk customers, prioritizes them and surfaces the list ready to act on. Terms covered: - AI Agent: an AI that acts on a goal rather than responding to a prompt - Agentic AI: systems that operate across multiple steps, decide along the way and adapt when something goes wrong - Agentic: what the word signals when it describes a workflow, a platform or a process Real examples: churn reduction in retail, patient follow-ups in healthcare, overnight compliance checks in financial services, live rerouting in logistics. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #AgenticAI #AIAgents #AI #AIGlossary #H2Oai

Most AI systems are not one model doing everything. They are several components working together, and something has to keep them in sync. Think of an insurance claim: one component extracts information from documents, another checks it against policy rules, another routes the case to the right team. Orchestration is the logic that decides which does what, in what order, and what gets passed between them. Terms covered: - Orchestration: the coordination layer across components - Agent Orchestration: orchestration applied to autonomous agents handing off work - Agent Runtime: the execution environment where agents actually operate - Agent Harness: the framework connecting agents, tools, memory and instructions into one system This is how AI goes from a single smart response to a coordinated system handling complex, multi-step work at scale. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #AgenticAI #AIOrchestration #AI #AIGlossary #H2Oai

Some agents are built for one job. Others reason across many. Knowing the difference is what helps you match the right architecture to the right problem. Terms covered: - Deep Research Agent: searches sources, reads documents, cross-references and produces a structured answer end to end - Coding Agent: writes, tests and debugs code on its own, iterating until it works - Vertical Agents: agents built for a specific industry, where the specialization is what makes them precise - RAG Agent and Agentic RAG: retrieval as an active decision rather than a fixed step - General-Purpose Agents: agents that decide for themselves whether a goal needs research, code, data or a combination Task-specific agents go deep. General-purpose agents go broad. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #AIAgents #AgenticAI #AI #AIGlossary #H2Oai

Knowing what an agent can do is one thing. Knowing how it works inside is another, and it is more structured than most people expect. Give a general-purpose agent a goal and it plans the steps, picks its tools, executes, checks its own results, self-corrects when something looks wrong, and delivers a summary or output files. Plan, act, check, self-correct, deliver. That loop is what separates an agent from a chatbot. Terms covered: - Agent Tools: the specific actions an agent can take, from web search to running code to calling APIs - Tool Builder: creating custom tools so an agent can reach internal databases, custom APIs and legacy systems - Agent Builder: the environment where you set the agent's goal, tools, instructions and constraints, without deep engineering The internal loop gives it intelligence. The tools define its reach. The builder is how you put it together. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #AIAgents #AgenticAI #AI #AIGlossary #H2Oai

Connecting an agent to a new tool used to mean custom integration work every single time. MCP is the standard that removes that. Terms covered: - MCP (Model Context Protocol): a standardized way for AI models to communicate with external tools and data sources. One standard, many integrations. - MCP Runner: the runtime component that manages those connections and makes sure the interactions happen reliably at scale In practice: a healthcare organization connects an agent to patient records, scheduling and compliance tools through the same protocol. A financial services firm links market data, risk systems and reporting without rebuilding the connection layer each time. MCP defines how agents talk to the world. The MCP Runner makes those conversations happen. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #MCP #AIAgents #AgenticAI #AIGlossary #H2Oai

AI does not automatically understand your organization. It does not know how your products relate to your customers or what your internal terminology means. It needs a map. Terms covered: - Knowledge Graph: information represented as a network of connected entities and relationships rather than rows in a table, so an AI system can reason over how things relate - Ontology: the formal framework that defines the concepts in your domain and how they connect, giving the graph its logic In healthcare, an ontology defines what counts as a symptom, a diagnosis and a treatment. In finance, it defines instruments, transactions, counterparties and risk categories. Together they give AI a shared understanding of your world, so it reasons accurately instead of guessing. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #KnowledgeGraph #Ontology #AI #AIGlossary #H2Oai

