LLM (Large Language Model)
A deep learning model trained on massive text corpora that can generate, summarize, translate, and reason about text. Examples include Gemini, PaLM, and GPT.
Foundation Model
A large AI model pre-trained on broad data that can be adapted to many downstream tasks. The base for most modern generative AI applications.
RAG (Retrieval-Augmented Generation)
A technique that retrieves relevant external documents and feeds them to a generative model, grounding its responses in factual, up-to-date information.
Grounding
Connecting model outputs to verifiable external data sources (Google Search, enterprise data) to reduce hallucinations and improve factual accuracy.
Prompt Engineering
The practice of designing and refining input prompts to guide a generative model toward desired outputs. Includes zero-shot, few-shot, chain-of-thought, and ReAct techniques.
Fine-Tuning
Further training a pre-trained model on a specific dataset to improve performance on a particular task or domain. Vertex AI supports supervised and RLHF fine-tuning.
Gemini
Google's most capable multimodal AI model family. Handles text, images, audio, video, and code natively. Available as Gemini Ultra, Pro, Flash, and Nano variants.
Gemma
Google's family of open-weight models built from the same research as Gemini. Lightweight and deployable on-device or in custom infrastructure.
Imagen
Google's text-to-image diffusion model. Generates photorealistic images from text descriptions. Available through Vertex AI for enterprise image generation.
Veo
Google's video generation model that creates high-quality videos from text or image prompts. Supports cinematic styles and creative video production workflows.
RLHF
Reinforcement Learning from Human Feedback. A fine-tuning technique where human preferences are used to train a reward model that guides the LLM toward more helpful, harmless outputs.
Temperature
A sampling parameter controlling output randomness. 0 = deterministic (most likely tokens), higher values = more creative/random outputs. Typical range 0–2.
Top-P (Nucleus Sampling)
A sampling parameter that limits token selection to the smallest set whose cumulative probability exceeds P. Lower values = more focused; higher = more diverse outputs.
Hallucination
When a generative model produces plausible-sounding but factually incorrect or fabricated information. Mitigated by grounding, RAG, and human-in-the-loop review.
SAIF (Secure AI Framework)
Google's framework for securing AI systems. Covers model security, data governance, access controls, and threat modeling specifically for AI/ML workloads.
Vertex AI
Google Cloud's unified AI platform for building, deploying, and managing ML and generative AI models. Includes Model Garden, Studio, Agent Builder, and RAG capabilities.
Model Garden
A curated catalog of foundation models in Vertex AI. Offers Google first-party models (Gemini, Gemma), open-source models, and third-party models for evaluation and deployment.
Agent Builder
A Vertex AI tool for creating AI agents that can use tools, search data, and take actions. Enables building conversational and task-oriented agents without extensive coding.
TPU (Tensor Processing Unit)
Google's custom AI accelerator designed for high-throughput training and inference. TPU v5e and v5p power Google's largest AI models and are available via Google Cloud.
Diffusion Model
A generative model that creates data by learning to reverse a noise-adding process. Used for image (Imagen), video (Veo), and audio generation.