- Foundation Model (FM)
- A large-scale model pretrained on broad data that serves as a base for many downstream tasks via prompting or fine-tuning.
- Large Language Model (LLM)
- A foundation model specialized in understanding and generating natural language text, trained on massive text datasets.
- Generative AI (GenAI)
- AI systems capable of creating new content such as text, images, audio, or code based on learned patterns from training data.
- Prompt Engineering
- The practice of designing and refining input prompts to elicit better or more accurate outputs from a generative AI model.
- Fine-Tuning
- Additional training of a pretrained model on a smaller, task-specific dataset to specialize its behavior.
- Embedding
- A numerical vector representation of text, images, or other data that captures semantic meaning, used for similarity search and RAG.
- Vector Database
- A database optimized for storing and querying embeddings by similarity, commonly used to power RAG retrieval steps.
- Token
- A unit of text, such as a word or subword piece, that a language model processes as input or output.
- Context Window
- The maximum amount of text (measured in tokens) a model can consider at once when generating a response.
- Hallucination
- An AI-generated output that is factually incorrect or fabricated but presented as though it were accurate.
- Amazon Bedrock
- A managed AWS service offering access to multiple foundation models via a unified API, along with tools for customization and building generative AI applications.
- Amazon SageMaker
- A fully managed AWS platform for building, training, tuning, and deploying machine learning models.
- Amazon Q
- AWS's generative AI-powered assistant that helps users with tasks like answering questions, generating content, and assisting with code.
- Amazon Titan
- AWS's family of first-party foundation models available through Amazon Bedrock, covering text and embedding use cases.
- Responsible AI
- An approach to developing and deploying AI systems that emphasizes fairness, transparency, safety, privacy, and accountability.
- Model Bias
- Systematic skew in a model's outputs that unfairly favors or disadvantages certain groups, often stemming from unrepresentative training data.
- Explainability
- The extent to which a model's decisions or predictions can be understood and interpreted by humans.
- Guardrails
- Configurable policies applied to generative AI applications to filter, block, or redact undesired prompts or responses, such as harmful content or PII.
- Amazon SageMaker Model Monitor
- A SageMaker capability that tracks deployed models in production for data quality issues and drift.
- Inference
- The process of using a trained model to generate predictions or outputs from new input data.
- Prompt
- The input text or instructions provided to a generative AI model to elicit a response.
- Shared Responsibility Model
- The AWS security framework in which AWS secures the cloud infrastructure and managed services while customers are responsible for securing their own data, configurations, and usage.