Every agent, language model and predictive system is built on the same foundation. Get this part right and the rest of the glossary is easier. Terms covered: - Artificial Intelligence: systems that learn, adapt and improve from data instead of following fixed rules written by a programmer - Machine Learning: the mechanism, where a model learns to tell fraud from non-fraud by seeing thousands of examples - Features: the variables and attributes that tell a model what to pay attention to - GPU: the hardware that makes training viable on text, images and video, built for thousands of calculations running at once AI sets the vision. Machine learning is the mechanism. Features are what the model learns from. GPUs are what make it scalable. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #MachineLearning #AI #GPU #AIGlossary #H2Oai

Processing language and truly understanding it are two different things. Here is the line between them, and the bigger ambition sitting behind both. Terms covered: - NLP (Natural Language Processing): translating, classifying, extracting and detecting sentiment. Taking raw text and doing something meaningful with it. - NLU (Natural Language Understanding): a subset of NLP focused on intent, meaning and context. Recognizing that "I want to cancel" signals a customer about to churn, not just four words typed. - AGI (Artificial General Intelligence): a hypothetical AI that reasons and performs across any domain the way a human can From processing, to understanding, to reasoning. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #NLP #NLU #AGI #AIGlossary #H2Oai

For most of AI's history the focus was analyzing and predicting. Generative AI creates. Here is what changed, and how to choose between proprietary and open source models. Terms covered: - Generative AI: AI that produces text, code, images, summaries and answers on demand - GPT (Generative Pre-trained Transformer): the architecture behind most large language models - Proprietary LLMs: powerful, easy to start with, priced per token, and your data goes to a third-party server - Open Source LLMs: weights you can download, deploy on your own infrastructure and fine-tune, with no per-token cost and no data leaving your environment Most serious deployments need both. Proprietary for speed and access, open source for control, privacy and cost at production scale. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #GenerativeAI #LLM #OpenSourceAI #AIGlossary #H2Oai

A language model does not know your internal documents, and it does not know what happened after it was trained. Retrieval is how you close that gap. Terms covered: - Information Retrieval: finding relevant content from a large collection in response to a query - RAG (Retrieval-Augmented Generation): retrieve relevant documents at the moment a question is asked, then generate a grounded answer from them - Graph RAG: uses a knowledge graph to retrieve, so the system navigates relationships between entities instead of matching keywords - Agentic RAG: treats retrieval as active reasoning. It rewrites queries, routes across sources, checks whether results are useful and retries when they are not. Standard RAG for direct queries. Graph RAG when relationships matter. Agentic RAG when the retrieval itself needs to reason. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #RAG #GenerativeAI #AgenticAI #AIGlossary #H2Oai

Language models do not read words. They read tokens. That one detail shapes output quality and your AI bill. Terms covered: - Token: the basic chunk of text a model processes. Short words are usually one token, longer words split into several, and punctuation counts too. - Tokenomics: the economics of token usage, and why costs add up fast across thousands of users and millions of interactions - NLG (Natural Language Generation): the capability that produces human-readable text from data or instructions - NER (Named Entity Recognition): pulling people, organizations, dates, locations and monetary values out of unstructured text From how text enters a model, to what it costs, to what comes out, to how meaning gets extracted. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #LLM #Tokenomics #NLP #AIGlossary #H2Oai

The same model gives very different answers depending on how you ask. Prompt engineering is how you take control of that. Terms covered: - Prompt Engineering: designing inputs that consistently produce useful, accurate, well-structured outputs. Specify the task, give context, set the format, anticipate edge cases. - System Prompt: the instructions given to a model before the conversation starts, defining its persona, scope, tone and constraints. Users rarely see it, but it governs everything. - Prompt Catalog: a library of tested, reusable prompts so teams stop reinventing the same one from scratch Prompt engineering is not about tricking the model. It is about communicating clearly with it. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #PromptEngineering #LLM #GenerativeAI #AIGlossary #H2Oai

