GCP GenAI Leader flashcards
GCP GenAI Leader · Build recall with a focused review session.
32 cards. Read the term and the definition here. The deck at the top of this page is optional.
- Token
- A token is the basic unit of text a model processes, roughly four characters of English; billing and context limits are measured in tokens.
- Context window
- The maximum number of tokens a model can attend to in a single request, covering the input prompt, conversation history, and generated output.
- Hallucination
- A fluent model output that is factually wrong or unsupported by sources; mitigated by grounding, tuning, and citation requirements.
- Temperature
- A sampling parameter controlling randomness: low values yield focused, repeatable output, while high values increase diversity and creative variation.
- Embedding
- A numeric vector representation of text or images that captures semantic meaning, enabling similarity search, clustering, and retrieval-augmented generation.
- Foundation model
- A large model pre-trained on broad data that can be adapted to many downstream tasks through prompting, grounding, or tuning rather than building from scratch.
- Multimodal model
- A model that accepts or generates more than one data type, such as text, images, audio, and video, within a single request.
- Diffusion model
- A generative image model that starts from random noise and iteratively denoises to produce an image, the approach behind Imagen.
- Vertex AI
- Google Cloud's unified machine learning platform hosting Model Garden, Vertex AI Studio, tuning, evaluation, vector search, pipelines, and prediction endpoints.
- Model Garden
- A Vertex AI catalog of Google, partner, and open foundation models that teams can browse, test in a console, and deploy behind managed APIs.
- Vertex AI Studio
- A console and API workspace for writing prompts, testing Gemini models, comparing outputs, and exporting code for generative AI prototypes.
- Gemini
- Google's multimodal foundation model family, offered in Pro, Flash, and lightweight variants, accessible through Vertex AI or the Gemini API.
- Grounding with Google Search
- A Vertex AI option that links model responses to fresh web results and returns supporting citations, reducing unsupported or outdated answers.
- Imagen
- Google's text-to-image generation model family on Vertex AI, producing and editing images from natural language prompts.
- Agent Builder
- A Vertex AI service for assembling generative AI agents that combine instructions, tools, and enterprise data connectors such as search or databases.
- Google AI Studio
- A browser-based developer environment for rapidly prototyping with Gemini models, then exporting API keys or code for production integration.
- Few-shot prompting
- Including a few worked input-output examples inside the prompt so the model copies the intended format, tone, and task without retraining.
- Chain-of-thought prompting
- A prompting technique that asks the model to reason through intermediate steps before answering, improving accuracy on multi-step math and logic problems.
- Retrieval-augmented generation
- A grounding pattern where relevant documents are retrieved from a datastore, inserted into the prompt, and used by the model to answer with enterprise knowledge.
- Supervised fine-tuning
- Further training a foundation model on labeled input-output pairs so it specializes in your task, style, or domain vocabulary.
- Adapter tuning
- Parameter-efficient fine-tuning that trains small additional weights instead of the full model, reducing cost and keeping the base model unchanged.
- Grounding
- Connecting model output to verifiable sources such as web search results or private datastores so answers are attributable and factually current.
- System instruction
- A prompt element that sets persistent role, tone, and behavioral constraints for the model across every turn of a conversation.
- Prompt injection
- An attack where adversarial text in user input or retrieved documents makes the model disregard its instructions; mitigated by filtering and isolating untrusted content.
- Google AI Principles
- Google's published commitments for AI development, covering beneficial use, fairness, safety, accountability, privacy, and scientific excellence, reviewed through formal governance.
- Secure AI Framework
- Google's framework for securing AI systems against risks including model theft, data poisoning, prompt injection, and unsafe or harmful outputs.
- Model Armor
- A Google Cloud service that screens prompts and responses for prompt injection, jailbreaks, sensitive data leaks, and harmful content before reaching users.
- VPC Service Controls
- A Google Cloud security perimeter that restricts data movement for Vertex AI resources, helping prevent exfiltration of sensitive prompts and training data.
- Center of Excellence
- A central team that defines standards, reusable patterns, and enablement for generative AI so business units scale without duplicating work.
- Human-in-the-loop
- A design pattern where a person reviews or approves model output before it triggers a consequential action, required for high-risk decisions.
- AI literacy program
- Structured training that gives employees the skills and confidence to use generative AI tools safely and productively in daily work.
- Use case prioritization
- Ranking generative AI candidates by business value, data readiness, technical feasibility, and risk so pilots chosen can realistically scale.