Generative AI Vocabulary

An AI trained on vast text data to understand and generate language

“Large language models can write essays, code, and answer complex questions.”

A large model trained on broad data that can be adapted to many tasks

“Foundation models serve as the base for numerous downstream applications.”

The amount of text a model can consider at once

“The expanded context window allows processing entire documents.”

A parameter controlling randomness in AI outputs

“Lower temperature produces more deterministic and focused responses.”

Limiting word choices to the k most likely options

“Top-k sampling prevents the model from choosing unlikely tokens.”

Prompting technique that encourages step-by-step reasoning

“Chain of thought prompting improved the model's problem-solving accuracy.”

Combining search results with generative AI for grounded responses

“Retrieval augmented generation reduces hallucinations by citing sources.”

AI capable of processing multiple types of input like text and images

“Multimodal models can describe images and answer questions about them.”

Performing tasks without any task-specific training examples

“Zero-shot learning enables the model to classify categories it hasn't seen.”

Learning from just a handful of examples

“Few-shot learning adapted the model using only five examples per category.”

Learning patterns from examples provided in the prompt

“In-context learning allows customization without retraining the model.”

Ensuring AI behavior matches human values and intentions

“Alignment research focuses on making AI systems safe and beneficial.”

Reinforcement Learning from Human Feedback for training AI

“RLHF helped the model produce more helpful and harmless responses.”

Constraints preventing AI from producing harmful outputs

“Guardrails block the model from generating dangerous content.”

An AI system that can take actions autonomously to achieve goals

“The AI agent booked flights and hotels to complete the travel planning task.”

AI capability to invoke external functions or APIs

“Tool use enables the model to search the web and run calculations.”

Multi-step AI processes that iterate and self-correct

“The agentic workflow reviewed and revised its own code until tests passed.”

Artificially generated data used for training AI models

“Synthetic data augmented our limited real-world examples.”

Training a smaller model to mimic a larger one

“Model distillation created a faster version suitable for mobile devices.”

Reducing model precision to decrease size and increase speed

“Quantization made the model run efficiently on consumer hardware.”

A representation of compressed data

“The model maps images to points in a latent space.”

A particular mode in which something exists or is experienced or expressed

“The model supports text and image modalities.”