Not every use case needs a large model. Smaller and purpose-built is often faster, cheaper and better suited to the job. Terms covered: - SLM (Small Language Model): fewer parameters, trading breadth for precision. Runs on-premises or on edge devices with no cloud dependency. - SLM Development: designing, training and optimizing a small model for a well-defined task or dataset - Fine-Tuning Engine: taking a pre-trained model and training it further on domain-specific data, so it learns the language of law, medicine or your business - LLMOps: the infrastructure and practices for deploying, managing, monitoring and updating language models over time A hospital running clinical documentation on local servers. A manufacturer processing quality reports on factory floor devices. These are where compact models win. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #SLM #FineTuning #LLMOps #AIGlossary #H2Oai

AI can be confidently wrong. Here is why it happens, how grounding fixes it, and how to measure whether it worked. Terms covered: - Hallucination: models are trained to generate statistically likely text, not to verify facts. When a model does not know, it fills the gap with something plausible. - Grounded Generation: anchoring responses in verified, retrieved sources so answers are traceable and far less likely to be fabricated - LLM Evaluation: assessing outputs for accuracy, relevance, coherence, cost and safety, as an ongoing practice rather than a one-time check - RAG Evaluation: measuring retrieval and generation separately, because a system can retrieve well and generate poorly, or generate fluently from the wrong sources The goal is not perfect AI. It is trustworthy AI, and that requires knowing how to measure it. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #LLM #Hallucinations #ResponsibleAI #AIGlossary #H2Oai

Language models are not just a chat technology. The same capability reads your documents and queries your databases. Terms covered: - Conversational AI: multi-turn dialogue that remembers context, handles follow-ups, adapts to intent and escalates when it hits its limits - Document Extraction: pulling structured information out of contracts, invoices, claims, medical records and filings, in seconds instead of hours - Text-to-SQL: asking a database a question in plain English and getting the answer back, with no technical intermediary and no waiting on a data team Conversation, documents and data. Three very different surfaces, made accessible through the same underlying capability. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #ConversationalAI #DocumentAI #TextToSQL #AIGlossary #H2Oai

Before you can talk about where AI adds value, you need to know which layer of the enterprise you are talking about. Terms covered: - System of Record: the authoritative source of truth for a data type. CRM for customers, ERP for financials, HR systems for employees. - System of Content: where unstructured content lives. Documents, emails, reports, contracts. Valuable, but harder to query and act on. - System of Intelligence: the AI and analytics layer sitting on top of both. It does not store data, it makes data useful. - AI Maturity: a framework for assessing how developed each layer is, and how effectively AI is applied across the organization Most organizations do not have all three working well at once. Knowing where an organization sits is what makes the AI conversation meaningful. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #EnterpriseAI #AIMaturity #DataStrategy #AIGlossary #H2Oai

Low code opens AI to the people who understand the business problem. Deep expertise is what makes low code possible in the first place. Terms covered: - Low-Code AI: visual interfaces, guided workflows and pre-built components instead of manual code. A business analyst trains a predictive model without writing a line of Python. - The roles behind it: data scientists, machine learning engineers, AI researchers and solutions engineers - Kaggle Grandmaster: the highest rank on the world's largest data science competition platform, held by fewer than a few hundred people worldwide Low code does not eliminate expertise. It removes the barrier that kept AI out of reach for most of an organization. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #LowCodeAI #DataScience #Kaggle #AIGlossary #H2Oai

Training a model is half the work. Getting it into production and keeping it reliable is where the real operational complexity lives. Terms covered: - MLOps: deployment, versioning, monitoring and retraining. What DevOps did for software engineering, MLOps does for machine learning. - LLMOps: the same principles extended to language models, adding prompt versioning, token cost control, hallucination monitoring and context window limits - Model Hub: a central repository where models are published, discovered, versioned and deployed from one place Without a model hub, models scatter across teams and machines with no visibility into what is running where, or which version is live. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #MLOps #LLMOps #AIinProduction #AIGlossary #H2Oai

