GCP GenAI Leader key terms

A category of AI systems that produce new content-text, images, audio, code, or video-by learning statistical patterns from large training datasets and generating outputs that resemble that data.

A large model trained on broad data that can be adapted to many downstream tasks through prompting or tuning, rather than being built for one narrow purpose.

A foundation model trained on massive amounts of text to predict and generate language, enabling tasks such as summarization, translation, question answering, and code generation.

A small unit of text-often a word piece or character sequence-that a model processes as input or output; context limits and usage pricing are usually measured in tokens.

A numeric vector representation of text, images, or other content that places semantically similar items close together in vector space, enabling search, clustering, and retrieval.

A neural network architecture that uses self-attention to weigh relationships across all parts of an input sequence in parallel, and underpins most modern large language models.

The maximum amount of text, measured in tokens, that a model can consider at once, covering the prompt plus the generated response.

A sampling setting that controls randomness in model output; lower values make responses more deterministic and focused, while higher values make them more varied and creative.

A model output that sounds plausible but is factually wrong or unsupported by any source, often arising when a model generates fluent text without reliable grounding.

A model that can accept or produce more than one type of data, such as text, images, audio, or video, within a single interaction.

A generative model that learns to create data such as images by starting from random noise and progressively removing noise until a coherent result emerges.

Google Cloud's unified platform for building, training, deploying, and managing machine learning models, including generative AI models and agents, with enterprise controls.

A Google Cloud console environment for prototyping with generative models: writing and comparing prompts, testing multimodal inputs, and exporting working code.

Google's family of multimodal generative AI models that handle text, images, audio, video, and code, available through consumer products, APIs, and Google Cloud.

An assistant integrated across Google Cloud products that helps users write code, analyze data, troubleshoot, and operate cloud resources using natural language.

A Vertex AI catalog of foundation and task-specific models from Google and third parties, with options to test, deploy, or tune them inside Google Cloud.

Google's family of text-to-image generation models on Vertex AI, used to create and edit images from natural language prompts.

Google's generative video model on Vertex AI that creates video clips from text prompts or reference images.

A Google Cloud service that builds search and retrieval experiences over an organization's own documents and data using embeddings and ranking, and can ground generative responses.

The practice of designing, structuring, and iterating on prompts-instructions, examples, and context-to get more accurate and useful model responses.

A prompting technique that includes a small number of example input-output pairs in the prompt so the model can infer the desired format or behavior.

A technique that asks a model to reason through intermediate steps before giving a final answer, which can improve performance on multi-step problems.

Guidance given to a model separately from user input that sets its role, tone, and constraints for the whole conversation.

An approach where relevant documents are retrieved from a knowledge source and inserted into the prompt so the model's answer is based on that evidence.

Connecting a model's response to verifiable external sources-such as search results or enterprise documents-so answers are more accurate and attributable.

Further training of a pre-trained model on a smaller, task-specific dataset so its behavior better matches a particular style, format, or domain.

Fine-tuning methods that update only a small set of adapter weights instead of the whole model, lowering cost and preserving the base model.

Reinforcement learning from human feedback: a training stage in which human preference ratings guide the model toward responses people find more helpful and safer.

The process of measuring a model's quality on defined criteria-such as accuracy, groundedness, and safety-using test datasets, human raters, or automated metrics.

An approach to designing and operating AI systems that considers fairness, safety, privacy, transparency, and accountability alongside performance and business value.

A control where people review, approve, or correct AI outputs at key points, especially for high-stakes decisions, rather than letting the system act autonomously.

Policies and controls that define how data is collected, classified, accessed, retained, and used, helping keep generative AI inputs and outputs compliant and secure.

Systematic unfairness in model outputs that disadvantages particular groups, often reflecting imbalances or stereotypes present in training data.

The degree to which people can understand why an AI system produced a given output, supporting trust, debugging, and regulatory obligations.

Use of AI tools by employees without organizational approval or oversight, which creates risks around data leakage, compliance, and inconsistent quality.

A centralized team that sets standards, shares reusable assets and best practices, and guides business units adopting a technology such as generative AI.

The structured process of preparing people and workflows for a new technology, covering communication, training, incentives, and support to drive adoption.

A security attack in which malicious instructions hidden in user input or retrieved content cause a model to ignore its intended instructions or leak information.