LLM and AI Glossary

Autonomous systems capable of planning, reasoning, memory usage, and executing multi-step workflows using tools and APIs.

Automates model building, feature selection, and hyperparameter tuning.

Identifies rare or unusual patterns in datasets.

Enhances real-world environments with digital overlays.

Allows models to focus on relevant parts of input for better predictions.

Ensures AI systems behave according to human values and intent.

Techniques to reduce unfairness in AI outcomes.

Probabilistic models for reasoning under uncertainty.

A scalable GAN architecture for high-quality image generation.

Combines decentralized trust with AI data pipelines.

AI systems that simulate human-like conversations.

Maximum input size an LLM can process at once.

Learning new tasks without forgetting old ones.

AI mimicking human reasoning.

Learns representations by comparing data samples.

Improves reasoning by guiding step-by-step outputs.

Multi-layer neural networks for complex pattern recognition.

AI-generated synthetic media.

Focus on improving datasets instead of models.

Pipeline design for collecting and preparing data.

Extracting insights from structured/unstructured data.

Generative models used in image/video synthesis.

Running AI locally on devices instead of cloud.

Making AI decisions interpretable.

Prevents forgetting in continual learning.

Combining models to improve performance.

Models that estimate uncertainty in predictions.

Vector representations capturing semantic meaning.

Adapting pre-trained models for specific use cases.

State-of-the-art, highly capable AI systems.

Large pre-trained models reusable across tasks.

Decentralized model training without sharing raw data.

Learning with very limited examples.

Enabling models to interact with APIs/tools.

AI that creates content (text, images, code, audio).

Probabilistic models with uncertainty estimation.

Connecting outputs to real-world data.

Models for graph-structured data.

Dual-network systems for realistic generation.

High-performance processors optimized for parallel computation, essential for training and running large AI and deep learning models.

Combines symbolic reasoning and neural networks.

Confident but incorrect outputs.

Techniques to identify false AI outputs.

Optimizing training configurations.

Breaking tasks into structured sub-tasks.

Understanding how models make decisions.

Dividing images into meaningful regions.

Training models to follow human instructions.

Using stored examples for prediction.

Continuous data modeling using neural networks.

Generating predictions using trained models.

Probability of multiple events occurring together.

Noise addition to prevent overfitting.

Accelerated machine learning framework.

Shared vector space for multiple data types.

Privacy technique ensuring indistinguishable records.

Transforms linear models to solve non-linear problems.

Structured data linking entities and relationships.

Compressing large models into smaller ones.

Binary classification algorithm.

Efficient image generation models.

Models trained on massive text corpora.

Efficient fine-tuning approach.

Compressed feature representation.

Operational practices for deploying, monitoring, and governing LLMs in production.

Learning how to learn new tasks quickly.

Poor generalization due to memorization.

Performance degradation over time.

AI handling text, image, audio, and video together.

Transform simple distributions into complex ones.

Brain-inspired computational models.

Combining logic and learning.

Language understanding and generation.

3D scene reconstruction from images.

Improving model parameters.

Learning from a single example.

Structured reasoning using domain knowledge.

Identifying unfamiliar data inputs.

Publicly available trained models.

Managing multi-step AI pipelines and agent workflows.

Designing inputs for better AI outputs.

Model evaluation metrics.

Dimensionality reduction method.

RL algorithm used in LLM alignment.

Efficient model adaptation.

Managing prompt lifecycle in production environments.

Reinforcement learning technique.

ML integrated with quantum computing.

Preparing models for low-precision deployment.

Combining retrieval systems with LLMs.

Learning via rewards and feedback.

Learning meaningful data features.

Deep networks with skip connections.

Aligning models with human preferences.

Search based on meaning and intent.

Predicting structured outputs.

Learning from unlabeled data.

Classification algorithm.

Artificial data for training models.

Splitting text into tokens.

Core architecture behind modern LLMs.

Reusing knowledge across tasks.

Latency metric in LLM responses.

Efficient decision-tree algorithms.

Model too simple to learn patterns.

Learning from unlabeled datasets.

Storage optimized for embeddings and similarity search.

Transformers applied to image data.

Generative latent-variable models.

Training difficulty in deep networks.

Reusing parameters across layers.

Numerical representation of language.

Training using imperfect labels.

Multilingual speech recognition system.

Efficient transformer implementations.

Popular gradient boosting algorithm.

Visualization tools for AI explainability.

Real-time object detection system.

Configuration-based ML workflows.

AI in agriculture forecasting outputs.

Performing tasks without training examples.

Standardizing data distributions.

Verifying data without revealing it.

Fault-tolerant distributed ML training.