Deploying an AI system is the start of a responsibility, not the end of one. Once a model or agent is live, you need to know what it is doing and why. Terms covered: - Observability: understanding what is happening inside a system from what it produces. A design property, built in rather than added later. - Monitoring: tracking defined metrics over time and alerting when something crosses a threshold - Model Monitoring: prediction accuracy, confidence scores, output distributions, latency - Agent Monitoring: what actions agents take, how they use tools, and whether they stay within their boundaries - Telemetry: the continuous collection of logs, traces and metrics that feeds all of it Without telemetry, monitoring is blind. Without monitoring, observability has nothing to analyze. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #AIObservability #Monitoring #MLOps #AIGlossary #H2Oai

A model that was accurate last year can quietly become unreliable this year. Drift is why, and interpretability is how you catch it. Terms covered: - Drift: production data starts to look different from training data. Consumer behavior shifts, conditions change, and the ground beneath the model moves. - Model Drift: the resulting degradation in performance, which leads to poor predictions being acted on as if they were still reliable - Risk Management: identifying and mitigating bias, security exposure, regulatory risk and unexpected edge case behavior - ML Interpretability: showing why a model made a decision, so it can be explained, audited and challenged Drift tells you when a model is changing. Risk management tells you what to protect against. Interpretability tells you what the model is actually doing. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #ModelDrift #ResponsibleAI #MLOps #AIGlossary #H2Oai

You cannot trust what you do not measure. Evaluation looks different depending on whether you are assessing a model, a retrieval system or an agent. Terms covered: - Evaluation: systematically measuring quality, accuracy and reliability. Not a one-time check at the end of development, but an ongoing practice. - LLM Evaluation: factual accuracy, relevance, coherence, tone and safety, measured across thousands of inputs and edge cases - RAG Evaluation: assessing retrieval and generation separately, because poor retrieval produces unreliable answers no matter how good the model is - Agent Evaluation: judging decisions and actions, not just text. Did it complete the task, use the right tools, handle the unexpected? - Tracing Matrix: a structured framework mapping inputs to outputs, so failure points can be isolated and tracked over time Evaluation is what turns AI from a promising technology into a trustworthy one. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #AIEvaluation #ResponsibleAI #LLM #AIGlossary #H2Oai

AI now decides who gets approved for a loan and which claim gets flagged. Governance is how organizations stay accountable for that. Terms covered: - Data Governance: how data is collected, stored, accessed and protected. Before you can govern AI, you need to govern the data it runs on. - AI Governance: the policies and controls that make sure AI systems are developed and deployed ethically, safely and in compliance. Increasingly a legal requirement in regulated industries, not a best practice. - Agentic Governance: keeping autonomous agents inside defined boundaries, with logged, auditable actions and meaningful human oversight - Responsible AI: the broader commitment to AI that is fair, transparent and accountable across the full lifecycle Governance is not a constraint on AI progress. It is what makes progress sustainable. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #AIGovernance #ResponsibleAI #DataGovernance #AIGlossary #H2Oai

Where AI runs decides who controls the data. Here is the full spectrum, from fully isolated to fully managed. Terms covered: - On-Premises: software and infrastructure inside your own data centers. No data leaves the building. - Air-Gapped: completely isolated from all external networks, including the internet. Used in defense, intelligence and critical national infrastructure. - Cloud and SaaS: computing delivered over the internet, scalable and flexible, with data on third-party servers - VPC (Virtual Private Cloud): a private, dedicated section inside a public cloud - Sovereign AI: infrastructure owned, operated and controlled within a specific country or jurisdiction, so data, models and compute never cross into foreign infrastructure - Kubernetes: the open-source platform that automates deployment and scaling of AI workloads wherever they run The right model depends on who you are, what data you handle and what your obligations are. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #SovereignAI #AirGapped #EnterpriseAI #AIGlossary #H2Oai

Compliance often decides whether a deal moves forward at all, regardless of how strong the technology is. These are the three US frameworks you will meet first. Terms covered: - SOC 2: a security framework covering security, availability, processing integrity, confidentiality and privacy. Effectively a baseline requirement for selling software to US enterprise customers. - HIPAA: the legal standard for protecting patient health information in the US. Any platform handling medical records or clinical data must demonstrate compliance. - FedRAMP High: the most stringent tier of the US federal cloud authorization program, required for the most sensitive government data Certifications get treated as administrative checkboxes. In practice they are what determines whether procurement can proceed. Equivalent requirements exist across Europe, APAC and Latin America. Those are covered in the next three videos. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #SOC2 #HIPAA #FedRAMP #AICompliance #AIGlossary #H2Oai

Europe sets standards that increasingly shape how organizations operate globally, not just inside the EU. Three frameworks cover most of it. Terms covered: - GDPR: governs how personal data of EU residents is collected, stored, processed and protected, and applies to any organization handling that data regardless of where it is based. Automated decision-making that significantly affects individuals requires transparency and human oversight. - EU AI Act: the first comprehensive regulatory framework for AI, taking a risk-based approach. Some uses are prohibited outright, low-risk applications face minimal requirements, and high-risk systems in hiring, credit scoring, healthcare and law enforcement face strict obligations. - ISO 27001: the internationally recognized standard for information security management, often a baseline expectation in European enterprise procurement Together these define what responsible AI deployment looks like in Europe. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #GDPR #EUAIAct #ISO27001 #AICompliance #AIGlossary #H2Oai

APAC has no single rulebook. Compliance across the region means understanding a diverse and fast-evolving set of national regulations. Terms covered: - PDPA (Singapore): governs collection and use of personal data, with a strong emphasis on consent and purpose limitation - MAS Technology Risk Management Guidelines: AI and technology requirements for Singapore financial services, covering governance, risk controls and resilience - PIPL (China): strict requirements on collection, processing and cross-border transfer, with data localization rules meaning some data cannot leave the country at all - Privacy Act (Australia): personal data across sectors, with additional APRA requirements for financial services, currently being updated toward GDPR standards - DPDP Act (India): consent-based processing obligations covering a market of over a billion people APAC is not one market. It is many, and the organizations that navigate it treat regional compliance as a strategic capability. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #AICompliance #DataPrivacy #APAC #AIGlossary #H2Oai

Latin America has been building data protection law longer than most people realize, and enforcement is accelerating. Terms covered: - LGPD (Brazil): modeled explicitly on GDPR, covering access, correction, deletion and portability rights, requiring a data protection officer and documented processing. The ANPD has been steadily increasing enforcement. - PDPA (Argentina): in force since 2000, currently being updated. Argentina holds an EU adequacy decision, which matters for multinational deployments. - Law 1581 (Colombia): consent-based processing with active enforcement from the Superintendence of Industry and Commerce - LFPDPPP (Mexico): private sector requirements covering notice, consent and data subject rights The clear trend is convergence toward GDPR-aligned standards, so existing GDPR work carries over. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #LGPD #AICompliance #DataPrivacy #AIGlossary #H2Oai

Years of historical data only pay off when they answer forward-looking questions. Which customers will leave? Which transactions are fraudulent? Which equipment is about to fail? Terms covered: - Predictive AI: turning accumulated history into decisions, shifting from reacting after events to acting before them - AutoML: automating algorithm selection, hyperparameter tuning and model comparison. Work that took weeks of specialist time now takes hours, and teams that ran one experiment can run dozens. - Machine learning automation: now extending across feature management, data preparation, deployment and app creation, with agents orchestrating the full lifecycle - Tabular Foundation Model (TFM): pre-trained on vast amounts of structured data, so it generalizes to new datasets faster and needs less labeled data Think of a TFM as the difference between hiring someone with no experience and someone who has already solved thousands of similar problems. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #PredictiveAI #AutoML #FoundationModels #AIGlossary #H2Oai

Some of the most valuable questions in any organization are about time. What will demand look like next quarter? When will this equipment need maintenance? Terms covered: - Forecasting: using historical data to predict future values, accounting for seasonality, trends, external events and interdependencies that traditional statistical methods struggle with at scale - Time Series Models: built for sequential, time-ordered data, designed to detect the temporal patterns that a standard model treating each data point independently would miss - Anomaly Detection: flagging when reality diverges from what is expected. A spike that signals fraud, a temperature reading before equipment failure, a behavior pattern suggesting a breach. Forecasting tells you what is coming. Time series models give you the accuracy to trust it. Anomaly detection tells you when something has gone wrong. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #Forecasting #TimeSeries #AnomalyDetection #AIGlossary #H2Oai

Modern AI did not come from traditional programming. It came from a series of architectural breakthroughs, and three terms trace the whole path. Terms covered: - Neural Networks: layers of interconnected nodes that adjust millions of connections based on the errors they make, learning from examples rather than following instructions - Deep Learning: stacking many layers so each level learns more abstract representations. Early layers detect edges and tones, middle layers find shapes, later layers recognize complex concepts. - Transformer: the 2017 architecture that introduced attention, weighing the relevance of every part of an input against every other part at once. Every major language model today is built on it. From a single connected node, to the power of depth, to the attention mechanism that made language AI possible. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #DeepLearning #NeuralNetworks #Transformers #AIGlossary #H2Oai

Machine learning is not limited to numbers in a spreadsheet. Text, images and unlabeled data all carry patterns a model can learn from. Terms covered: - NLP Prediction: making predictions from text rather than generating it. Routing support tickets by urgency, classifying sentiment in financial news, flagging patients at elevated risk from clinical notes. - Vision Modeling: the same principle applied to images and video. Detecting production line defects in milliseconds, identifying anomalies in medical scans, tracking infrastructure change from satellite imagery. - Clustering: finding natural groupings based on similarity with no predefined categories. The model does not classify, it discovers. Text, images and unlabeled data. Three very different inputs, made meaningful through the same underlying capability. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #MachineLearning #ComputerVision #Clustering #AIGlossary #H2Oai

The model gets most of the attention. In practice it is often the smallest part of the work, and everything around it decides whether it delivers value. Terms covered: - Data Science Lifecycle: the end-to-end process, from defining the right problem through data preparation, modeling, validation, deployment, monitoring and maintenance - Feature Engineering: transforming raw data into what the model actually learns from. A timestamp becomes customer tenure. A transaction amount becomes a ratio against that customer's historical average. This work often shapes performance more than the choice of algorithm. - Model Scoring / Inference: the model making predictions on data it has never seen. Real-time fraud scoring, nightly batch demand forecasts, millions of recommendations at once. Understanding the full pipeline is what separates AI that works in a demo from AI that works in production. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #DataScience #FeatureEngineering #MLOps #AIGlossary #H2Oai

Over 90 AI terms covered, from machine learning basics and neural networks through generative AI, agentic systems, observability, platform architecture and compliance across four regions. The vocabulary of AI moves fast. You now have a foundation solid enough to keep up with it and contribute to the conversation. What to do next: - Take the quiz at the end of this course to earn your H2O.ai University certification - Go deeper with full courses and learning paths at https://h2o.ai/university - Rewatch any term you want to lock in, the full playlist is linked below — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #AI #AIGlossary #MachineLearning #GenerativeAI #H2Oai

How a model learns depends on whether your data comes with answers attached. Four terms cover almost every machine learning model built. Terms covered: - Supervised Learning: training on labeled data where both input and correct output are known, so the model learns patterns it can apply to cases it has never seen - Unsupervised Learning: no labels, no predefined categories. The model explores the data and finds natural groupings and structures on its own. - Classification: predicting which category something belongs to. Churn or stay, fraudulent or legitimate, benign or malignant. - Regression: predicting a continuous numerical value. A sale price, next month's inventory need, a customer's projected lifetime value. Supervised or unsupervised, classification or regression. This is the foundation everything else is built on. — AI Fundamentals Glossary | H2O.ai University Full playlist: https://www.youtube.com/playlist?list=PLAWKlIKS-66E Free courses and certifications: https://h2o.ai/university #MachineLearning #SupervisedLearning #Classification #AIGlossary #H2Oai